AMD CPU speculation... and expert conjecture

[cdrkf]

The discussion about AMD losing things by going bulk was exaggerated, but it is true. The general consensus on Kaveri desktop is that the IPC gains were wiped out by the fact that Kaveri can't clock as high. I don't see too many architectural changes that would have caused that. It seems far more likely to be related to the process instead.

Yeah, there are good OCing Kaveri's out there. But the fact remains that at the end of the day we had to see a stock clockspeed reduction thanks to going bulk and that the point of diminishing returns when raising TDP is much lower.

AMD lost a lot going from Richland to Kaveri in terms of clock speeds. It wasn't some catastrophic unable to break 3ghz type of event, but it was enough to hurt overall CPU performance.

Evidently part of the IPC increment introduced by Steamroller architectural improvements was dropped by the lower base clock. Nobody is negating that! What is being discussed here is why AMD underclocked Kaveri.

We know that Kaveri average OC on air is 4.5GHz. We have seen Kaveri benchmarked @4.7GHz:

http://www.hardcoreware.net/kaveri-7850k-overclocked-benchmarks/

Anand showed that Kaveri can be OC up to ~4.2GHz without changing the voltage (1.25V) and with almost same power consumption 147W vs 148W

We also know that early talks and docs from AMD mentioned explicitly that Kaveri was a 100W APU originally.

Thus the question is why did AMD downrated Kaveri to 95W and reduced the base frequency from the expected 4.0GHz (check my BSN article) to final 3.7GHz? Why did AMD do that when Kaveri can hit 4.5/4.7GHz on air?

This is the million dollar question that nobody is answering!

I think you're own information actually answers that question: 4ghz base is *the fastest* they could run Kaveri at reasonable power consumption levels. Given AMD's new strategy, they appear to be trying to release *something* on a very regular cadence these days (every 6 to 9 months), using minor refreshes in between the major technology changes. This is similar to Intel's strategy and is something that helps keep people interested, and more importantly builds investor confidence that things are progressing (and something AMD was very poor at before the regime change).

I personally think that Kaveri will get a mid cycle refresh to counter the Haswell update and keep things ticking over before Carizo can be released (much like they did with Richland), so keeping the initial release of Kaveri modest gives them some room to scale things up in a few months time when Kaveri 2.0 gets shipped (probably with some other tweaking under the hood as well). This is pure speculation at this point, however it would fit with their actions lately and would make sense from the business point of view as it should help keep the stock prices a bit more stable.

 

[de5_Roy]

Kaveri desktop has been released only in two quad-core SKU. Only one mobile SKU is dual-core whereas the other six mobile SKUs are quad-core.

The no existence of more dual-core salvage parts from quad core dies is clearly indicating extremely good yields.

delay of kaveri launch, delay of mobile skus, shortage of fully functional mobile skus, absence of a8 7600, power consumption, high tdp of only two available dt skus this late in the cycle - all point to process and yield problems. and then there's the un-seeable track record of glofo.

It was clearly a typo. I did mean Skybridge. I edited and corrected it ^^. You would learn the difference between making an occasional typo and the systematic spreading of daily FUD/nonsense such as "it is going to suck in performance terms.".

Spoiler

E.g. this is dictionary entry for FUD:

Fear, Uncertainty and Doubt. Also known as scare tactics, either accomplished by threat or making the opponent doubt his standpoint. Not only used in lawsuits, but also in politics and military propaganda.

And this is entry for typo:

A misspeeling of a word when typing or texting

Its annoying when people point out your erorrs and correct you, because some times you don't focus on your grammar and punctuation like some fucking nerd or maybe you just didn't know how to speel the damn word.

Note the fine irony in this last definition

LOL. this is probably the most arrogant, ham, hostile, pathetic admission of typo i've seen this year. at least makes to top 50 list of admission of mistakes from [strike]trolls[/strike] people who got caught red(geddit?) handed during trolling/trollbaiting/flaming.

 

[-Fran-]

I would say AMD chose to use 95W instead of 100+W because of OEMs.

MoBo makers can cheap out in some components to hit that TDP (going on a limb; never seen a graph or article telling how watts scale in price for OEMs) and usually, from what I understand, cheaper components for low TDP have a similar life span as the more expensive ones (VRMs and capacitors, in particular). Also, 95W was (is?) a very common OEM TDP from old (or Intel as well), so AMD going to 100W was weird at the time for Llano. It could have been HP or Dell that told AMD to hit 95W instead of 100W.

Anyway, the trade off for Kaveri was not to just say "look, we went under 32nm as well!". I believe they said it very clearly: SOI wasn't dense enough to include GCN and keep the same die size. They had to sacrifice speed because of die size; I would say that is the correct trade off they made.

Cheers!

 

[de5_Roy]

I would say AMD chose to use 95W instead of 100+W because of OEMs.
MoBo makers can cheap out in some components to hit that TDP (going on a limb; never seen a graph or article telling how watts scale in price for OEMs) and usually, from what I understand, cheaper components for low TDP have a similar life span as the more expensive ones (VRMs and capacitors, in particular). Also, 95W was (is?) a very common OEM TDP from old (or Intel as well), so AMD going to 100W was weird at the time for Llano. It could have been HP or Dell that told AMD to hit 95W instead of 100W.

i think so too.

may be llano had a lower tjmax.? i don't know for sure. llano owners will know better.

Anyway, the trade off for Kaveri was not to just say "look, we went under 32nm as well!". I believe they said it very clearly: SOI wasn't dense enough to include GCN and keep the same die size. They had to sacrifice speed because of die size; I would say that is the correct trade off they made.

Cheers!

imo, it might be more like glofo couldn't do it. they had problem with 32nm as well. 28nm is new and still seems to have room for improvement. amd's tradeoff was right considering the existing circumstances.

 

[juanrga]

Kaveri desktop has been released only in two quad-core SKU. Only one mobile SKU is dual-core whereas the other six mobile SKUs are quad-core.

The no existence of more dual-core salvage parts from quad core dies is clearly indicating extremely good yields.

delay of kaveri launch, delay of mobile skus, shortage of fully functional mobile skus, absence of a8 7600, power consumption, high tdp of only two available dt skus this late in the cycle - all point to process and yield problems. and then there's the un-seeable track record of glofo.

