AMD CPU speculation... and expert conjecture

[cowboy44mag]

Well the FX9590 has been met with rather muted acceptance but I think there are a few impressive features;

To run 5ghz across all 8 cores at the voltage needed and the total power used is actually good considering a 4.5ghz 3930K uses more power. On the benching side a lot of benches show it competitive with i7's and Intel Extreme processors and it cannot claim core advantage as the Extremes have 12 vs 8 threads so that aspect is impressive, one other feature I liked was that at 5.2ghz the most that HS/f used in the HWC benches could handle at those clocks the 9590 was around 15-20% faster than its baseclock speed so perhaps its reached that point where the massive clocks overcame the IPC limitations on the architecture.

Pricing wise its hard to knock AMD choosing that mark albeit I think maybe $500 and $330 respectively would have been better publicity but again these are not average joe chips.

Considering the large changes in architecture coming with steamroller can you imagine what an enthusiast Steamroller FX chip will be? With the Pildedriver architecture the 9590 is able to achieve 5.0-5.2 GHz overclock on all 8 cores, if you can hit better clocks with the substantial improvements to IPC.... that would really be awesome. Even though an enthusiast level Steamroller FX chip would be very, very impressive I doubt I would shell out $800+ for one. For my needs the 9590 is already overkill

 

[hcl123]

That benchmark deals with lots of writes I believe hence top end ssd destroys hdds or low performing ssd. Is it so hard for people to understand, I am not on about intel or amd being better on these benchmarks just the ssd performance vs hdd or slow ssd.

Deals with a lot of "TRANSACTIONS" and that is from/for "system memory"... i believe if "coded" (again SOFTWARE) with "transactional memory" support, the HSW-E (if supported as planned) could revert the situation.

But don't count HTM (hardware transactional memory) "coding" be here so soon for current applications (edt).

UPDATE:
Matter of fact "vectorization" of that code has the effect, by compiler, of joining together a lot of memory accesses, those "queries" done by vectors could improve performance a lot, being FMA3 or 4 even better... i still believe its what happens, most probably FMA4 which Inyel doesn't support natively ( the same happened upon SSE4 of intel vs SSE2/3 of AMD in the time of K10 vs core2, was not the intel chip that was better, it was that particular SOFTWARE (with SSE4) that run much better on intel)...

 

[cowboy44mag]

4.5ghz 3930k is around 70% faster and fyi it probably takes similar power when i7 is oc'd. At stock the 3960x takes 62w less then the fx9590.

http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/62166-amd-fx-9590-review-piledriver-5ghz-17.html

Whats more impressive is the i7 4770k with its igpu stressed to the limit takes 101w and has at stock similar performance to the fx9590 that takes 248w without much headroom for overclocking.

You just don't get it do you? Seriously were you dropped on your head repeatedly as a child? No one on this forum cares about performance per watt. I would wager that everyone here, except for yourself, lives in an area where power is plentiful and cheap. The FX-9590 is an enthusiast processor period. You have to have a serious psu, motherboard combo to even run it. Yes it uses a lot more juice to run it, but enthusiast don't really give a hoot in hell about power consumption, all they care about is performance plain and simple.

As far as the i7 4770K goes the FX-9590 has better performance at stock than the Hasfail chip. The only place where the i7 4770K can beat the FX-9590 is single core performance as expected (and it doesn't get beaten by much). Multi-core performance the i7-4770K is beat plain and simple. Furthermore what make the Hasfail chip Hasfail is the fact that when it comes to overclocking the i7 4770K doesn't have much "headroom" either as they get too hot and don't overclock well at all. With a top end motherboard (Asus) and proper cooling I'm sure the FX-9590 can get higher clock than 5.2GHz.

More importantly you add on another 15-30% increase in performance for Steamroller FX the Hasfail chip is beaten across the board (except in power consumption).

