AMD CPU speculation... and expert conjecture

[Reepca]

If I understand correctly, a big core probably can get more done per unit of time than a small core? That's what the name implies to me anyway. In which case, I think I'd prefer big cores. At least in the APU line, since anything that can be parallelized can probably be better-parallelized via the GPU. Since some loads are unavoidably serial, I'd rather they fix that single-threaded weak spot than put more cores in. Granted some stuff is task-parallel, but 4-6 cores should be enough to cover any task-parallel loads a single application would probably use.

Still on the HSA hype train.

 

[esrever]

@tourist: I was referring to the 28nm apus.

@cazalan: Zen will not entirely be APUs, they can share a socket with APUs because all the gpu is on die in APUs.

Yea, the lack of L3 really does not help FM2/FM2+ CPU's.

but where doses it make a difference ?

. cinebench 11.5 s/t
fx4130 = 0.92 8mb l3
6600k = 1.07
3850k = 0.89
11.5 m/t
fx4130 = 3.09 8mb l3
6600k = 3.4
3850k = 3.47

You are comparing bulldozer to piledriver. Also tests like this will have very little need for a L3 since most of the data should fit into the L2 no problem. If you were running actual programs in parallel, the L3 will get use. It is why the Piledriver FX chips outperforms kaveri's cpu in gaming by a slight factor even tho kaveri is using steamroller.

 

[8350rocks]

@juan:

When they were realistically considering 16 core cpus for zen, that was true, it had to be. That number is now half at 8 cores meaning what happens to those cores when there is no gpu on die?

Hopefully they go back to having some L3 because even the Apple A8 has 4MB L3.

Well...that was the plan for the non GPU parts. It will not just be deactivated or fused off silicon. The entire die is supposed to be used on those parts as I understand it (lengthy discussion over the matter...).

 

[juanrga]

@juan:

When they were realistically considering 16 core cpus for zen, that was true, it had to be. That number is now half at 8 cores meaning what happens to those cores when there is no gpu on die?

Even assuming it is true, I doubt it, it could mean that the cores are now bigger, or that the die size is now smaller or that there is more LLC, or that there are more stuff on the SoC or...

Since you are saying nothing, really, I will add to my guess what I think will be the Zen compute unit. If I am not mistaken, it implies there are no six-core processors

• 
  ■Mini-core ~5mm2 on Samsung/Glofo 14FF node
  ■~40% higher IPC over Jaguar/Piledriver
  ■2x128 bit FP/SIMD
  ■DDR4; HBM optional
  ■3.0--3.5GHz
  ■CMP (no SMT)
  ■Lego-like design
  ■ 4-cores plus shared L2 cache form the Zen compute unit

 

[juanrga]

Next table is probably the origin of the Sweclockers 'leak' about the Kaveri refresh

Besides the red flags that I mentioned before, it is worth to mention that the 870K 'refresh' has lower frequencies than the existent 860K. Another red flag.

 

[-Fran-]

The 870K looks like a harvested part that clocked way lower than expected, but you can't conclude the "fine" batch will be as bad as that. The other interpretation is that the process is turning out to be quite good and they're not forced to use "good chips" as cannon fodder.

Well, a half baked interpretation, but still plausible.

Also, the 8850K looks fine in my opinion.

Cheers!

 

[juanrga]

The 870K looks like a harvested part that clocked way lower than expected, but you can't conclude the "fine" batch will be as bad as that. The other interpretation is that the process is turning out to be quite good and they're not forced to use "good chips" as cannon fodder.

Well, a half baked interpretation, but still plausible.

Also, the 8850K looks fine in my opinion.

Cheers!

And the 850 that replaces the 840 is also clocked lower and the A6-8550K that replaces the A6-7400K is also clocked lower and so on, and the 65W 8650 is clocked higher than the 95W 8650K...

The 8850K looks fine, except when one considers all the red flags I mentioned before about it, such as 856MHz not being a valid frequency for Kaveri.