Apart from you ignoring my main argument (yields are good because there is almost no dual-cores), my sources give other reasons for the delays. E.g. Kaveri DT delay was caused by HSA issues. In any case even if we accept your 'hypothesis' that delays were caused by yield problems, this doesn't imply that the problems remain when products are released.

Absence of a8 7600 means nothing because this is a low-'binned' SKU derived from the top part which is already selling. You cannot really pretend that "terrible yields" are the reason why 3.1/3.3GHz 6CU parts are not in the market, when higher 3.7GHz 8CU parts are being sold. Can you?

Power consumption is another non-issue. 95W Kaveri consumes much less power than Richland 100W. Check the Tomshardware chart given.

"Track record of glofo" also doesn't count because this is bulk not FDSOI. (FDSOI is more complex to deal with, I already explained why).

In short, none of your arguments make sense. None of them explain why AMD downrated Kaveri from 4.0GHz/100W to 3.7GHz/95W

Luckily other people in this thread has given plausible arguments.

It was clearly a typo. I did mean Skybridge. I edited and corrected it ^^. You would learn the difference between making an occasional typo and the systematic spreading of daily FUD/nonsense such as "it is going to suck in performance terms.".

LOL. this is probably the most arrogant, ham, hostile, pathetic admission of typo i've seen this year. at least makes to top 50 list of admission of mistakes from [strike]trolls[/strike] people who got caught red(geddit?) handed during trolling/trollbaiting/flaming.

You used an evident typo to pretend that it was "spreading FUD/nonsense". You posting this new personal attack is not surprising me.

 

[juanrga]

The discussion about AMD losing things by going bulk was exaggerated, but it is true. The general consensus on Kaveri desktop is that the IPC gains were wiped out by the fact that Kaveri can't clock as high. I don't see too many architectural changes that would have caused that. It seems far more likely to be related to the process instead.

Yeah, there are good OCing Kaveri's out there. But the fact remains that at the end of the day we had to see a stock clockspeed reduction thanks to going bulk and that the point of diminishing returns when raising TDP is much lower.

AMD lost a lot going from Richland to Kaveri in terms of clock speeds. It wasn't some catastrophic unable to break 3ghz type of event, but it was enough to hurt overall CPU performance.

Evidently part of the IPC increment introduced by Steamroller architectural improvements was dropped by the lower base clock. Nobody is negating that! What is being discussed here is why AMD underclocked Kaveri.

We know that Kaveri average OC on air is 4.5GHz. We have seen Kaveri benchmarked @4.7GHz:

http://www.hardcoreware.net/kaveri-7850k-overclocked-benchmarks/

Anand showed that Kaveri can be OC up to ~4.2GHz without changing the voltage (1.25V) and with almost same power consumption 147W vs 148W

We also know that early talks and docs from AMD mentioned explicitly that Kaveri was a 100W APU originally.

Thus the question is why did AMD downrated Kaveri to 95W and reduced the base frequency from the expected 4.0GHz (check my BSN article) to final 3.7GHz? Why did AMD do that when Kaveri can hit 4.5/4.7GHz on air?

This is the million dollar question that nobody is answering!

I think you're own information actually answers that question: 4ghz base is *the fastest* they could run Kaveri at reasonable power consumption levels. Given AMD's new strategy, they appear to be trying to release *something* on a very regular cadence these days (every 6 to 9 months), using minor refreshes in between the major technology changes. This is similar to Intel's strategy and is something that helps keep people interested, and more importantly builds investor confidence that things are progressing (and something AMD was very poor at before the regime change).

I personally think that Kaveri will get a mid cycle refresh to counter the Haswell update and keep things ticking over before Carizo can be released (much like they did with Richland), so keeping the initial release of Kaveri modest gives them some room to scale things up in a few months time when Kaveri 2.0 gets shipped (probably with some other tweaking under the hood as well). This is pure speculation at this point, however it would fit with their actions lately and would make sense from the business point of view as it should help keep the stock prices a bit more stable.

I think information shows that AMD could have released an 4Ghz/100W Kaveri whereas maintaining power levels below Richland 100W

Richland was not planned initially. The original roadmap was Trinity --> Kaveri. Richland was released to fill the hole left by Kaveri delay.

The current roadmap is Kaveri --> Carrizo and so far as I know Carrizo has been already tapped-out. Thus I don't expect delays.

In any case I find interesting your idea that AMD did release a low clocked Kaveri because could release Kaveri 2.0 refresh with higher clocks in some months.

My hypothesis for the reduction in clocks is completely different. I believe (pure speculation) that AMD downrated of Kaveri is closely related to Carrizo being a 65W APU.

I have made some numbers for my forthcoming Excavator article and I believe that Excavator architectural improvements will be compensated by a new reduction in clocks. If my numbers are accurate (no warranty, because there is not enough data), Excavator would barely improve over a 4.0GHz/100W Kaveri, but will bring ~10% improvement over a 3.7GHz/95W Kaveri.

Of course if AMD releases Kaveri 2.0 in next months, then my hypothesis is automatically destroyed. ;-)

 

[juanrga]

I would say AMD chose to use 95W instead of 100+W because of OEMs.

MoBo makers can cheap out in some components to hit that TDP (going on a limb; never seen a graph or article telling how watts scale in price for OEMs) and usually, from what I understand, cheaper components for low TDP have a similar life span as the more expensive ones (VRMs and capacitors, in particular). Also, 95W was (is?) a very common OEM TDP from old (or Intel as well), so AMD going to 100W was weird at the time for Llano. It could have been HP or Dell that told AMD to hit 95W instead of 100W.

It could be. Do you believe Carrizo 65W is another OEM requirement?

Anyway, the trade off for Kaveri was not to just say "look, we went under 32nm as well!". I believe they said it very clearly: SOI wasn't dense enough to include GCN and keep the same die size.

Yes! I already gave the link to the slide mentioning that one of the reasons why AMD chose bulk was its optimization for density. I also give some data: 32nm --> 28nm only brings about 31% better density, when GCN in Kaveri required 85% better density.