 

[hcl123]

IMO, this FX 9590 is what SteamRoller performance will be, except with higher power consumption. Also, I found this...

http://www.rwlabs.com/newsitem.php?id=6829

I'm looking forward to buying a v8 in the future. I'll wait for the price to drop a little first. (If I wanted to burn my wallet, I'd buy an Intel CPU)

It depends on what "FAB process" the Steamroller FX will be made of. If it is "bulk" don't expect so higher clock potential. They better have a more clock prone design and or much higher IPC design (so to speak).

The FX9590 is for "enthusiasts" OCers... those already grab a "common" piece a OC it over 200W, some times over 300W in some OC exercises... this will be much better...

Its not uncommon at all going over 200W, the i7 3770 goes already over 200W at 4.8Ghz... above the FX9590, no matter the TDP rating
http://pctuning.tyden.cz/ilustrace3/obermaier/4770K/scaling_sandy.png

 

[Ranth]

cowboy44mag: You say hasfail when haswell is intels biggest innovation in last decade in all aspects of the chip.

AMD steamroller has to improve about 100% extra performance and perform 2x better performance per watt to get to haswell levels, thats why I mention power. Haswell 8 core 130w tdp cpu will be out next year and because amd is no competition expect enthisiasts to get it.

edit:fx9590 doesn't beat out the 4770k on multithreaded performance they are about similar.

Steamroller doesn't need to do that............ As long as AMD get above Intel performance AMD will have the upperhand, Intel can't use the perf/watt to anything if AMD beats them in performance.

 

[szatkus]

If you want multithread performance you can buy i7-3930K. FX-9590 is only a toy. Good to beat overclocking records or something.

 

[cowboy44mag]

cowboy44mag: You say hasfail when haswell is intels biggest innovation in last decade in all aspects of the chip.

AMD steamroller has to improve about 100% extra performance and perform 2x better performance per watt to get to haswell levels, thats why I mention power. Haswell 8 core 130w tdp cpu will be out next year and because amd is no competition expect enthisiasts to get it.

edit:fx9590 doesn't beat out the 4770k on multithreaded performance they are about similar.

Steamroller doesn't need to do that............ As long as AMD get above Intel performance AMD will have the upperhand, Intel can't use the perf/watt to anything if AMD beats them in performance.

In hafijur's make believe world Intel Haswell is twice as powerful in performance than Piledriver. He has quoted several times that Steamroller would have to be 100% better (than Piledriver) to compete with Haswell. If Steamroller were 100% better performance wise than Piledriver then it would be competing with Intel's $1000 plus processors and beating them.

Like I said dropped on his head many, many times. Haswell is as much a technological leap forward as the HMS Hood was. Much better power consumption than others of its class but everything else.... not so much. Even Intel loving sites quote Haswell as not being that much of an improvement over Ivy Bridge. Yet he still trolls and trolls about how great a "leap" forward it was. If that is Intel's idea of moving "forward" AMD will surpass them by no little margin by Excavator.

 

[szatkus]

@hafijur:
*4 modules

 

[gamerk316]

FX 9590 tested thoroughly:

http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/62166-amd-fx-9590-review-piledriver-5ghz-16.html

Who was talking about Company of Heroes 2 not doing well on AMD systems? They must have been crazy...

I simply pointed out CoH2 tended to favor Clock/IPC over more cores. Looks like the extra clock pushed the 9590 over the 4770k at stock. See:

http://www.techspot.com/review/689-company-of-heroes-2-performance/page4.html

and

Game scales well with increased clock for AMD, meaning the game is likely CPU bottlenecked.

Compare that to the 4770k:

Besides the somewhat obvious GPU bottleneck (@41 FPS, worth noting the 8350 maxed at 40, so that was probably bottlenecked too), note how low the clock can get without affecting performance, especially compared to the 8350. This highlights AMD's poor IPC relative to Intel: In a GPU bottlenecked situation, a 2.5GHz 4-core chip performs essentially the same as a 4.5GHz 8-core chip.

Which again, is EXACTLY why I've been saying IPC/Clock matters more then cores.

