AMD CPU speculation... and expert conjecture

[richardgal]

It will come out as another complete failure if they don't change platform. You can't build a castle from a pile(driver) of ***. (ba dum tss!)

But seriously, if they keep on messing around with the FX arch, they might aswell just "line up on the rooftop and commit seppuku."

 

[amdfangirl]

Wasn't Jaguar as a whole the successor to Bobcat? Which means an OOO CPU and improvements tweaks?

Cheers!

Pretty sure Bobcat was OOO.

 

[griptwister]

It will come out as another complete failure if they don't change platform. You can't build a castle from a pile(driver) of ***. (ba dum tss!)

But seriously, if they keep on messing around with the FX arch, they might aswell just "line up on the rooftop and commit seppuku."

Those are mighty strong words coming from a guy who's signature build consists on an i3

 

[sarinaide]

My bet is that it'll come close to Ivy, but won't beat Haswell in CPU. If still 28nm and not 20nm in 2H 2014, I'd also bet any iGPU improvements might only amount to 30% or less, depending on how big a die and TDP AMD is willing to go with.

If Intel had been willing to make Ivy an actual 95W TDP CPU, instead of 77W, I'd bet that the HD4K would have matched or beaten Trinity's as they could have stuck 24 or maybe even 32 EU's on it..

As to x86, most AMD fans would be really delighted if it matched SB/IB in IPC and general x86 performance so I would take that happily.

On the iGPU debates, don't think Intel releasing HD4K would have actually know what Trinity had to offer, the focus was closing up to top Llano parts which they still linger behind, in some aspects HD4K is relatively close to Trinity, but in most important aspects HD4k is around 3-4x slower, not that this is disappointing but one must bear in mind that AMD are accused of lowballing with what is regarded as extremely low end and inefficient VLIW parts, with no reason to be there.

Trinity vs Llano brought general 35% improvements to the iGPU on a aging VLIW architecture which is neither efficient in power or performance. Richland is still believed to go ahead with all the slated changes just minus Steamroller cores, so that will see AMD attempting to more than double the bandwidth and improve performance in the same TPD window. The changer here is GCN, as revolutionary as SB was to x86, take my long favorite HD6970 compared to its replacement HD7970 and the performance differential is almost 3x and uses a lot less power. If Richland is correct it will use a HD8660D on the highest end part, with the HD8750 said to be 15% faster than the HD7770 and use 45% less power, a HD8660D will be a watered down version of that. I did say Richland would at least bring 25% gains on iGPU performance over Trinity, but that could very well be significantly higher, knowing AMD's experience in GPU technology this is very possible and will likely see it dominate for some time yet. I initially heard somewhere around 700-900 radeon stream processors on the highest end D part, though its been said to cap out around 720 stream processors, which is double 7660D at half the power, factor in unifying the IMC between and other tweeks to bandwidth, it is very possible that this could bridge to mainstream quite efficiently. Right now other changes to the iGPU/Memory interface is unknown but I doubt it would be any less pertinent to AMD to improve this considering it is the bread and butter.

 

[-Fran-]

Pretty sure Bobcat was OOO.

Yep, that's my point. AMD has no other OOO design AFAIK.

As to x86, most AMD fans would be really delighted if it matched SB/IB in IPC and general x86 performance so I would take that happily.

On the iGPU debates, don't think Intel releasing HD4K would have actually know what Trinity had to offer, the focus was closing up to top Llano parts which they still linger behind, in some aspects HD4K is relatively close to Trinity, but in most important aspects HD4k is around 3-4x slower, not that this is disappointing but one must bear in mind that AMD are accused of lowballing with what is regarded as extremely low end and inefficient VLIW parts, with no reason to be there.

Trinity vs Llano brought general 35% improvements to the iGPU on a aging VLIW architecture which is neither efficient in power or performance. Richland is still believed to go ahead with all the slated changes just minus Steamroller cores, so that will see AMD attempting to more than double the bandwidth and improve performance in the same TPD window. The changer here is GCN, as revolutionary as SB was to x86, take my long favorite HD6970 compared to its replacement HD7970 and the performance differential is almost 3x and uses a lot less power. If Richland is correct it will use a HD8660D on the highest end part, with the HD8750 said to be 15% faster than the HD7770 and use 45% less power, a HD8660D will be a watered down version of that. I did say Richland would at least bring 25% gains on iGPU performance over Trinity, but that could very well be significantly higher, knowing AMD's experience in GPU technology this is very possible and will likely see it dominate for some time yet. I initially heard somewhere around 700-900 radeon stream processors on the highest end D part, though its been said to cap out around 720 stream processors, which is double 7660D at half the power, factor in unifying the IMC between and other tweeks to bandwidth, it is very possible that this could bridge to mainstream quite efficiently. Right now other changes to the iGPU/Memory interface is unknown but I doubt it would be any less pertinent to AMD to improve this considering it is the bread and butter.