 

[-Fran-]

The trade off I mention sounds about right given the "bet" AMD has with HSA: They need GCN right now (already in, of course) so they can start giving out on their promise and blah. If they wouldn't have put GCN and stuck to VLIW4/5 in the same 28 [strike]HKMG[/strike] SHP bulk, then the die size would have been very small (good thing), but they wouldn't have had advanced a single inch with the HSA integration part (very, very bad).

Anyway, I feel like 125W, 95w, 65W, 45W (notebook territory for regular cooling), 35W, 20W and 15W are "common" for OEMs. Intel has been kind of "breaking the mold" with their new TDPs of late, but both AMD and Intel used to follow those numbers with OEMs. That's how I recall it at least.

Cheers!

EDIT: Corrected mistake.

 

[de5_Roy]

sorry, googling took too long. the internet is vast and full of puppies and deer cubs.

Apart from you ignoring my main argument (yields are good because there is almost no dual-cores),

your "main argument" only counts if the apu die only has cpu cores. it has more components.

my sources give other reasons for the delays. E.g. Kaveri DT delay was caused by HSA issues. In any case even if we accept your hypothesis that delays were caused by yield problems, this doesn't imply that the problems remain when products are released.
Absence of a8 7600 means nothing because this is a low-'binned' SKU derived from the top part which is already selling. You cannot really pretend that "terrible yields" are the reason why 3.1/3.3GHz 6CU parts are not in the market, when
higher 3.7GHz 8CU parts are being sold.

Power consumption is another non-issue. 95W Kaveri consumes much less power than Richland 100W.

"Track record of glofo" also doesn't count because this is bulk not FDSOI.

terrible yields? where did i say that? you're outright lying. and now you got caught, again.
i mention absence of a8 with shortage of fully enabled mobile skus because glofo seem to have failed to rein in leakage with their new process node. the same factor (among others) could be preventing amd from enabling fully functional mobile skus. as for yields, it seems amd is taking longer than usual to stockpile apus to maintain stable supply and retail prices for dt apus haven't gone down much in addition to previous delays. i also base my assumption (geddit? subject to error) based on amd selecting low-mid range clockrates for mobile skus.
track record is also important because glofo has always had more troubles with new nodes. they shouldn't have as much with delayed-launched half node shrink. since they're using bulk silicon with design that was supposed to be for soi, their problems increased. it'll take glofo some time to iron out the problems but as usual, it'll put amd at an unfortunate disadvantage despite having a good device.
edit2: as for 95w and 100w kaveri - tdp isn't the same as peak or average power consumption. a 95w tdp kaveri can use 100w and vice versa. besides, none of my statements had anything to do with why amd rated a10 7850k at 95w. this is just a red herring.

I agree on that you taking an evident typo and pretending that it was "spreading FUD/nonsense" is in the top 50 of your silliest attacks against me. Specially, when I have been discussing Intel Skylake for months. E.g. I was the guy who leaked that the chief architect of Phi KL is the guy who is designing the new graphics in Skylake, for instance.

see, you're resorting to twisting my words. it's called a straw man argument - a fallacy-trollbait. and... intel skylake with no context to amd in an amd thread? that's way off-topic.
you're using off-topic stuff to justify your mistakes and arrogance to cover the fact that you got caught near-instantly while trolling. just cool your head somwhere and come back with something logically relevant.

edit: missed the late edits. doesn't change my stance or you getting caught lying and trolling. if you consider calling out those as personal attacks, i can't help you there.

 

[jimmysmitty]

Intel users never admit that anything that comes from Intel is a failure. .

I am an Intel user and I say crapburst was an epic fail. Sadly, AMD tried to imitate that fail with faildozer. Somewhere along the line AMD lost their way and made the same mistakes they used to bash Intel for.

I kinda get the analogy however the 'goal' of bulldozer if you like is very different to netburst. Netbursts design was to achieve the highest possible frequency on a single core. Bulldozer's goal was to cram as many cores onto a single bit of silicon within a specific TDP range. Unfortunately in the pursuit of more cores they decided to keep the individual cores 'light' making the bet that software as a whole would be multi-threaded enough to get a big boost of performance from the extra threads despite the lower instruction / clock of the individual cores. It proved to be a bad bet (although I would point out that AMD's 8 core fx chips are ageing quite well as software is slowly starting to use the threads).

The problem is that AMD did do the same thing as NetBurst and put long pipelines in to increase the core frequency and it ended up in the same exact result, it got higher clocks but had lower IPC and used way more power. By default a Haswell CPU uses half of what a Piledriver core does. Overclock that Piledriver core and the performance gains are minimal for the power usage jump.

Honestly it is pretty mush a 180 right now with Intel and AMD. AMD has high frequency, high power draw and low IPC parts while Intel has low frequency, low power draw and high IPC parts. It is pretty much Athlon 64 vs Pentium 4 in reverse.

I am sure AMD will do better with the talent they got back but it is still kind of funny to see them doing something they trashed Intel for. Much like when they made their MCM Opterons to get more cores per chip after trashing Intel for making the MCM quad cores.

Bulldozer was just the first (and somewhat flawed) version, sort of like Agena and Piledriver would be Deneb, Steamroller is Thuban, etc.

Piledriver and Steamroller fixed some flaws of Bulldozer design (specially Steamroller which is a step backward from the CMT design). But to be fair, the entire arch is unfixable. This is why AMD is not releasing any Warsaw CPU based in Steamroller modules for servers and also why the new K12 core is a clean design that abandon CMT.

I kinda get the analogy however the 'goal' of bulldozer if you like is very different to netburst. Netbursts design was to achieve the highest possible frequency on a single core. Bulldozer's goal was to cram as many cores onto a single bit of silicon within a specific TDP range. Unfortunately in the pursuit of more cores they decided to keep the individual cores 'light' making the bet that software as a whole would be multi-threaded enough to get a big boost of performance from the extra threads despite the lower instruction / clock of the individual cores. It proved to be a bad bet (although I would point out that AMD's 8 core fx chips are ageing quite well as software is slowly starting to use the threads).

