AMD CPU speculation... and expert conjecture

[de5_Roy]

since amd uses snooping based cache coherency protocol, is it also partly responsible for up to 8 cores per cpu and 16 cores per socket(mcm cpus) practical/economical limit? i thought that adding more cores would complicate cache coherency. i've suspected that it may also be the reason amd's console socs, octocore arm cpus won't go higher than 8 cores per cpu. tbh i haven't come across in-depth details on how arm cpus work.

 

[gamerk316]

MIPS is back, boys and girls:

http://www.tomshardware.com/reviews/imagination-technologies-mips-powervr,3890.html

In the comparison with competing high-end ARM CPUs above (which is again provided by Imagination, so do with the information what you will), we see its MIPS P5600 “Warrior” CPU manage to beat the ARM competition (likely Cortex-A15) in performance/MHz, performance/mW, and also performance/area. Keeping in mind that the other factors were most likely normalized in these comparisons, it’s still impressive that MIPS-based CPUs can be this competitive with ARM in the mobile market already.

And according to some here, this means MIPS > ARM > X86.

 

[cdrkf]

since amd uses snooping based cache coherency protocol, is it also partly responsible for up to 8 cores per cpu and 16 cores per socket(mcm cpus) practical/economical limit? i thought that adding more cores would complicate cache coherency. i've suspected that it may also be the reason amd's console socs, octocore arm cpus won't go higher than 8 cores per cpu. tbh i haven't come across in-depth details on how arm cpus work.

Uh AMD's console socs are x86 based on their small 'jaguar' cores (same as in Beema and Mullins apu's).

 

[cdrkf]

MIPS is back, boys and girls:

http://www.tomshardware.com/reviews/imagination-technologies-mips-powervr,3890.html

In the comparison with competing high-end ARM CPUs above (which is again provided by Imagination, so do with the information what you will), we see its MIPS P5600 “Warrior” CPU manage to beat the ARM competition (likely Cortex-A15) in performance/MHz, performance/mW, and also performance/area. Keeping in mind that the other factors were most likely normalized in these comparisons, it’s still impressive that MIPS-based CPUs can be this competitive with ARM in the mobile market already.

And according to some here, this means MIPS > ARM > X86.

I'm sure I read that the Chinese government has been funding the development of home grown high performance CPUs that are based around MIPS? What's the betting that if MIPS does start to gain traction, the new 'ambidextrous' AMD will probably buy a licence?

 

[de5_Roy]

Uh AMD's console socs are x86 based on their small 'jaguar' cores (same as in Beema and Mullins apu's).

yeah, i said that knowing that.

 

[cdrkf]

Uh AMD's console socs are x86 based on their small 'jaguar' cores (same as in Beema and Mullins apu's).

yeah, i said that knowing that.

Ah sorry I think I read your comment wrong

 

[gamerk316]

NVIDIA IS DOOMED! DOOMED I TELL YOU!

http://www.fudzilla.com/home/item/35457-nvidia-posts-strong-results-beats-estimates

 

[blackkstar]

That's why A57 is more less two times bigger than A15?

Apple A7 is only 100mm^2. That's still not remotely massive and it leaves a lot of room for cramming in cores. But I wasn't even talking about ARM, I was talking about x86 cores and how Intel and AMD are both doing the same thing.

Except, again, as was discovered in the late 70s, you can't use more then a handful of cores to work on general computing tasks. You simply won't see the performance gain for the vast majority of tasks out there. So adding cores is also a dead end.

Hence why I suspect we're going to hit "peak computing" in 3-4 years, where we essentially run out of ways to improve performance through HW. At that point, we're all going to be waiting form Quantum computing to become viable.

Things have changed a lot since the 70s. People are going to have to find a solution. We're seeing things that can share memory much more efficiently and we're coming up with workloads that do scale well to lots of cores. Rendering and compiling all scale well and I don't even think we had ray tracing back then. We also have things like Apache which can serve thousands of pages a second and every page request runs in its own thread. All things that didn't exist in the 70s when the claim that more cores doesn't help.