Like palladin said way way back, it's not about hardware on Intel's side. It's software. Since they have the process advantage, they can pretty much slap all the things they want to their designs (exaggeration, but you know what I mean, haha). They have to work out the issues with the support and driver's performance gradually; they can't produce a massive iGPU part without proper understanding and scaling in the development of the companion software.

Final derailment: CBO predicts a 9.1% unemployment rate if fiscal cliff occurs. Deficit would be cut in half though.

Don't worry, FOXCONN said they'll put a plant in the States. I'm sure they'll solve everything with child labor (?)

Cheers! xP

 

[sarinaide]

Like palladin said way way back, it's not about hardware on Intel's side. It's software. Since they have the process advantage, they can pretty much slap all the things they want to their designs (exaggeration, but you know what I mean, haha). They have to work out the issues with the support and driver's performance gradually; they can't produce a massive iGPU part without proper understanding and scaling in the development of the companion software.

Cheers! xP

I did raise this issue before, software developers don't have any interest in intel iGPU support, looking at my games boxes the "This game may not support intel integrated graphics solutions please consult blah blah blah". 40EU's, 9 Billion EU's, the issue is one company knows how to produce the hardware and software/firmware to support it, Intel don't have that luxury and I am sure as day Nvidia and AMD will not let them get the IP's to them. AMD Vision opperates of Catalyst drivers so driver support is not an issue. I can see AMD vision escalating quite a lot over the years to come, no doubts that SR will feature a iGPU that is by and large lower mainstream level, which is very impressive all things considered. The more important issue for the long term of AMD is not the gaming performance but the HSA performance which is the most fundamental reason for the APU and Vision.

 

[mayankleoboy1]

Intel is not interested in HSA, AFAICT.
They are more interested in adding specialised hardware, aka Quicksync. But that has even lesser support than HSA.

 

[gamerk316]

I did raise this issue before, software developers don't have any interest in intel iGPU support, looking at my games boxes the "This game may not support intel integrated graphics solutions please consult blah blah blah". 40EU's, 9 Billion EU's, the issue is one company knows how to produce the hardware and software/firmware to support it, Intel don't have that luxury and I am sure as day Nvidia and AMD will not let them get the IP's to them. AMD Vision opperates of Catalyst drivers so driver support is not an issue. I can see AMD vision escalating quite a lot over the years to come, no doubts that SR will feature a iGPU that is by and large lower mainstream level, which is very impressive all things considered. The more important issue for the long term of AMD is not the gaming performance but the HSA performance which is the most fundamental reason for the APU and Vision.

Games don't "target" GPU's. They follow some minimum graphics specification (say, DX10, SM3.0), and any GPU that meets those requirements should be able to play the game (performance aside at least). Intels problem for the longest time was very slow adoption of new GPU standards (SM3.0 support, for example), leading to lots of Intel GPU's not being supported for many games.

 

[esrever]

Yep, that's my point. AMD has no other OOO design AFAIK.

Everything AMD made since the K5 has been OoO. Only x86 cpus on the market without OoO is atoms.

 

[-Fran-]

Everything AMD made since the K5 has been OoO. Only x86 cpus on the market without OoO is atoms.

Oh really? I did not know that. Color me surprised. I never thought Atom was the only modern IO CPU 0.0

Cheers!

 

[fazers_on_stun]

Everything AMD made since the K5 has been OoO. Only x86 cpus on the market without OoO is atoms.

Dunno if Atom is the only non-OoO x86 out there, but yes it is currently in-order to lower the power. IIRC I read somewhere that out-of-order processing takes a huge amount of power, maybe 30% or more of the total CPU power consumption, as well as large amounts of die area depending on how effective it's supposed to be. Itanium (not x86 obviously) is in-order and IIRC requires the compiler to resolve branches and other data dependencies to avoid stalling out.