Either the whole leak is fake or AMD is running in full crazy mode now!

 

[esrever]

@tourist: I was referring to the 28nm apus.

@cazalan: Zen will not entirely be APUs, they can share a socket with APUs because all the gpu is on die in APUs.

Yea, the lack of L3 really does not help FM2/FM2+ CPU's.

but where doses it make a difference ?

. cinebench 11.5 s/t
fx4130 = 0.92 8mb l3
6600k = 1.07
3850k = 0.89
11.5 m/t
fx4130 = 3.09 8mb l3
6600k = 3.4
3850k = 3.47

You are comparing bulldozer to piledriver. Also tests like this will have very little need for a L3 since most of the data should fit into the L2 no problem. If you were running actual programs in parallel, the L3 will get use. It is why the Piledriver FX chips outperforms kaveri's cpu in gaming by a slight factor even tho kaveri is using steamroller.

stars is in there also. but remember this ? http://www.tomshardware.com/reviews/gaming-fx-pentium-apu-benchmark,3120-9.html
no l3 did not hurt stars as much, it was the low clock

Llano had extra L2 to make up for the lack of L3, the Phenom II x4 955 is 30% faster than Athlon II x4 645 in sc2 while only being clocked 100mhz higher, the only difference is the L3. The L3 does make a difference even if its not always very much.

 

[esrever]

I can't tell the difference between a FX8320 and a A6-3620 either without benchmarks. The difference between CPUs is so small now I doubt anyone can really tell an i7 from a pentium without running benchmarks. Even in most games, if you have a powerful enough GPU, there really isn't that much difference.

 

[con635]

I can't tell the difference between a FX8320 and a A6-3620 either without benchmarks. The difference between CPUs is so small now I doubt anyone can really tell an i7 from a pentium without running benchmarks. Even in most games, if you have a powerful enough GPU, there really isn't that much difference.

Here's one for you, I went from practically an oc'd 750k to an ivy xeon 1230, the xeon crushed all my 750k benches and literally doubled my fps in some scenarios/games. However, if I turn off fps counter your completely right in a blind test I probably wouldn't see the difference 98% of the time. At the time I bought the xeon me and a brother were running 750k/760k with hd7770s, he bought an r9 280 instead, guess who gets a noticeably better gaming experience?

 

[juanrga]

a really good article by barrons.

http://online.barrons.com/articles/amd-bets-with-a-new-stack-of-chips-1422688235

That article would be better suited in the other thread, the one about AMD as company, whereas this thread would be better leave to discussing products, but since you posted here I will reply here.

The article confirms what I said about the new direction of AMD, migrating from PC market towards the embedded/semicustom market:

After a tough slog in recent years, Advanced Micro Devices has begun to make real progress in its transition from semiconductor designer for consumers’ personal computers to supplier of chips powering everything from advanced avionics to the latest Microsoft Xbox.

[...]

The aim is to cut AMD’s reliance on household PCs, whose sales seem to be locked in a decline as consumers opt for tablets and smartphones.

[...]

MORE IMPORTANTLY, AMD is starting to look like a different company. Its business mix has shifted from more than 90% PC-related in 2012 to about 60% at the end of 2014. AMD expects revenues from its computing and graphics unit to level off near those of its enterprise, embedded, and semi-custom group. Computing and graphics includes desktop, notebook, and graphic processors, while the other, faster-growing unit makes chips for servers, game consoles, slot machines, and everything from airplane-cockpit consoles to digital signs towering over Times Square.

It confirms that AMD is in bad financial situation:

Neither AMD’s stock nor its credit-default swaps offer any hint of the recent positive news. The shares have fallen 23% since October, when AMD sharply lowered its forecast for fourth-quarter earnings, and 44% since last summer. Credit-default swaps based on its debt trade in the low 600s, a level suggesting AMD “is in real financial distress,” says Rob Spivey, a director at Valens Credit, an independent research outfit.

[...]