I agree on that 8 core FX chips are aging well, this is why AMD extended the current FX line up to 2015. The new gen games are scaling up well. This is a new example of scaling up to 8 cores:

Yet on the other side is showing how badly it is actually scaling compared to Intel. THG got stock FX 8350 at i5 3550 performance levels with the best being a 6 core i7 (makes sense as it is a real 6 core)

It is not surprising that from a 3 module 6 thread to a 4 module 8 thread it scales well. What is surprising is that it scales horribly compared to Intel.

Consider that Intel is planning Broadwell this year with Sky Lake next year as well as Haswell-E this year. AMD is falling behind fast and even the consoles wont save them as those optimizations don't work for PC (both consoles have unified memory which alone screws up any optimizations for PC).

That's right, woodchuck-chuckers - it's Groundhog Day!

LOL, ya think... these two argue back and forth back and forth about the same point over and over and over again. Great post for the release day of Edge of Tomorrow.

Lets rename the forum "juan and 8350 argue and say the same thing with some AMD news sometimes"

Oh by the way AMD owns the graphics card charts, pretty surprising from Tom's

It is not that surprising. If you have been on THG as long as I have (although I didn't sign up until later for the forums I have been viewing since Tom ran it) you would know that while some of the writers and editors have preferences, it is overall a fair site. I have seen AMD (ATI) dominate the charts before. It all depends on if what they are producing is actually good. The GeForce 8000 series was superior to the HD2000 series. I still owned a HD2900Pro 1GB because, well it was near the same as a 2900XT just a bit slower clocks and cheap as hell. Plus it had 1GB of VRAM which was insane at the time.

Now a fan site like say AMDZone, or IntelZone those are just biased no matter what.

 

[jdwii]

I actually bought a 8350FX had it for a couple days first thing i noticed was in gaming it is 20% faster even with my 1100T at 3.9Ghz but i also notice that in Fritz-CineBench it was slower by 5% and with Wprime being really slower by around 20%(weird). However its important to note that this CPU is using less power then my phenom Overclocked. And everything on the desktop feels faster as well as a speed up in shutdown and boot up times. Going to try and OC this thing to 4.4Ghz at least.

However i know without a doubt compared to the I5 per clock this Piledriver is 50% behind, i tested it compared to the I5 haswell my friend got and its clocked at 3.2Ghz i lowered the clock to this chip to compare. Even in multithreaded(handbrake) work loads this thing still loses at that clock rate or barley comes even. Compared to my Phenom its around 10% per clock slower but i noticed that Piledriver scaled better with increase in frequency just a 100mhz jump brought my Cinebench core up 2.6% which is perfect scaling something the phenom did not do.

 

[de5_Roy]

^^ what o.s. (7 sp1 or 8.1), which cinebench (hopefully not the intel-friendly one), what kind of handbrake workload (1st pass, haswell should be faster. fx should be faster in 2nd pass), which tests used stock clockrates and which used o.c. etc.. details.

edit:
for some reason the 28nm shp info isn't available on glofo's website. instead i found 28nm hpp (something i thought was cancelled)
http://www.globalfoundries.com/technology-solutions/28nm-high-k-metal-gate
even roadmaps show only 28nm lpm and hpp. no shp. no 16nm finfet either, only 14nm finfet. and 20nm lpm, no shp or hpp. i am not really good with these acronyms. weren't steamroller and kaveri supposed to be fabbed on 28nm hpp? this distracted me for a while. and puppies.

 

[wh3resmycar]

I still recall when an 'expert' here pretended to convince me that "APUs are only for cheap gaming" Yeah sure!

actually they are.

 

[jdwii]

^^ what o.s. (7 sp1 or 8.1), which cinebench (hopefully not the intel-friendly one), what kind of handbrake workload (1st pass, haswell should be faster. fx should be faster in 2nd pass), which tests used stock clockrates and which used o.c. etc.. details.

edit:
for some reason the 28nm shp info isn't available on glofo's website. instead i found 28nm hpp (something i thought was cancelled)
http://www.globalfoundries.com/technology-solutions/28nm-high-k-metal-gate
even roadmaps show only 28nm lpm and hpp. no shp. no 16nm finfet either, only 14nm finfet. and 20nm lpm, no shp or hpp. i am not really good with these acronyms. weren't steamroller and kaveri supposed to be fabbed on 28nm hpp? this distracted me for a while. and puppies.

Hand brake is just compressing 1 video file also i'm on 8.1 and he is on 7. The Stock voltage on this is 1.31250V so i should be able to take this thing on when it comes to overclocking(if it stay's warm low VID means hotter CPU). Also i compared the tests to my phenom overclocked and the fx at stock and my friends I5 had turbo disabled and it can't be overclocked being a non K edition.
Edit my overclocking results so far sorry mods thought i would post this here.
1.3125V Stock Voltage LLC High
4400 8:55 65C Processor, 70C Core
Stable 1 Hour Prime
Failed but 4.3Ghz works at 1.3125V CPU gets to 62C however after running for 3 hours on prime, might go water to get anything more. Anyways at least at 4.3Ghz this CPU beats my old one in Cinebench and Fritiz but still slower in Wprime by a decent amount for some reason.

 

[juanrga]

your "main argument" only counts if the apu die only has cpu cores. it has more components.

Really? Are you sure that APUs have more components than CPU cores? If what you say is true this is revolutionary news. :lol: Sarcasm apart, the CPU on Kaveri occupies about a 50% of the die. Yield problems would affect to the CPU and the almost absence of dual-cores implies that this is not the case.

terrible yields? where did i say that? you're outright lying. and now you got caught, again.

In reality I was being sarcastic. Again let me help you

http://www.urbandictionary.com/define.php?term=sarcastic

i mention absence of a8

and you ignored the argument reply

i also base my assumption (geddit? subject to error) based on amd selecting low-mid range clockrates for mobile skus.

Richland mobile turbo did span 2.6--3.5GHz whereas Kaveri mobile turbo spans 3.0--3-6GHz. I see higher clocks for this new gen.

Your Kaveri clocks are lower because (sarcasm -->) "yields are terrible" (<-- sarcasm) kind of argument doesn't fit too well with reality.

track record is also important because glofo has always had more troubles with new nodes.

Troubles with FDSOI doesn't necessarily translate to bulk.

they shouldn't have as much with delayed-launched half node shrink. since they're using bulk silicon with design that was supposed to be for soi, their problems increased.