Moving to medium settings, which are more CPU bound:

More CPU bound, so dropping below 3.5GHz starts affecting performance, so the 4770k is clearly still GPU bound at medium settings somewhere between 3 and 3.5 GHz or so. Compared to the 8350 at 4.5GHz, which is still not putting out the same FPS as the 4770k, showing the FPS is being held back by the CPU.

Again, proving my point.

---------------

So heres my issue: Why I would expect raising the clock of PD would cause CPU performance to raise, eventually matching the 4770k, I would not expect the results being shown in your benchmark, due to the fact the 4770k is already bound by the GPU (as shown in the two 4770k graphs) at highest settings. Its also worth noting that hardwarecanucks has the GTX 670 grossly outperforming the GTX Titan from the techspot review. Techspot shows the Titan GPU bound at 41FPS at highest settings regardless of CPU, while hardwarecanucks shows a GTX 670 reaches at least 55 FPS at highest settings with a 9590, and 50 FPS with a 4770k.

So yeah, something is REALLY wrong with someones result set. I'm not taking sides, just pointing out the techspot and hardwarecanucks result sets are impossible to rectify. In short: Someone's test methodology is incorrect.

EDIT

Just sampled a few review sites. The FPS numbers for a Titan GPU tend to max out at about 40-45 FPS or so, which matches Techspot. Right now, I'm leaning toward hardwarecanucks results being inflated, which raises some serious questions about their testing methodology.

 

[cowboy44mag]

Intel is not using its power efficiency advantage till 8 core haswell. Intel don't need to improve drastically as they are already on a sensational level. Its true that amd need 8 core to compete against intel 4 cores.

AMD is making much larger gains per generation than Intel is. At the rate at which they are both "progressing" Intel's lead will be non-existent by Excavator. From what we can tell right now Steamroller will be competing at 4 cores so no it won't take 8 Steamroller cores to compete against Intel 4 cores. When Steamroller FX 8 core processors become available we will see just how good those Hasfail 8 core processors are (especially in performance per price). With Hasfails overclocking problems expect a delay with the 8 core processors.

Cling to your all important power consumption numbers if you need to, but as explained to you in deep detail power consumption in a desktop computer isn't as important as performance.

 

[etayorius]

lol, intel are so far ahead. 4 years ago it was close now its the biggest gap in the history between intel and amd. Piledriver and bulldozer are massive flops. AMD would have been better off downshrinking the older phenom 2 cpus to 32nm. I find it staggering how from 45nm core 2 series penryn cpus on laptops to ivy bridge 22nm cpus they have improved so much. I think sandy bridge took amd by suprise and since then intel haven't looked back and amd engineers are scratching there heads to wonder how intel are doing this.

Intel nehalem was a big flop but at 3ghz the i7 competes against the fx8350 and this is on 45nm. At 4ghz it should match the fx8350 as well.
http://www.anandtech.com/bench/Product/100?vs=697

Sandy bridge gigantically improved effiency and ipc and added igpu for i7's and took half power consumption to destroy nehalem cpus. Thats when amd were blown away.

Can´t you go into a Intel thread and stay there? i been silently following this thread since page 25 and i am getting sick of your constant Intel trolling... Are there any mods out there willing to ban or at least warn this guy to stay away from this thread? he is going offtopic, trolling and hijacking this thread for a while... it is getting annoying reading his constant Pro Intel rambling... for Christ sake just get the F out of this thread already!

This guy is ruinning this thread... seriously... if you love Intel so much just go and buy a damn Intel CPU, this thread was supposed to be about AMD Steamroller and not about how revolutionary your damn intel haswell pimp is.

 

[The Q6660 Inside]

lol, intel are so far ahead. 4 years ago it was close now its the biggest gap in the history between intel and amd. Piledriver and bulldozer are massive flops. AMD would have been better off downshrinking the older phenom 2 cpus to 32nm. I find it staggering how from 45nm core 2 series penryn cpus on laptops to ivy bridge 22nm cpus they have improved so much. I think sandy bridge took amd by suprise and since then intel haven't looked back and amd engineers are scratching there heads to wonder how intel are doing this.