Thought that Atom was going OoO either at 22nm or 14nm in order to speed it up. With the lower power consumption and the smaller transistor size, Intel should be able to keep the power draw low enough for the phone & tablet market. The latest ARM CPUs are OoO as well.

 

[mayankleoboy1]

Dunno if Atom is the only non-OoO x86 out there, but yes it is currently in-order to lower the power. IIRC I read somewhere that out-of-order processing takes a huge amount of power, maybe 30% or more of the total CPU power consumption, as well as large amounts of die area depending on how effective it's supposed to be. Itanium (not x86 obviously) is in-order and IIRC requires the compiler to resolve branches and other data dependencies to avoid stalling out.

Thought that Atom was going OoO either at 22nm or 14nm in order to speed it up. With the lower power consumption and the smaller transistor size, Intel should be able to keep the power draw low enough for the phone & tablet market. The latest ARM CPUs are OoO as well.

It would be much better that all the heavy thing is done at the compile time. Leaving teh main CPU in-order and with a smaller die/power.

 

[viridiancrystal]

http://www.tgdaily.com/hardware-features/63917-report-ps4-powered-by-amds-liverpool-apu
^interesting rumor, but I'm keeping my salt close by.

http://www.xbitlabs.com/news/cpu/display/20121108224721_AMD_Substantially_Cuts_Prices_on_Entry_Level_Desktop_Chips.html
^expected, but informative.

 

[fazers_on_stun]

It would be much better that all the heavy thing is done at the compile time. Leaving teh main CPU in-order and with a smaller die/power.

I would think also much more expensive too, but I dunno. Besides, like it or nutz , we're still stuck with x86 legacy instructions after what - 30 years now. AFAIK the backwards compatibility doesn't take up too much die area or transistor budget - somebody mentioned maybe a million or two extra transistors which is tiny on a billion-plus CPU, a while back.

Or we could all go out and buy Itaniums . IIRC it's still a $4BN a year business for Intel so I doubt they'll be giving it up despite Oracle trying to kill it off..

 

[fazers_on_stun]

http://www.tgdaily.com/hardware-features/63917-report-ps4-powered-by-amds-liverpool-apu
^interesting rumor, but I'm keeping my salt close by.

http://www.xbitlabs.com/news/cpu/display/20121108224721_AMD_Substantially_Cuts_Prices_on_Entry_Level_Desktop_Chips.html
^expected, but informative.

Until some must-have games come out strictly for PS4 or XBox Next, I'm sticking with my PS3 ..

From the 2nd article:

According to Mercury Research, for the first time in several years the share of AMD’s microprocessors on the x86 market dropped to 16.1% in the third quarter of 2012 from 18.8% in Q3 2011. Market share of Intel increased to 83.3% in Q3 2012, up from 80.6% in the same quarter a year before. It is noteworthy that Intel gained on AMD mostly in desktops thanks to the roll-out of Intel Core i-series 3000-family “Ivy Bridge” microprocessors that started in April, 2012. At the time, AMD virtually had nothing to compete with against its arch-rival in Q3 2012 as it A-series “Llano” chips and FX-series “Bulldozer” central processing units were either slow or suffered from low supply of mainboards. The price-slash should give AMD an opportunity to fight back the unit market share from Intel.

Pretty much what everybody expected from AMD's Q3 earnings report. I suspect their DT share is way, way down from the 41% of a year ago..

 

[noob2222]

Pretty much what everybody expected from AMD's Q3 earnings report. I suspect their DT share is way, way down from the 41% of a year ago..

41%! ... don't think AMD has been that high since the athlon days.

http://www.engadget.com/2011/08/02/amds-market-share-tiptoes-higher-intel-still-ruler-of-the-roos/

states 28.9% in april 2011

and to contradict that statement

http://www.techpowerup.com/164915/AMD-Gains-CPU-Market-Share-at-Intel-s-Expense-in-Q1.html

43% for 2 years in april 2012 ...

and to contradict that statement

http://www.tomshardware.com/news/amd-intel-cpu-processor,15041.html

toms said 26% in march 2012

somebody is lying and I'd place bets its on the people who claimed 43%, many more sites state DT market share was only in the 20s.

 

[NoUserBar]

I'm sure they look at everything they can think of. It's not stated that AMD's Steamroller is coming out on the same socket, but AMD has also not stated or even rumored why Steamroller would NOT come out on both sockets next year either. Or at least for FM2.