There will be problems where more cores doesn't help though. Please do not misunderstand. But our single thread is so good right now that it's acceptable. Also, do not forget that your GPU is just hundreds or thousands of SIMD cores and those have no problem scaling to lots of cores. It can be done. It won't be easy. But a lot of things will benefit from more cores.

Lol no more or less the same is like -5%-+5%. I7 920 vs I7 3770K which haswell is even faster in dolphin emulator over IPC improvements, I7 920 is 15% less efficient in IPC compared to Ivy and probably closer to 20% slower per clock compared to Haswell. That is from 2008-2014 only 6 years would you like me to pull out some horrible Pentium 4 IPC benchmarks(from 04) i hope not. Amd on the other hand has been around the same.
http://alienbabeltech.com/main/platform-upgrade-core-i7-920-vs-i7-3770-at-4-2ghz-featuring-ecs-golden-series-motherboard-and-kingston/7/
http://www.anandtech.com/show/2658/16
I7 965 was around 10%-15%(some cases 30%) faster ipc IPC compared to the Intel core 2 extreme Qx 93770 at the same clock rates

Then we can see the improvements on the quad core extreme vs Pentium dual core from 05 sure its around 10-15% as well. Overall from 04-14 we saw a solid 40-50% boost in IPC compared to a pentium 4 extreme and a I7 4770K.
We can also see sandy to haswell is around 10-15% faster in IPC

http://www.hardocp.com/article/2013/06/01/intel_haswell_i74770k_ipc_overclocking_review/1#.U-O-2fldXJY

I'm not to sure how much longer these core improvements will continue.

Compare the instruction sets of Haswell to predecessors and you will understand that Haswell's performance in Dolphin does not come from architecture improvements, but from new instructions. Here's a little shocker for you. When I custom compile Blender in Gentoo for both my i7 920 AND by FX 8350, they see massive improvements. i7 920 is 50% faster and FX 8350 is over 80% faster. Dolphin's performance is not some sort of Haswell miracle, it is the Dolphin Team doing a very good job at compiler optimizations, specially because they have something that won't even run on older hardware well so they have no reason to even care about being compatible with Pentium 3 level instruction sets like 99% of Windows applications.

Also, I was arguing that after Nehalem things slowed down massively. P4 was a blunder and an exception but the differences between the AMD and Intel cores of the time were generally huge improvements.

That was fascinating.

But, really, do you believe that? I don't see any reason single-thread performance is at a standstill, because it has steadily gone upward since the beginning of CPUs.

Haswell did improve single thread performance, and as the size of the process technology drops lower and lower (14nm, 10 nm, etc.) we will only be able to fit more processing components and transistors, really.

I don't think it will stop, and even as the industry leans toward more and more cores, the single thread performance will only improve as the techology improves.

Yeah, I do.
http://www.tomshardware.com/reviews/core-i7-4770k-haswell-review,3521.html

Look at what changed with Haswell:

out of order window increased 14%
an extra ALU
an extra store
larger L2 TLB

And you don't see a gain over 10% unless you're using new instructions anywhere. Intel has already plucked the low hanging fruit from branch predictors and stuff.

Take a look at the [H] review since they locked frequency and you don't have to be concerned with CPU turbos mucking up scores and not having any clue what the chips frequency actually is during a benchmark.

4% faster in Hyper Pi, 5% faster in wPrime, 12% cinebench (which is multi-threaded), 21% in POV-Ray (multi-threaded), 9% in LAME, hardly anything in WinRar. The only time Haswell shows big gains is when it's using more than one thread, which shows me more that Intel increased performance of how cores work together more than they did with increasing actual single thread performance. Those are huge gains and Haswell was supposed to be a big "tock" architectural change and we get a whole lot of tiny increases with a few oddball gains.

Compare to a review of FX-57 or E6600
http://www.anandtech.com/show/1722/4
http://www.trustedreviews.com/Intel-Core-2-Duo-Conroe-E6400-E6600-E6700-X6800_PC-Component_review_photoshop-virtualdub-results_Page-5#tr-review-summary

X6800 is 53% faster than P4EE 955 in Photoshop. Those are the kinds of changes we will never see again that I am talking about.