What "design"?

edit2: as for 95w and 100w kaveri - tdp isn't the same as peak or average power consumption. a 95w tdp kaveri can use 100w and vice versa.

Cof Cof, but what is being discussed is something different, Capt. obvious.

see, you're resorting to twisting my words. it's called a straw man argument - a fallacy-trollbait. and... intel skylake with no context to amd in an amd thread? that's way off-topic.
you're using off-topic stuff to justify your mistakes and arrogance to cover the fact that you got caught near-instantly while trolling. just cool your head somwhere and come back with something logically relevant.

edit: missed the late edits. doesn't change my stance or you getting caught lying and trolling. if you consider calling out those as personal attacks, i can't help you there.

I see that your irony detector continues to fail. I wonder why You also miss some posts on purpose for being able to post that of above. Your tactic is not new, thus I am not really impressed.

For your information, I have been discussing in this thread some aspects of Skylake such its new graphics architecture when discussing which will be the competition of AMD new products. If the rumor of Larrabee coming back on Skylake is true, then Intel will be not only fighting on pure graphics workloads but could treat AMD HSA.

Neo-heterogeneity vs HSA

for some reason the 28nm shp info isn't available on glofo's website. instead i found 28nm hpp (something i thought was cancelled)
http://www.globalfoundries.com/technology-solutions/28nm-high-k-metal-gate
even roadmaps show only 28nm lpm and hpp. no shp. no 16nm finfet either, only 14nm finfet. and 20nm lpm, no shp or hpp. i am not really good with these acronyms. weren't steamroller and kaveri supposed to be fabbed on 28nm hpp? this distracted me for a while. and puppies.

28nm SHP is a custom process for AMD, why would it appear in Glofo roadmaps? Can you see XboxOne APU on AMD desktop roadmaps?

16nm finfet is from TSMC. On Glofo website you will only find info about their 14nm finfet process.

There is no SHP 20nm on track at Glofo. How could you find info about it?

Steamroller/kaveri never were supposed to be fabbed on 28nm hpp.

After showing that you are very well informed about Glofo processes I can see understand now why you are inventing your (sarcasm -->) "yields are terrible" (<-- sarcasm) argument.

 

[juanrga]

I still recall when an 'expert' here pretended to convince me that "APUs are only for cheap gaming" Yeah sure!

actually they are.

The link that you deleted just explains why this is not true. I also recommend you to learn for what will be used the new Berlin APU. Hint: It is not for gaming.

 

[juanrga]

I agree on that 8 core FX chips are aging well, this is why AMD extended the current FX line up to 2015. The new gen games are scaling up well. This is a new example of scaling up to 8 cores:

Yet on the other side is showing how badly it is actually scaling compared to Intel. THG got stock FX 8350 at i5 3550 performance levels with the best being a 6 core i7 (makes sense as it is a real 6 core)

It is not surprising that from a 3 module 6 thread to a 4 module 8 thread it scales well. What is surprising is that it scales horribly compared to Intel.

Consider that Intel is planning Broadwell this year with Sky Lake next year as well as Haswell-E this year. AMD is falling behind fast and even the consoles wont save them as those optimizations don't work for PC (both consoles have unified memory which alone screws up any optimizations for PC).

Very roughly

1 PD module ~ 1 SB core + HT

Thus, not a surprise that a 12 threads i7 beats a 8-threads FX.

The reason why the $1000 chip is faster than cheaper FX-8350 doesn't have anything to do with any vacuous discussion about the definition of core. The cores in a CMT design are so real as the cores in a CMP design, just each are a different kind of cores.

The reason why the $1000 chip is faster is because it has more execution units, more cache, more... If AMD did release a module with doubled execution units (e.g. 4 ALU+ 4 AGU per core) then a 4-module chip would match a SB six-core, independently of if you consider AMD cores real or not.

Precisely I am analyzing now a supposed die shot of a new module that shows doubled execution units compared to Steamroller module. I believe it is could be Excavator module, but I cannot verify.

The above graphs are more good news for FX-8000/9000 users. About two years ago, the FX-8350 had performed as the FX-4170 and both would be slower than the i3. The prediction made by some of us that modern games would be better threaded (thanks to new consoles) and would run better on a 8-core FX than a 4-core FX turned out to be true.

Excavator is on track to compete against Broadwell. With better threaded game and MANTLE, AMD desktop will remain competitive up to that SkyBridge/K12 see the light.

Now a fan site like say AMDZone, or IntelZone those are just biased no matter what.

IntelZone = IntelTech?

 

[de5_Roy]

Really? Are you sure that APUs have more components than CPU cores? If what you say is true this is revolutionary news. :lol: Sarcasm apart, the CPU on Kaveri occupies about a 50% of the die. Yield problems would affect to the CPU and the almost absence of dual-cores implies that this is not the case.

see, you keep trying to ignore the igpu, which contains over 40% of the die. your poor attempt at sarcasm can't hide that fact.

In reality I was being sarcastic. Again let me help you

http://www.urbandictionary.com/define.php?term=sarcastic

in reality, you got caught lying, red handed, once again. you feigning sarcasm can't hide that fact.

and you ignored the argument reply

i see you have nothing technologically relevant, so you claim ignoring. i did reply. i have nothing new to add to that as you have nothing relevant to add to your statement.

Richland mobile turbo did span 2.6--3.5GHz whereas Kaveri mobile turbo spans 3.0--3-6GHz. I see higher clocks for this new gen.
Your Kaveri clocks are lower because (sarcasm -->) "yields are terrible" (<-- sarcasm) kind of argument doesn't fit too well with reality.

mobile richland igpu 8650G's max. turbo clockrate is at 720mhz with 384 max. shaders.
once again you avoid the igpu, which traded off it's clockrate. mobile kaveri's lowest cpu max. turbo clockrate is 3.0ghz on the 1m/2c sku, topping out at 3.6 ghz with fx7600p. it's igpu max. turbo clockrate is at 686mhz for 512 shaders; no mention of base clockrate (richland a10 had 533 mhz). the next sku, fx 7400P, has lower cpu clockrate, less shader, still can't have higher igpu clockrate. that wasn't the case with richland, which was fabbed on a mature process. llano actually maintained full range of shaders for top mobile skus and decent clockrates despite being a revolutionary new design on an immature process. it remains to be seen how much of the turbo clcokrates (cpu and igpu) these apus can hold in real devices.

you're repeating the word "sarcasm" now. i'll address this later in this post.