That’s not to minimize AMD’s problems. Since the end of 2012, overall PC sales have dropped at a 5% annualized rate, and revenue from AMD’s PC-related x86 microprocessors has fallen even faster. AMD’s market share fell to 8.7% from 13.5% at the end of 2013. At its low point in 2012, AMD lost $1.2 billion, or 16 cents a share, as revenue fell 17%, to $5.4 billion. Its $2.2 billion of long-term debt exceeds its $2 billion market cap. The reduced fourth-quarter forecast was a reminder of the challenges. Short sellers hold 22% of AMD’s shares.

In short, the article says more of the same, nothing new really...

 

[truegenius]

a really good article by barrons.

http://online.barrons.com/articles/amd-bets-with-a-new-stack-of-chips-1422688235

whenever i see this line

The aim is to cut AMD’s reliance on household PCs, whose sales seem to be locked in a decline as consumers opt for tablets and smartphones.

it makes me angry and i mutter in my head that these analysts do even ever analysed anything in their whole life, why don't they force on the reason more than the sales decline, anyone with sub 80 IQ can see this that sales are declining just by looking at graphs/figures (because it takes more effort, thats why they don't give reasons)
why would an average joe buy something when there is no performance increase or minimal performance increase
we see only 1 new gpu die (top one) and all other just rebranded old thing and performance only increase significantly when die shrink occurs
we see no performance and price/performance and performance/watt increase in cpu performance with new series ( but surely see performance decrease (buldozer) (even in case of intel only new affordable laptops i see in my area are made up of lower clocked "U" series chips, intel atom based pentium with less performance and higher cost)
so how come they expect sales to rise while component performance is falling
not everyone need 1 day battery backup for laptop, battery backup belongs to mobile more

but i case of mobile/tablet, we can get performance of previous year flagship in current low-mid range phone ( but it is also slowing down because now we don't get enough performance increase now, we just get more screen resolution, bigger size, less battery backup )
nexus s 1ghz single core 512mb ram
> nexus 1.2ghz dual 1GB
> nexus 4 1.5ghz quad krait 2GB
> nexus 5 2.26ghz quad krait 2GB (which is not an performance upgrade when we consider performance hungry 5" 1080p )
> nexus 6 2.7ghz quad krait 3GB (which is more performance loss than prebious one if we consider horsepower heeded to handle QHD and 6"

6" ? seriously google, do you give package of (ranging or exceeding) crores (Rupees) to your employs just to increase screen size and resolution
in that case it looks like next nexus will get 7" screen size
hey look next gen nexus spotted nexus 7
and its cheaper than current gen

and guess what, we will see this in future

Analysts says mobile sales declining and people are opting for smart undergarments, smart shoes, floaters, glasses which talks to each other and predicts when you need to pee and reports when you get near or see a toilet and makes log of how much was fallen on shoes and will remind you to flush and wash hands

 

[blackkstar]

I can't tell the difference between a FX8320 and a A6-3620 either without benchmarks. The difference between CPUs is so small now I doubt anyone can really tell an i7 from a pentium without running benchmarks. Even in most games, if you have a powerful enough GPU, there really isn't that much difference.

You won't see a difference between a Pentium G and 5960X unless you're using software that will use a lot of cores.

The barons article is another analyst who doesn't think the DIY market exists. It's another "Dell and Lenovo shipped fewer PCs than they did 5 years ago, PC market is dying!" things.

http://hardware.slashdot.org/story/15/01/13/2110215/pc-shipments-are-slowly-recovering
I don't know what they are talking about Gartner is saying PC shipments from OEMs will rise. And all of this is ignoring DIY market.

In the U.S., PC shipments increased 13.1 percent year-over-year, the fastest increase in four years, thanks to holiday purchases.

There was an increase of 13.1% in PC shipments holiday season 2014 than there was holiday season 2013 in the USA. That is not what a dying market looks like.

Hopefully AMD releases Fiji soon. The GTX 970 blunder is a huge opportunity for AMD and they seem to be capitalizing on it rather well.