Intel nehalem was a big flop but at 3ghz the i7 competes against the fx8350 and this is on 45nm. At 4ghz it should match the fx8350 as well.
http://www.anandtech.com/bench/Product/100?vs=697

Sandy bridge gigantically improved effiency and ipc and added igpu for i7's and took half power consumption to destroy nehalem cpus. Thats when amd were blown away.

Did you even care to realize that all improvements after Core are minor? Sandy Bridge was a minor step up, about 10%. An overclocked Q9650 on a 790i motherboard is still quite close to a 3770K rig with the same GPU. Nehalem was a small increase as well and one of Intel's best lines of processors, you know the last ones that had the following.

Did not kill overclocking on anything less than a top end i5.
Actually had good value for money.

Piledriver was not a flop whatsoever. A a ULV P7350 against an i5, seems fair.....................................

 

[cowboy44mag]

lol, intel are so far ahead. 4 years ago it was close now its the biggest gap in the history between intel and amd. Piledriver and bulldozer are massive flops. AMD would have been better off downshrinking the older phenom 2 cpus to 32nm. I find it staggering how from 45nm core 2 series penryn cpus on laptops to ivy bridge 22nm cpus they have improved so much. I think sandy bridge took amd by suprise and since then intel haven't looked back and amd engineers are scratching there heads to wonder how intel are doing this.

Intel nehalem was a big flop but at 3ghz the i7 competes against the fx8350 and this is on 45nm. At 4ghz it should match the fx8350 as well.
http://www.anandtech.com/bench/Product/100?vs=697

Sandy bridge gigantically improved effiency and ipc and added igpu for i7's and took half power consumption to destroy nehalem cpus. Thats when amd were blown away.

Intel is ahead, no question there, but it isn't by the huge margin that you or any other overcompensating Intel fanboy wants to believe. The best benchmarks of AMDs last generation (Piledriver) and Intels current generation (Haswell) are separated be mere seconds and a few FPS. That is not "destroying" the competition.

Furthermore if Intel has any more "leaps forward" like Haswell AMD will have no trouble advancing right by them. Anytime you have a product that performs worse in any way than the product that came before it then it is a failure. Bulldozer was a failure because it wasn't as good as Phenom II, Haswell is a failure because it isn't as good as Ivy Bridge. Even Intel loving sites have said the same in their reviews. You try to overclock Haswell and it gets too hot and doesn't scale as well as Ivy Bridge--> failure.

Intel is prime to be caught off guard by a Steamroller processor that will make their "lead" even less significant than it already is. Ivy Bridge-->Haswell isn't anything compared to the leap forward of Piledriver-->Steamroller. If AMD can replicate that again from Steamroller-->Excavator Intel won't have a lead anymore.

 

[szatkus]

It's failure because people were waiting for Tock like Sandy Bridge. It's good step forward, but not revolution.

Looks like Steamroller will make AMD more competetive, but still they have long way to go.

 

[cowboy44mag]

I just corrected cowboy44mag nonsense. AMD you could say have nearly standed still for the last 4 years except a die shrink and he makes it out like they are improving so much when its intel who are.

edit: I am hoping amd improve like 100% performance per watt then intel have some competition and finally stop releasing low end parts. Its a bit like nvidia releasing a gt 640 as a gtx680 if the competition was so low.

My nonsense and you say AMD hasn't made any improvements in 4 years and has "stood still"???? Are you serious?? You are saying that a Phenom II is exqual to a FX-8350 and the Phenom II will benchmark the same as an FX-8350? Man I don't think I could drink enough to visit the reality you live in

Seriously you got to step out of the hash bar and get some fresh air, save the very few undamaged brain cells that remain. Piledriver was anything but a flop and closed the margin between AMD and Intel considerably. By your own benchmark scores you have posted from Intel friendly sites the margin is quite small in the real world. On a chart a few seconds can look quite large, but in the real world a few seconds is nothing. FPS is the same as our imperfect human eyes can't see the difference. If Steamroller can deliver 20-30% better performance than Piledriver then Steamroller FX will be neck and neck with Intel. It will come down to what compiler the benchmark used and which benchmarks are used (single vs multi) as to which is better.

