AMD CPU speculation... and expert conjecture

[hcl123]

The contrary, the cluster approach, by permitting to divide the workload in "separated" type of instructions and data, is the "feature" (edt) that more helps "clock cycle" (since long, academia simulations shown this)... on AMD module only the FP code "type" is off the CPU code paths.

But its already clever, prone to achieve higher clocks AS DEMONSTRATED... and prone to future enhancements... and no its NOT BD/PD that is P4, matter of fact BD has less pipeline stages than SB/HW, 15 compared to 17 for SB/Hasfail. Only they toke the possibility of separated code paths to streamline the CPU cores further (they went a little too far i think).

By overclocking I was referring to the TDP limits and turbo modes. With a Jaguar/Athlon each core can be clocked up or down without any dependencies. Like you can have Core 0 at 4Ghz and Core 1 at 1Ghz. For many workloads you don't need every core running at the base clock. Ideally you could get 1 or 2 cores to clock even higher than usual, while preserving TDP with the remaining cores at much lower clock speeds.

If you look at Intel mobile processors some of them can turbo 1Ghz over their base clock. I'd like to see that grow even larger to like 2Ghz. A lot of what we do on computers tends to be bursty in nature. Temporary super-Turbos would make PCs feel even more responsive.

What you describe as *nothing* to do with the "module" topology.

It has to do with the power management schemes, that with AMD more centered around "server" for this chips, is half/half, that is, is only clock gated and managed in halfs (since Lisbon/Thuban), not by single *Modules*.

About TDP limits, SOI has it much higher, IBM chips are around 250W for those big servers and must function 24/7 (imagine all days all along always close to max load ?) for many years.

IBM doesn't need Turbo, it doesn't make sense... AMD is half way to both ends.

If you want *REAL* low power (for the "bursty" usage model)... simply go ARM (there you wouldn't need too sophisticated power managements... and the software tends to be much less bloated and faster to) .period.

Sophisticated Power Management is only an attempt to "ease" the power constraints of x86.

 

[hcl123]

http://www.tomshardware.com/news/AMD-Kaveri-APU-Gaming,22947.html

juangra 15-20% better ipc means they are about the equivalent performance of a 2.9ghz core i series cpu is a 4ghz kaveri cpu. So at 3.3ghz the i5 3350p will be equivalent to around something like a 4.4ghz kaveri cpu. I don't know why amd fanboys overhype amd's own targets which is less then what you predicted.

I find it hilarious that an intel [strike]fanboy[/strike] who has been posting nonsensical performance and power figures for weeks is now trying to give lessons to others. LOL

The difference in performance between the A10-6800k and the i5-3350p is of about a 10-15%. Therefore the CPU in Kaveri will be slightly faster than the i5 IB.

http://www.kitguru.net/components/cpu/henry-butt/amd-a10-6800k-richland-apu-review/10/
lol, are you people mad. Its crazy 10-15% in your world in real world is around 80%. The a10-6800k competitor for cpu performance is i3, stop kidding yourself.

edit look at i3 3220 3dmark11 physics score:
http://www.hardwaresecrets.com/article/A10-5800K-vs-Core-i3-3220-CPU-Review/1646/17

I would say the i3 3220 and a10-6800k cpu are on par.
one more thing:
http://www.pcper.com/reviews/Processors/Intel-Core-i5-3350P-Ivy-Bridge-Processor-Review-No-Integrated-Graphics/Overclocki
'To reiterate - the Core i5-3350P uses 110 watts LESS POWER than the FX-8150 while outperforming it in many cases. Crap.' Good luck a10 apu competing againt the i5 lol.

Are you for real ? ... sometimes i wonder if you are a payed shill (or something) or if you lost the capability to "think".

The links you point SHOW NOTHING of the conclusions you take... give time and money, and i can think of ways to code those benchmarks faster on Deneb than on Hasfail lol... on BD/PD is then so easy that even gives shills, like with Cinebent(sse on non-intel, avx on intel), just use XOP/FMA4 as much as possible lol

UPDATE:
[ but then it will be "fraud" wouldn't you say ? ... coding with XOP/FMA4 when intel doesn't support it natively, and so it would by the SOFTWARE side painful slow in their chips... and more "fraud" because nobody out there is really using XOP/FMA4 for their "current" software that ppl use every day, and so a benchmark coded with with XOP/FMA4 wouldn't be representative at all of any software ppl use, right ?? ... but isn't that already the case with those benchmarks, most specially the "synthetics" ? ... no representative at all of the software that ppl use ?? ]

 

[szatkus]

Stop trolling people. Let's make it clear:

Fact 1. i3-3220 is 20-28% faster in SINGLE THREAD.
Source: http://www.tomshardware.com/reviews/a10-6700-a10-6800k-richland-review,3528-3.html
Fact 2. Kaveri will struggle with Haswell parts: +10%
Fact 3. i5 has turbo: +15%
Fact 4. i5 has 4 true cores, Kaveri only 2 modules.