 

[jdwii]

I can't tell the difference between a FX8320 and a A6-3620 either without benchmarks. The difference between CPUs is so small now I doubt anyone can really tell an i7 from a pentium without running benchmarks. Even in most games, if you have a powerful enough GPU, there really isn't that much difference.

With a lot of games lately they are really taxing one core pretty heavily were stuttering is quite noticeable on weaker single core performance. I really hope Zen/K12 make this problem go away.

 

[wh3resmycar]

I can't tell the difference between a FX8320 and a A6-3620 either without benchmarks. The difference between CPUs is so small now I doubt anyone can really tell an i7 from a pentium without running benchmarks. Even in most games, if you have a powerful enough GPU, there really isn't that much difference.

cute. i had this r7 260x on an athlon ii x3. worlds apart compared to when i put it on my i5. so i'm sorry but what you're saying is BS.

ps. we own a net cafe.

 

[esrever]

I can't tell the difference between a FX8320 and a A6-3620 either without benchmarks. The difference between CPUs is so small now I doubt anyone can really tell an i7 from a pentium without running benchmarks. Even in most games, if you have a powerful enough GPU, there really isn't that much difference.

cute. i had this r7 260x on an athlon ii x3. worlds apart compared to when i put it on my i5. so i'm sorry but what you're saying is BS.

ps. we own a net cafe.

Really now? Well, I guess we might do different stuff on our computers. I have had no problem on mine and it runs everything outside of gaming well compared to pretty much every computer I used in the past 3 years. I can't tell the nanoseconds the computer is able to load a webpage or open ms word since the bottlenecks are all elsewhere. In gaming, I don't see the difference between 60 FPS and 80 FPS since my monitor is only 60hz anyways.

 

[jdwii]

I can't tell the difference between a FX8320 and a A6-3620 either without benchmarks. The difference between CPUs is so small now I doubt anyone can really tell an i7 from a pentium without running benchmarks. Even in most games, if you have a powerful enough GPU, there really isn't that much difference.

cute. i had this r7 260x on an athlon ii x3. worlds apart compared to when i put it on my i5. so i'm sorry but what you're saying is BS.

ps. we own a net cafe.

Really now? Well, I guess we might do different stuff on our computers. I have had no problem on mine and it runs everything outside of gaming well compared to pretty much every computer I used in the past 3 years. I can't tell the nanoseconds the computer is able to load a webpage or open ms word since the bottlenecks are all elsewhere. In gaming, I don't see the difference between 60 FPS and 80 FPS since my monitor is only 60hz anyways.

Its always the same argument from these guys so sad.

Really i have no problems reminds me of "its not hot in hear to me" bs

60 vs 80 fps like l
https://yourlogicalfallacyis.com/strawman

 

[jdwii]

"I don't see the difference between 60 FPS and 80 FPS since my monitor is only 60hz anyways."

NEWS everyone a Athlon II x3 can run farcry 4 and many other games at 60fps stable all the time. And if it can't its over the game being boring or crappy or some other excuse i come up with.

Sorry just sick of hearing such BS from people.

 

[de5_Roy]

guys, specify the situations/workloads where you do/not notice any difference. all your posts look like blanket statements.

for example: i notice no difference among single core, dual core and quad core when i am doing basic stuff in the operating system. during file operations like renaming/copying/moving, i'm usually memory and storage bound. the latest 4C pcs has the fastest ram and highest capacity, so that one works the fastest (cpu activity shows almost nothing).

during media playback, the systems are mostly gpu bound depending on rendering method/codec etc. the 1C pc chokes quickly in cpu-based rendering, then the 2C.

during de/compression the bottleneck shifts among the cpu, ram and storage. if you have two of those fast enough, the third one becomes the bottleneck. cpu bottlenecks are the most noticeable in this type of workload - the 4C one is the fastest and it shows every time. also depends on the software - 7zip scales with cores, some others don't scale higher than 2C.

multitasking with high priority av scan in the background - the 4C is noticeably the fastest.

video transcoding (handbrake): 4C hands down the fastest although all 4 cores are loaded and the whole pc chokes up.