 

[sarinaide]

lol, intel are so far ahead. 4 years ago it was close now its the biggest gap in the history between intel and amd. Piledriver and bulldozer are massive flops. AMD would have been better off downshrinking the older phenom 2 cpus to 32nm. I find it staggering how from 45nm core 2 series penryn cpus on laptops to ivy bridge 22nm cpus they have improved so much. I think sandy bridge took amd by suprise and since then intel haven't looked back and amd engineers are scratching there heads to wonder how intel are doing this.

Intel nehalem was a big flop but at 3ghz the i7 competes against the fx8350 and this is on 45nm. At 4ghz it should match the fx8350 as well.
http://www.anandtech.com/bench/Product/100?vs=697

Sandy bridge gigantically improved effiency and ipc and added igpu for i7's and took half power consumption to destroy nehalem cpus. Thats when amd were blown away.

You are like a broken record and the repetition is starting to hurt my brain and I am not liking it anymore.

 

[cowboy44mag]

It's failure because people were waiting for Tock like Sandy Bridge. It's good step forward, but not revolution.

Looks like Steamroller will make AMD more competetive, but still they have long way to go.

Its not a step forward if it can't overclock as well as Ivy Bridge. PERIOD. If overclocked Ivy Bridge = to what Haswell can overclock stable then its not a step forward. If sites as bias as anandtech recommend not upgrading to Haswell if you are on Ivy Bridge then it is not a step forward.

Steamroller will make AMD more competitive, but they don't have as far to go as you Intel fanboys like to think. Like I said a couple FPS and a few seconds isn't a huge, gigantic, overwhelming lead. Steamroller FX should make Intel's lead inconsequential when taking performance/price into consideration.

 

[cowboy44mag]

lol, intel are so far ahead. 4 years ago it was close now its the biggest gap in the history between intel and amd. Piledriver and bulldozer are massive flops. AMD would have been better off downshrinking the older phenom 2 cpus to 32nm. I find it staggering how from 45nm core 2 series penryn cpus on laptops to ivy bridge 22nm cpus they have improved so much. I think sandy bridge took amd by suprise and since then intel haven't looked back and amd engineers are scratching there heads to wonder how intel are doing this.

Intel nehalem was a big flop but at 3ghz the i7 competes against the fx8350 and this is on 45nm. At 4ghz it should match the fx8350 as well.
http://www.anandtech.com/bench/Product/100?vs=697

Sandy bridge gigantically improved effiency and ipc and added igpu for i7's and took half power consumption to destroy nehalem cpus. Thats when amd were blown away.

You are like a broken record and the repetition is starting to hurt my brain and I am not liking it anymore.

+1 power consumption is not the end all be all of performance

 

[cowboy44mag]

I think we have strayed into Intel land and Hawell long enough. This thread is supposed to be about Steamroller and comparing Piledriver to Steamroller. The actual benchmarks when Steamroller FX releases will show us without a doubt how it compares to Intel.

I am really hoping to get a little news on Steamroller FX soon. I can understand AMD wanting to keep a lid on things for now, but it gets harder and harder to wait and not upgrade to FX-8350!! The FX-8350 is a very good processor, but this close to Steamroller I want to see what Steamroller FX and a Sabertooth 990FX can do!!

 

[lilcinw]

And the 3930x costs 5x what the 8350 does...

Somebody needs to track down BaronMatrix. I don't know if there is enough popcorn in the States to sustain me for the flame war these two could have.

 

[cowboy44mag]

How would an 8 core phenom 2 perform on 32nm, answer similar to the fx8350. Theres a few differences yes architecturally but nothing major. The i7 3960x already beats piledriver around 50% so improving 20-30% doesn't make it neck and neck. Since sandy bridge intel improved like 15-20% ipc wise to haswell. So at best amd might get to sandy bridge levels ipc wise but not efficiency.