Troubles with FDSOI doesn't necessarily translate to bulk.

why do you keep calling their 32nm node fdsoi?

What "design"?

steamroller, and kaveri.

Cof Cof, but what is being discussed is something different, Capt. obvious.

yes, and you cut off that part where i clarify your attempting (and failing) to start another straw man argument.

I see that your irony detector continues to fail. I wonder why You also miss some posts on purpose for being able to post that of above. Your tactic is not new, thus I am not really impressed.

there is no irony, sarcasm to detect. your repeated mentioning of sarcasm can't hide the fact that you got caught lying. and now you're piling more lies (e.g. "it was sarcasm") on top of that. i never missed any addresseable part of your post unless it was a very late edit. my replies aren't meant to impress you, they're simply replies. if you're looking for impressive, look elsewhere.

For your information, I have been discussing in this thread some aspects of Skylake such its new graphics architecture when discussing which will be the competition of AMD new products. If the rumor of Larrabee coming back on Skylake is true, then Intel will be not only fighting on pure graphics workloads but could treat AMD HSA.

Neo-heterogeneity vs HSA

oh those, those never fit with the thread context because indepth info is unavailable as of now.

28nm SHP is a custom process for AMD, why would it appear in Glofo roadmaps? Can you see XboxOne APU on AMD desktop roadmaps?

amd may be the only customer for 28nm shp, for that both amd and glofo optimized their designs and tools but that does not mean the process was amd-exclusive. and it did appear on glofo roadmaps. console socs, otoh, are co-owned by console vendors. this is yet another straw man argument.

16nm finfet is from TSMC. On Glofo website you will only find info about their 14nm finfet process.

yeah, i found it out late in googling. there was an article about tsmc and glofo's processnodes but ambiguously mentioned 20nm and 16nm which lead to my confusion and error.

There is no SHP 20nm on track at Glofo. How could you find info about it?

from this B.S.ofnews link:

During the GlobalFoundries Technology Conference (GTC) at the end of last month, Gregg Bartlett, senior VP of technology, presented the company’s latest technology roadmap.

http://www.brightsideofnews.com/2011/09/11/rumors-14nm-node-and-450mm-wafers-by-2015/
if this is a fake, then i've been misdirected. that'd be the last time i'd believe B.S.ofnews content. it was a 2011 link, processes take time to develop and ramp up for production, so the timeline matches for the node to be production-ready. oh well...

Steamroller/kaveri never were supposed to be fabbed on 28nm hpp.

After showing that you are very well informed about Glofo processes I can see understand now why you are inventing your (sarcasm -->) "yields are terrible" (<-- sarcasm) argument.

sr/kvr was on track for 28nm hpp for a long time before glofo's failure. i am not very well informed about chip manufacturing. i just learned what doublegate/trigate really means and how it's formed. reading stuff takes a lot of time.

if you disagree with what i write, that's fine by me - agree to disagree. but straw man arguments won't work. i appreciate sarcasm, too, generally being subject to it for a long time - that's how i can tell when it's not sarcasm, just failed attempt to hide falsehood. i'll call out such attempts the moment i see them.

 

[wh3resmycar]

I still recall when an 'expert' here pretended to convince me that "APUs are only for cheap gaming" Yeah sure!

actually they are.

The link that you deleted just explains why this is not true. I also recommend you to learn for what will be used the new Berlin APU. Hint: It is not for gaming.

i don't recall deleting any link. but for the sake of what's "actually" out there, APUS are indeed for cheap gaming.

 

[juanrga]

I still recall when an 'expert' here pretended to convince me that "APUs are only for cheap gaming" Yeah sure!

actually they are.

The link that you deleted just explains why this is not true. I also recommend you to learn for what will be used the new Berlin APU. Hint: It is not for gaming.

i don't recall deleting any link. but for the sake of what's "actually" out there, APUS are indeed for cheap gaming.

Well maybe you deleted it unconscientiously , but the link is still in my original message, whereas is not in your reply

And you continue without understanding the concept of APU, despite my recommendation to learn what the Berlin APU is for.

 

[juanrga]

Really? Are you sure that APUs have more components than CPU cores? If what you say is true this is revolutionary news. :lol: Sarcasm apart, the CPU on Kaveri occupies about a 50% of the die. Yield problems would affect to the CPU and the almost absence of dual-cores implies that this is not the case.

see, you keep trying to ignore the igpu, which contains over 40% of the die. your poor attempt at sarcasm can't hide that fact.

You continue without get it. It is not about ignoring the gpu, it is about if there is "terrible yield" (<-- sarcasm) problems they would appear in CPU is less modular than the GPU by a ratio of 2:8.

In reality I was being sarcastic. Again let me help you

http://www.urbandictionary.com/define.php?term=sarcastic

in reality, you got caught lying, red handed, once again. you feigning sarcasm can't hide that fact.

I see your inability to get the concept of sarcasm. I must confess I am not surprised.

Richland mobile turbo did span 2.6--3.5GHz whereas Kaveri mobile turbo spans 3.0--3-6GHz. I see higher clocks for this new gen.
Your Kaveri clocks are lower because (sarcasm -->) "yields are terrible" (<-- sarcasm) kind of argument doesn't fit too well with reality.

mobile richland igpu 8650G's max. turbo clockrate is at 720mhz with 384 max. shaders.
once again you avoid the igpu, which traded off it's clockrate. mobile kaveri's lowest cpu max. turbo clockrate is 3.0ghz on the 1m/2c sku, topping out at 3.6 ghz with fx7600p. it's igpu max. turbo clockrate is at 686mhz for 512 shaders; no mention of base clockrate (richland a10 had 533 mhz). the next sku, fx 7400P, has lower cpu clockrate, less shader, still can't have higher igpu clockrate. that wasn't the case with richland, which was fabbed on a mature process. llano actually maintained full range of shaders for top mobile skus and decent clockrates despite being a revolutionary new design on an immature process. it remains to be seen how much of the turbo clcokrates (cpu and igpu) these apus can hold in real devices.

you're repeating the word "sarcasm" now. i'll address this later in this post.