 

[szatkus]

That's why A57 is more less two times bigger than A15?

Apple A7 is only 100mm^2. That's still not remotely massive and it leaves a lot of room for cramming in cores. But I wasn't even talking about ARM, I was talking about x86 cores and how Intel and AMD are both doing the same thing.

Actually you were talking about ARM:

People spend so much time running around the internet going "LOL MOAR COARS HAHAHAHA" when that's all we have left. ARM is doing it too.

The rest of quotes aren't mine

 

[gamerk316]

Things have changed a lot since the 70s. People are going to have to find a solution. We're seeing things that can share memory much more efficiently and we're coming up with workloads that do scale well to lots of cores. Rendering and compiling all scale well and I don't even think we had ray tracing back then. We also have things like Apache which can serve thousands of pages a second and every page request runs in its own thread. All things that didn't exist in the 70s when the claim that more cores doesn't help.

And what did we do with most of those tasks? Offload them to GPUs or Servers.

At the end of the day, the majority of the work you are going to do within an application is SERIAL PROCESSING. Sure, parts of an application you can thread really, really well; we already do it for Rendering and Physics. But AI? Not so much. Same deal with audio. Parallel I/O? Forget that. And then there's the main game engine, which you really can't thread well at all.

And even then, you have the host OS to consider. You have significant overhead there, which gets exponentially worse as you throw in more cores. Maintaining coherency becomes a major problem. And so on.

Point is, the stuff the does scale already gets offloaded to specialized processors. The rest of the stuff is serial, and there's nothing you can do to make that benefit from more cores.

 

[jdwii]

NVIDIA IS DOOMED! DOOMED I TELL YOU!

http://www.fudzilla.com/home/item/35457-nvidia-posts-strong-results-beats-estimates

You just wait until Intel releases their iGPU its going to beat 4 titans according to some users.

I'm one of the few i feel like who wants Nvidia and Amd and Intel to remain competitive, Nvidia makes some amazing things and i for one am happy with them as i was with Amd.

 

[gamerk316]

Optical computers on the horizon:

http://www.extremetech.com/extreme/187746-by-2020-you-could-have-an-exascale-speed-of-light-optical-computer-on-your-desk

 

[michaelwebb]

amd all the way i love mine.
Graphics Card
3GB Sapphire Radeon R9 280X TOXIC, 28nm, 6400MHz GDDR5, GPU 1100MHz, Boost 1150MHz, 2048 Streams, DVI/HDMI/mDPort

Processor
AD785KXBJABOX - AMD A10 7850K Black Edition, S FM2+, Kaveri Core, Quad Core, 4.0GHz, AMD Radeon R7, 95W, Retail

Power Supply
850W Corsair RM Series RM Series, Full Modular, 80 PLUS Gold, 1x135mm Fan, ATX v2.4, PSU

Motherboard
A88X-G45 GAMING Assassin?s Creed Liberation HD MSI A88X-G45 GAMING Assassin?s Creed Lib HD, AMD A88X, FM2+, DDR3, SATA III 6Gb/s, RAID, PCIe 3.0, DSub/DVI/HDMI, ATX

Memory/R.A.M
8GB (2x4GB) Corsair DDR3 Vengeance Pro Series Red, PC3-17066 (2133), Non-ECC Unbuffered, CAS 9-11-11-31, XMP, 1.65V

Case
Cooler Master CM Storm Trooper XL-ATX Case with Window, USB 3.0, Black

Hard Drive
1TB WD WD10EFRX RED 24x7, SATA 6Gb/sec, 64MB Cache, IntelliPower, 8ms NAS/Enterprise

Operating System
Microsoft Windows 8.1 64Bit DVD English International OEM -

DVD Writer
LiteOn IHAS124-14 24x DVD±R, 8x DVD±DL, DVD+RW x8/-RW x6, DVD-RAM x12, SATA, Black, OEM

 

[jdwii]

That A10 is bottlenecking that 280X without a doubt in many...many games. At least its OC a bit

 

[michaelwebb]

sorry for the silly reply explain newby at all this buddy sorry..just brought im happy with it anyway think its great.