Conclusion: Kaveri can fight with i3, but has no chance with i5.

 

[juanrga]

http://www.tomshardware.com/news/AMD-Kaveri-APU-Gaming,22947.html

juangra 15-20% better ipc means they are about the equivalent performance of a 2.9ghz core i series cpu is a 4ghz kaveri cpu. So at 3.3ghz the i5 3350p will be equivalent to around something like a 4.4ghz kaveri cpu. I don't know why amd fanboys overhype amd's own targets which is less then what you predicted.

I find it hilarious that an intel [strike]fanboy[/strike] who has been posting nonsensical performance and power figures for weeks is now trying to give lessons to others. LOL

The difference in performance between the A10-6800k and the i5-3350p is of about a 10-15%. Therefore the CPU in Kaveri will be slightly faster than the i5 IB.

I think it would be more 2550K-ish performance wise.

I agree with you. The i5-2550k ~ i5-3350p. I am considering a difference in performance of about a 5%, although this intel enthusiast claims a 6%

http://www.tomshardware.co.uk/forum/349920-28-3350p-2500k

 

[The Q6660 Inside]

http://www.tomshardware.com/news/AMD-Kaveri-APU-Gaming,22947.html

juangra 15-20% better ipc means they are about the equivalent performance of a 2.9ghz core i series cpu is a 4ghz kaveri cpu. So at 3.3ghz the i5 3350p will be equivalent to around something like a 4.4ghz kaveri cpu. I don't know why amd fanboys overhype amd's own targets which is less then what you predicted.

I find it hilarious that an intel [strike]fanboy[/strike] who has been posting nonsensical performance and power figures for weeks is now trying to give lessons to others. LOL

The difference in performance between the A10-6800k and the i5-3350p is of about a 10-15%. Therefore the CPU in Kaveri will be slightly faster than the i5 IB.

I think it would be more 2550K-ish performance wise.

I agree with you. The i5-2550k ~ i5-3350p. I am considering a difference in performance of about a 5%, although this intel enthusiast claims a 6%

http://www.tomshardware.co.uk/forum/349920-28-3350p-2500k

That is for the 2500K Dat lack of IGP on the 2550K and 100MHz added :3 Anyways, it should be safe to estimate Kaveri being around 2550K performance. inb4hajifursaysitwillbelessthanbloomfield

 

[juanrga]

The difference in performance between the A10-6800k and the i5-3350p is of about a 10-15%. Therefore the CPU in Kaveri will be slightly faster than the i5 IB.

http://www.kitguru.net/components/cpu/henry-butt/amd-a10-6800k-richland-apu-review/10/
lol, are you people mad. Its crazy 10-15% in your world in real world is around 80%. The a10-6800k competitor for cpu performance is i3, stop kidding yourself.

edit look at i3 3220 3dmark11 physics score:
http://www.hardwaresecrets.com/article/A10-5800K-vs-Core-i3-3220-CPU-Review/1646/17

I would say the i3 3220 and a10-6800k cpu are on par.
one more thing:
http://www.pcper.com/reviews/Processors/Intel-Core-i5-3350P-Ivy-Bridge-Processor-Review-No-Integrated-Graphics/Overclocki
'To reiterate - the Core i5-3350P uses 110 watts LESS POWER than the FX-8150 while outperforming it in many cases. Crap.' Good luck a10 apu competing againt the i5 lol.

LOL

A cinebench score is compared to a cinebench score of a OC i5!!!! Substract the 15% coming from the overclock and then about a 10-15% from the cripple _AMD function and you will get the performance of the i5.

I know that kitguru writes some of most funny reviews, but the hardwaresecrets link also did make me laugh. They give a CPU score for the trinity A10-5800k (3.8GHz) slower than for the Llano A8-3870K (3.0GHz; no turbo) and they are happy!!!!

Your ridiculous, laughable, nonsensical, trolling claim that the i3 is 80% faster than the trinity A10 has to be cut by a factor of ten up to 8%. Look to this 3dmark physics score

Regarding CPU: Kaveri > A10-6800k > A10-5800k > i3 3220.