Richland uses VLIW4, Kaveri uses GCN. GCN is a more efficient arch and doesn't need to be clocked so high. This is why with lower clocks the new GPU is much more powerfull.

The lower clocks are not due to "terrible yields" (<-- sarcasm). Again check the CPU...

Troubles with FDSOI doesn't necessarily translate to bulk.

why do you keep calling their 32nm node fdsoi?

I keep calling things as they are, unlike you.

What "design"?

steamroller, and kaveri.

Ok, then you are plain wrong.

I see that your irony detector continues to fail. I wonder why You also miss some posts on purpose for being able to post that of above. Your tactic is not new, thus I am not really impressed.

there is no irony, sarcasm to detect. your repeated mentioning of sarcasm can't hide the fact that you got caught lying. and now you're piling more lies (e.g. "it was sarcasm") on top of that. i never missed any addresseable part of your post unless it was a very late edit. my replies aren't meant to impress you, they're simply replies. if you're looking for impressive, look elsewhere.

In reality this new attempt to personal attacks began when you found a typo, but were unable to detect it, since then you are posting laughable attempts to continue attacking me, because you half brain is unable to accept that you failed the first time.

For your information, I have been discussing in this thread some aspects of Skylake such its new graphics architecture when discussing which will be the competition of AMD new products. If the rumor of Larrabee coming back on Skylake is true, then Intel will be not only fighting on pure graphics workloads but could treat AMD HSA.

Neo-heterogeneity vs HSA

oh those, those never fit with the thread context because indepth info is unavailable as of now.

In your opinion? Maybe, but left rest of people decide.

28nm SHP is a custom process for AMD, why would it appear in Glofo roadmaps? Can you see XboxOne APU on AMD desktop roadmaps?

amd may be the only customer for 28nm shp, for that both amd and glofo optimized their designs and tools but that does not mean the process was amd-exclusive. and it did appear on glofo roadmaps. console socs, otoh, are co-owned by console vendors. this is yet another straw man argument.

In your imagination AMD custom SHP process appears in roadmaps... reality is different. This is why you cannot find it.

16nm finfet is from TSMC. On Glofo website you will only find info about their 14nm finfet process.

yeah, i found it out late in googling. there was an article about tsmc and glofo's processnodes but ambiguously mentioned 20nm and 16nm which lead to my confusion and error.

Excuses. You have simply no idea, but you were spreading FUD.

There is no SHP 20nm on track at Glofo. How could you find info about it?

from this B.S.ofnews link:

During the GlobalFoundries Technology Conference (GTC) at the end of last month, Gregg Bartlett, senior VP of technology, presented the company’s latest technology roadmap.

http://www.brightsideofnews.com/2011/09/11/rumors-14nm-node-and-450mm-wafers-by-2015/
if this is a fake, then i've been misdirected. that'd be the last time i'd believe B.S.ofnews content. it was a 2011 link, processes take time to develop and ramp up for production, so the timeline matches for the node to be production-ready. oh well...

It is not fake. It is the old roadmap, presented at 2011 conference. Both the 28nm SHP and the 20nm SHP in that ancient roadmap were canceled. And a new 28nm SHP process custom for AMD was developed.

:lol:

Steamroller/kaveri never were supposed to be fabbed on 28nm hpp.

After showing that you are very well informed about Glofo processes I can see understand now why you are inventing your (sarcasm -->) "yields are terrible" (<-- sarcasm) argument.

sr/kvr was on track for 28nm hpp for a long time before glofo's failure. i am not very well informed about chip manufacturing. i just learned what doublegate/trigate really means and how it's formed. reading stuff takes a lot of time.

This is all in your imagination.

 

[juanrga]

First reviews of Haswell refresh confirm very small overclocking advantage if any

http://www.pcper.com/reviews/Processors/Intel-Core-i7-4790K-Devils-Canyon-Review-and-Overclocking/Overclocking-Experience

Intel advertised "over 5.5.GHz on air" at Computex was in reality only 15MHz over 5.5Ghz and obtained on water with special golden chip and only 1c1t.

Real world sample with all cores actives hits 4.7GHz using water and too high voltage.

The "over 5.5GHz on air" was pure lie. I notice that the same people in this thread who has been accusing AMD of lying about jaguar servers or even of "stolen mantel" is now silent against this blatant lie from Intel. I wonder why.

 

[8350rocks]

The discussion about AMD losing things by going bulk was exaggerated, but it is true. The general consensus on Kaveri desktop is that the IPC gains were wiped out by the fact that Kaveri can't clock as high. I don't see too many architectural changes that would have caused that. It seems far more likely to be related to the process instead.

Yeah, there are good OCing Kaveri's out there. But the fact remains that at the end of the day we had to see a stock clockspeed reduction thanks to going bulk and that the point of diminishing returns when raising TDP is much lower.

AMD lost a lot going from Richland to Kaveri in terms of clock speeds. It wasn't some catastrophic unable to break 3ghz type of event, but it was enough to hurt overall CPU performance.

Evidently part of the IPC increment introduced by Steamroller architectural improvements was dropped by the lower base clock. Nobody is negating that! What is being discussed here is why AMD underclocked Kaveri.

We know that Kaveri average OC on air is 4.5GHz. We have seen Kaveri benchmarked @4.7GHz:

http://www.hardcoreware.net/kaveri-7850k-overclocked-benchmarks/

Anand showed that Kaveri can be OC up to ~4.2GHz without changing the voltage (1.25V) and with almost same power consumption 147W vs 148W

We also know that early talks and docs from AMD mentioned explicitly that Kaveri was a 100W APU originally.

Thus the question is why did AMD downrated Kaveri to 95W and reduced the base frequency from the expected 4.0GHz (check my BSN article) to final 3.7GHz? Why did AMD do that when Kaveri can hit 4.5/4.7GHz on air?

This is the million dollar question that nobody is answering!

Because the process of bulk does not scale well, some silicon will draw less power at 4.0 GHz than others, they have to be able to guarantee TDP will fall within acceptable MoE for stock samples.