OMG seriously how many illegal narcotics do you shove up your ... a day? Your comparing i7 3960 an $1100 processor to the FX-8350 a processor that can be bought for $150?? Comparing the i7 4770K is already a stretch at over $300. If Steamroller has 20-30% increase over Piledriver it will be neck and neck with i7 4770K which is still almost double the cost.

Now seriously can we get back to AMD STEAMROLLER and comparing Piledriver to Steamroller? Sorry to inform ya pal, but this aint a everybody loves to suck Intel's .... page.

 

[cowboy44mag]

And the 3930x costs 5x what the 8350 does...

Somebody needs to track down BaronMatrix. I don't know if there is enough popcorn in the States to sustain me for the flame war these two could have.

Your probably right, but I'm done with talking about Intel. You like Intel, by all means buy Intel. I am much more interested in getting some actual Steamroller news (hopefully Steamroller FX ) and comparing Steamroller to Piledriver, Bulldozer, Phenom II, ect....

 

[szatkus]

It's failure because people were waiting for Tock like Sandy Bridge. It's good step forward, but not revolution.

Looks like Steamroller will make AMD more competetive, but still they have long way to go.

Its not a step forward if it can't overclock as well as Ivy Bridge. PERIOD. If overclocked Ivy Bridge = to what Haswell can overclock stable then its not a step forward. If sites as bias as anandtech recommend not upgrading to Haswell if you are on Ivy Bridge then it is not a step forward.

Most people don't overclock. For average user Haswell is better option than IB.

Steamroller will make AMD more competitive, but they don't have as far to go as you Intel fanboys like to think. Like I said a couple FPS and a few seconds isn't a huge, gigantic, overwhelming lead. Steamroller FX should make Intel's lead inconsequential when taking performance/price into consideration.

The problem is still single thread. The gap is higher than 30%. There's still significiant amount of new games which use only two threads. That's why discussions like "Phenom X4 vs. i3" are usually one big flame, because in some titles first is better and second is better in others.
Also gap is even larger when you compare top CPUs in proffessional tasks. AMD still doesn't have anything competetive to Intel hex-cores.

 

[szatkus]

According to this its not that impressive.
http://www.xbitlabs.com/news/cpu/display/20130702235056_AMD_Kaveri_Engineering_Sample_Found_in_Cosmology_Research_Database.html

There's something wrong with that ES or tests. It's even lower than similar Piledrivers.

 

[mayankleoboy1]

2.) OpenCL Programs:

-Adobe Photoshop
-Adobe Premiere Pro
-AIDA64
-Battlefield 3
-Final Cut Pro X
-Corel Video Studio
-Darktable
-Handbrake
-HMPP Workbench
-Image Magick
-Indigo Renderer
-LuxMark software
-MAGIX Movie Edit
-Mathematica 8
-Mental RAY
-Movie Studio HD 11
-MulticoreWare
-Linux OS
-Mac OS
-Windows 8.1
-H.264/AVC Pro Encoder
-Panorama Maker Pro 6
-Par4All
-PhotoMonkee
-POV-RAY
-PowerDirector
-ReconstructMe
-Ruby
-Stage Presence
-THEIA RT
-TotalMedia Theatre 5
-Triton Ocean SDK
-Vantage Transcode
-Vegas Pro 11
-Viewdle Face Recognition
-VLC Media Player
-V-RAY
-vReveal
-WinZip
-7zip
-WinRAR
-Wireless Security Auditor

Additionally CLCC is the Open Source OpenCL compiler for x86

winrar and 7zip dont support OpenCL. Additionally, 7zip has no plans to use OpenCL in the foreseeable future. Reason : OpenCL is not suitable for the algorithm of 7zip (as said by the developer). This may sound like heresy to HSA/AMD fanbois, but this is really how real world coding is done.
I am not sure if VLC uses any opencl.
Handbrake has experimental, pre alpha support for OpenCL for only a few filters. And IIRC, opencl reduced performance in most cases.