That A10 is bottlenecking that 280X without a doubt in many...many games. At least its OC a bit

 

[jdwii]

sorry for the silly reply explain newby at all this buddy sorry..just brought im happy with it anyway think its great.

That A10 is bottlenecking that 280X without a doubt in many...many games. At least its OC a bit

that is all what matters i guess is if you are happy with it but 40% of your APU is just setting their doing nothing but wasting space. More cores or L3 cache or bigger cores would of helped you out more.

 

[szatkus]

That A10 is bottlenecking that 280X without a doubt in many...many games. At least its OC a bit

Oh no, not again...

 

[esrever]

sorry for the silly reply explain newby at all this buddy sorry..just brought im happy with it anyway think its great.

That A10 is bottlenecking that 280X without a doubt in many...many games. At least its OC a bit

that is all what matters i guess is if you are happy with it but 40% of your APU is just setting their doing nothing but wasting space. More cores or L3 cache or bigger cores would of helped you out more.

Space doesn't matter to a consumer. The chip is the same size no matter what you buy.

 

[michaelwebb]

sorry for the silly reply explain newby at all this buddy sorry..just brought im happy with it anyway think its great.

That A10 is bottlenecking that 280X without a doubt in many...many games. At least its OC a bit

that is all what matters i guess is if you are happy with it but 40% of your APU is just setting their doing nothing but wasting space. More cores or L3 cache or bigger cores would of helped you out more.

Space doesn't matter to a consumer. The chip is the same size no matter what you buy.

ok yea new to me to be honest is there anyway to get more out of that apu and other stuff. new o this so bare with me. lol thanks for the reply

 

[de5_Roy]

if the owner is satisfied with his pc's performance, there's no need to push it for the time being.

 

[michaelwebb]

if the owner is satisfied with his pc's performance, there's no need to push it for the time being.

no your very true im running 4.3 at the min on overclock and the bored what iv got i can press a button and it will over clock it more but does everything for you but makes sure its safe thats the good thing with these new boreds from msi if that is what you mean...very new to all this would more ram help in any way at all.. cheers

 

[jdwii]

if the owner is satisfied with his pc's performance, there's no need to push it for the time being.

no your very true im running 4.3 at the min on overclock and the bored what iv got i can press a button and it will over clock it more but does everything for you but makes sure its safe thats the good thing with these new boreds from msi if that is what you mean...very new to all this would more ram help in any way at all.. cheers

Overall its pretty solid and should last plus i think FM2+ might be further compatible with newer CPU's

Space doesn't matter to a consumer. The chip is the same size no matter what you buy.
True but the point of a APU is its graphics so its kinda of important if the igpu wont be used you would be better off getting a 8320fx(which is around the same price).

 

[jdwii]

That A10 is bottlenecking that 280X without a doubt in many...many games. At least its OC a bit

Oh no, not again...

yes again and again and again until i don't see people picking 100$ quad core Amd CPU's and pairing it with 300$ Video cards. BIG bottleneck and the new ones are just as fast as a A10 6800K in most cases in CPU tasks. So sorry you can defend your choices of thinking a cheap CPU and a high end GPU is a good choice but the facts are not on your side benchmarks do not lie the people conducting them do or sadly the ones looking at them.

 

[esrever]

Depends on the game but if you actually look at benchmarks. Most games aren't even CPU dependent these days. Any quad or dual with hyper threading seems to play most of games without bottleneck.

 

[michaelwebb]

if the owner is satisfied with his pc's performance, there's no need to push it for the time being.

no your very true im running 4.3 at the min on overclock and the bored what iv got i can press a button and it will over clock it more but does everything for you but makes sure its safe thats the good thing with these new boreds from msi if that is what you mean...very new to all this would more ram help in any way at all.. cheers

Overall its pretty solid and should last plus i think FM2+ might be further compatible with newer CPU's

Space doesn't matter to a consumer. The chip is the same size no matter what you buy.
True but the point of a APU is its graphics so its kinda of important if the igpu wont be used you would be better off getting a 8320fx(which is around the same price).

right i see now so if i got the 8320fx chip instead it will run alot better being 8 core might just look into that instead of buying ather one of the graphics cards iv got.