THAT is why they clocked it lower. Because they could not guarantee the TDP would conform on ALL samples, where as with the previous process they could.

 

[8350rocks]

Spoiler

I agree on that 8 core FX chips are aging well, this is why AMD extended the current FX line up to 2015. The new gen games are scaling up well. This is a new example of scaling up to 8 cores:

Yet on the other side is showing how badly it is actually scaling compared to Intel. THG got stock FX 8350 at i5 3550 performance levels with the best being a 6 core i7 (makes sense as it is a real 6 core)

It is not surprising that from a 3 module 6 thread to a 4 module 8 thread it scales well. What is surprising is that it scales horribly compared to Intel.

Consider that Intel is planning Broadwell this year with Sky Lake next year as well as Haswell-E this year. AMD is falling behind fast and even the consoles wont save them as those optimizations don't work for PC (both consoles have unified memory which alone screws up any optimizations for PC).

Very roughly

1 PD module ~ 1 SB core + HT

Thus, not a surprise that a 12 threads i7 beats a 8-threads FX.

The reason why the $1000 chip is faster than cheaper FX-8350 doesn't have anything to do with any vacuous discussion about the definition of core. The cores in a CMT design are so real as the cores in a CMP design, just each are a different kind of cores.

The reason why the $1000 chip is faster is because it has more execution units, more cache, more... If AMD did release a module with doubled execution units (e.g. 4 ALU+ 4 AGU per core) then a 4-module chip would match a SB six-core, independently of if you consider AMD cores real or not.

Precisely I am analyzing now a supposed die shot of a new module that shows doubled execution units compared to Steamroller module. I believe it is could be Excavator module, but I cannot verify.

Spoiler

The above graphs are more good news for FX-8000/9000 users. About two years ago, the FX-8350 had performed as the FX-4170 and both would be slower than the i3. The prediction made by some of us that modern games would be better threaded (thanks to new consoles) and would run better on a 8-core FX than a 4-core FX turned out to be true.

Excavator is on track to compete against Broadwell. With better threaded game and MANTLE, AMD desktop will remain competitive up to that SkyBridge/K12 see the light.

Now a fan site like say AMDZone, or IntelZone those are just biased no matter what.

IntelZone = IntelTech?

K12.

EDIT:

@juanrga: 32nm node was PD-SOI! NOT FD-SOI! If you are going to "call things as they are", do it right next time.

 

[de5_Roy]

You continue without get it. It is not about ignoring the gpu, it is about if there is "terrible yield" (<-- sarcasm) problems they would appear in CPU is less modular than the GPU by a ratio of 2:8.

see, i never typed the phrase "terrible yields", didn't even insinuate such. it's your own made up Lie that you keep putting on me. repeated writing "sarcasm" beside it doesn't make you any less of a Liar and a Troll. you just got called out after getting caught. now you're piling more lies upon the old one to cover up. you're being pitiful.

the igpu takes up around 47% of the die area and has 512 max. shaders in it. if amd had good yields, they'd have low leakage, full 2 modules, fully functional, high clocked igpu with all it's components. they might have to clock them lower to fit into low tdp, but parts would be functional nonetheless. the fact that they have only one fully enabled dt and one mobile sku speak for themselves. defects don't appear only in the cpu are, the whole die is subject to defects.

I see your inability to get the concept of sarcasm. I must confess I am not surprised.

sarcasm has nothing to do with it. you got caught lying, got called out, now you're embarrassing yourself by trolling about it.

Richland uses VLIW4, Kaveri uses GCN. GCN is a more efficient arch and doesn't need to be clocked so high. This is why with lower clocks the new GPU is much more powerfull.

The lower clocks are not due to "terrible yields" (<-- sarcasm). Again check the CPU...

ah, but this has little to do with gcn's uarchitectural efficiency or performance, it has to do with whether glofo can fab it well or not. and evidently, they're having issues. that's something amd themselves pointed out during dt kaveri launch. glofo has always fumbled anything with gpu. amd wanted them to fab some of the older gcn 1.0 gpus but they had to cancel them and have tsmc fab the whole lineup.

I keep calling things as they are, unlike you.

clarify what "as they are means".

Ok, then you are plain wrong.

simply saying that doesn't make you right. since you evidently failed to back your claim up with a logically sound, credible explanation, i don't see it worth taking into consideration.

In reality this new attempt to personal attacks began when you found a typo, but were unable to detect it, since then you are posting laughable attempts to continue attacking me, because you half brain is unable to accept that you failed the first time.

i only pointed out your mistake, that's done long time ago.
your trolling started when i caught you lying about my statement (in a reply to someone else) and called you out on it. then you started to add even more lies to that and trolling. you've cried "personal attack" before, every time someone successfully calls you out on your lies or mistakes. i never attacked you personally, have always argued your ideas. i don't care about your person (i know it sounds a bit harsh but that's how it is). i am sorry that you feel you're being personally attacked when in reality, i am only pointing out your lies and trolling.

oh those, those never fit with the thread context because indepth info is unavailable as of now.
In your opinion? Maybe, but left rest of people decide.

my opinion? no. those statements never fit with thread topic because you'd need to go indepth on how both work and go into deeper speculation like memory access and context switching. vague statements don't count as they're too broad.

Excuses. You have simply no idea, but you were spreading FUD.

if that's what makes you feel better. agree to disagree.

In your imagination AMD custom SHP process appears in roadmaps... reality is different. This is why you cannot find it.

i did find it on official roadmap.

It is not fake. It is the old roadmap, presented at 2011 conference. Both the 28nm SHP and the 20nm SHP in that ancient roadmap were canceled. And a new 28nm SHP process custom for AMD was developed. :lol:

foundries start developing fab. processes years before they start selling asics in the market. i hope someone more knowledgeable can shed more lights on this.

This is all in your imagination.

since this is open to speculation and only amd-insiders can verify, and since i am not an amd insider, let it be in my imagination then...

@8350rocks: i remember reading a pcper.com editorial stating glofo 32nm was pd-soi and 32nm would be as far as pd-soi would go. but that wasn't official statement so i can't use it as proof. amd's website only mentions 32nm SOI (no pd/fd). iirc hcl123 earlier also explained 32nm to be pd-soi.