 

[jdwii]

"I don't see the difference between 60 FPS and 80 FPS since my monitor is only 60hz anyways."

NEWS everyone a Athlon II x3 can run farcry 4 and many other games at 60fps stable all the time. And if it can't its over the game being boring or crappy or some other excuse i come up with.

Sorry just sick of hearing such BS from people.

are you drunk ? your above 2 posts lead me to believe so

It's just the same nonsense over and over, hear it all the time. If he said the basic user doesn't feel the difference i'd understand. I felt skips on web pages such as IGN on weaker recent(within 3 years) hardware. I will delate my last 2 post they were a bit vicious. Its quite obvious most would benefit from a SSD over anything else in their laptop/desktop.

Also i was a bit upset over the exaggeration as well and simply pointed out the fallacy in his argument
"I don't see the difference between 60 FPS and 80 FPS since my monitor is only 60hz anyways."

 

[juanrga]

I already provided before in this same thread a link to an article that showed that Jaguar and Piledriver have similar IPC, with difference of less than 3% on non-synthetic benchmarks.

the quote is Even if we omit the synthetic tests where Kabini does quite well, the core is still extremely competitive with AMD’s “big core” architecture, with an efficiency gap of less than 10%. not 3% ^
I will link it again

http://www.extremetech.com/computing/174980-its-time-for-amd-to-take-a-page-from-intel-and-dump-steamroller

I had to rebuild your post, with so many unmatched tags, improper quoting, and useless bold for separating your post from mine. Cannot you properly format? There is a preview button you can hit to check everything is correct before submitting posts.

The findings found by Extremetech article are summarized in the next graphic

Jaguar was 2.69% better than Piledriver on non-syntethic benchmarks, which matches my "less than 3% on non-synthetic benchmarks" and, evidently, also matches the "less than 10%".

Once again you are paying attention to ridiculous details (was 3%? 5%? 8%?) and ignoring the main message: Jaguar is still extremely competitive with AMD’s “big core” architecture. I will ignore further irrelevant posts about this.

I also would recommend you to stop attacking jdwii, just because he is right.

 

[noob2222]

I find it funny when people cant even read an article properly. Gamerk hit the nail on the head, ipc is a theoretical "maximum" number thats affected by multiple aspects during any given test program.

And to quote extremetech's ICP.

The simplest way to measure the efficiency of the two chips is to divide their respective benchmark scores in a given application by (CPU Frequency * Core Count). This normalizes both variables and gives us a measure of intrinsic core performance. 

 

[truegenius]

I find it funny when people cant even read an article properly. Gamerk hit the nail on the head, ipc is a theoretical "maximum" number thats affected by multiple aspects during any given test program.

And to quote extremetech's ICP.

thats why i sometimes use terms like per clock performance per core
though i think that we know ipc is theoretical but we use it as a ratio ( ex, SB is 40% better than PH2 ) based on benches which we prefer to find performance ( for ex, CB, 7zip, x264 for me because they scale with clock, support all core, push all core to max )

 

[noob2222]

Good point tg. One thing ill point out also, IPC is independent of benchmarks, how can processors swap places during certain benchmarks? Benchmarks arent testing IPC. In todays cpu designs, raw IPC is a useless term when talking actual performance.

 

[cdrkf]

Good point tg. One thing ill point out also, IPC is independent of benchmarks, how can processors swap places during certain benchmarks? Benchmarks arent testing IPC. In todays cpu designs, raw IPC is a useless term when talking actual performance.

I'm not sure that's entirely true. IPC varies on a particular design based on the specific task. For example, whilst it's generally accepted that Piledriver is similar or slightly inferior to Phenom II in IPC in general, there are some tasks that can make use of the new features on PD where it would have an IPC advantage. Still, a broad comparison can highlight the average IPC a particular processor has and that is comparable to other designs.