(double neg is a gammer thing not a logic thing lols)

Although with Richland coming out on FM2, It's possible that richland will be on FM2 then Steamy on a different chipset/socket I suppose..

....But I've been thinking, there is a new process in the mix for a MUCH cheaper way to produce CPU's and other nanotech.

Link here: http://phys.org/news/2012-08-small-scale-solution-large-scale-impact.html

If this takes off then far more companies will be able to, perhaps, enter the microchip business.

 

[Cazalan]

Thought that Atom was going OoO either at 22nm or 14nm in order to speed it up. With the lower power consumption and the smaller transistor size, Intel should be able to keep the power draw low enough for the phone & tablet market. The latest ARM CPUs are OoO as well.

Next year the 22nm Atoms will be OoO and a full SoC. No extra chip sets needed.

2013 is looking to be quite active. New 64bit ARM chips, Haswell/Avoton, Steamroller APUs.

Well have to see how well ARM scales up to real server workloads, and if Intel can scale the wattage down.

 

[skitz9417]

i dont think tht amd will beat intel cpu when it comes out in 2014

 

[jdwii]

http://www.tgdaily.com/hardware-features/63917-report-ps4-powered-by-amds-liverpool-apu
^interesting rumor, but I'm keeping my salt close by.

http://www.xbitlabs.com/news/cpu/display/20121108224721_AMD_Substantially_Cuts_Prices_on_Entry_Level_Desktop_Chips.html
^expected, but informative.

since steamroller is most likely delayed that is all what's going to happen is price cuts Intel wont ever do it but a I3 that is unlocked would make Amd pretty irrelevant in the CPU market if Intel beats Amd in GPU integrated performance Amd will be done for.

 

[griptwister]

i dont think tht amd will beat intel cpu when it comes out in 2014

u dont think tht?

Anyways, I think AMD Needs to release Steamroller around the same time Intel releases the 2nd Tick. The clock is running out, AMD has no time... As someone mentioned earlier, If Intel does beat AMD in the APU department... AMD will be finished.

 

[skitz9417]

yea they might be finshed but who know

 

[fazers_on_stun]

41%! ... don't think AMD has been that high since the athlon days.

There's a number of ways to measure marketshare - # chips shipped, # chips sold and $ revenue. So I don't think anybody is outright lying, they're just not clear on how they are measuring it. It's possible the lower percentage stems from revenue since AMD in general has lower ASPs.

Anyway, what's important here is the trend downwards, albeit in the stagnant DT market.

 

[fazers_on_stun]

Next year the 22nm Atoms will be OoO and a full SoC. No extra chip sets needed.

2013 is looking to be quite active. New 64bit ARM chips, Haswell/Avoton, Steamroller APUs.

Well have to see how well ARM scales up to real server workloads, and if Intel can scale the wattage down.

Well the rumors are that Steamy won't see the light of day until sometime in 2014, so basically as far as desktop goes, what you see right now is what you'll be seeing this time next year.

Intel has already stated they are going for sub-10W Haswells, so that'll be interesting to compare to ARM. While I'm willing to trade performance on my Asus transformer ePad for all-day battery life, it would be nice to have both ..

 

[jdwii]

Well the rumors are that Steamy won't see the light of day until sometime in 2014, so basically as far as desktop goes, what you see right now is what you'll be seeing this time next year.

Intel has already stated they are going for sub-10W Haswells, so that'll be interesting to compare to ARM. While I'm willing to trade performance on my Asus transformer ePad for all-day battery life, it would be nice to have both ..

Most likely if they do have 10 watt chips they will be cutting performance left and right.


Either way haswell is only going to be 10% better on average(higher clock speed, better turbo or IPC? in CPU benchmarks even according to Intel troll sites,

Haswell is all about power consumption upgrades and Graphics. I'm wondering if it will overclock well on the desktop lets see if they can get 5.0Ghz on air with a mid-range cooler i'll be like wow.

As for steamy i'm honestly disappointed Trinity is such a great product and the CPU inside of it really does fight a I3 pretty well its still weaker but not by much in most cases and steamy would of allowed Amd to pull ahead but of course something that great has to be delayed and when it comes out its going to get outclassed by Intel once again. And the stock continues to fall and the lay off's will probably be another 15% by this time next year. What's that CEO doing again? Starting to think Steve ballmer is better.