Stop trolling people. Let's make it clear:

Fact 1. i3-3220 is 20-28% faster in SINGLE THREAD.
Source: http://www.tomshardware.com/reviews/a10-6700-a10-6800k-richland-review,3528-3.html
Fact 2. Kaveri will struggle with Haswell parts: +10%
Fact 3. i5 has turbo: +15%
Fact 4. i5 has 4 true cores, Kaveri only 2 modules.

Conclusion: Kaveri can fight with i3, but has no chance with i5.

Fact 1. i3-3220 is 8% faster in SINGLE THREAD.
Source: http://cpuboss.com/cpus/Intel-Core-i3-3220-vs-AMD-A10-6800K

Fact 1b. AMD claims Steamroller is 15-20% faster.

Fact 2. Haswell i5 is about a 5% faster than Ivy Bridge

Fact 2b. Several reviews are reporting performance regressions. E.g. this one

http://www.pcgameshardware.de/screenshots/original/2013/05/Test-Haswell-4770K-4670K-4570-Crysis-3-v2.png

Fact 3. i5 lacks HT

Fact 3b. Benchmarks used for i5 already consider turbo.

Fact 4. There is not anything as true or false cores, because neither Intel nor you have the monopoly of the definition of core.

Fact 4b. Again, this semantics is all irrelevant, because we are using benchmarks for A10, i3, and i5 and benchmarks don't care about cores/modules definitions neither which has what.

Conclusion: Kaveri fights i5 regarding CPU alone. With HSA software, kaveri will [strike]fight[/strike] destroy i7s.

 

[noob2222]

http://www.tomshardware.com/news/AMD-Kaveri-APU-Gaming,22947.html

juangra 15-20% better ipc means they are about the equivalent performance of a 2.9ghz core i series cpu is a 4ghz kaveri cpu. So at 3.3ghz the i5 3350p will be equivalent to around something like a 4.4ghz kaveri cpu. I don't know why amd fanboys overhype amd's own targets which is less then what you predicted.

you and your ever changing stories.

Wow, you are embarrasing yourself now. Firstly I am on about 4 cores of amd apu at 4ghz being similar to a 2.9ghz core i series cpu, maybe it more a 3.1ghz 2c 4t intel will beat a 4ghz apu quad core. Oh and one more thing.

...
On here the same amd wins 1 more but they are about the same overall within close proximity.

quite a different story going from kaveri will be slower than a 2.9 ghz i series to richland right now being EQUAL TO A 3.3 GHZ i series. Maybe you missed where your fanboy claims that adding 15-20% ipc will actually make kaveri slower than what richland is now.

or maybe your trying to imply that the i3 is not an i-series cpu and that your only comparing it to the i5 wich is ~$100 more.

 

[jdwii]

Are we really comparing a I3 to a A10 or a I5 to a 8 core FX again? When a program is more single threaded Intel wins when you can use all the cores Amd wins End of story.
IPC and clock speed matters and to a extent cores matter as well to a point in programming.
Performance per watt matters a lot in laptops and tablets not nearly as much in desktops and it would take 10 years to even make up the difference between a I5 and 8 core fx.

Intel is only at Trinity level when it comes to its graphics at best and you're dealing with spotty drivers that are worse than even Amd ones.

Future is more cores if it wasn't Intel would not be making a 8 core processor, game developers would be saying different, but clock speed and IPC are still going to matter a lot in Gaming PC's since they use a brute force method of computing games and they have a extra layor to worry about and that is the API and windows

 

[noob2222]

^^ good review, especially the 1100 euro price tag for both the iris pro and the gtx 750m+ haswell that performs 25% faster. so ... why aren't oems rushing to ship every laptop with iris pro?

and sure, being 5 watts more efficient than the A10 6700 (and a whopping 7.5% faster) is really something to brag about. at that cost ($130 for the a10 to $440 for the iris pro) it will take 64 years (24/7 @ $0.11/kwh) to pay the $310 difference in initial cost.

 

[8350rocks]

http://www.youtube.com/watch?v=iEZhU-p2Mak
http://www.youtube.com/watch?v=A7mX3Ysek7g
http://translate.google.co.uk/translate?client=firefox-a&hl=en&ie=UTF8&oe=UTF8&u=http://www.computerbase.de/artikel/grafikkarten/2013/intel-iris-pro-5200-grafik-im-test/
This is what haswell igpu hass done so it beats amd apus. I was comparing before to cinebench r11.5 but now realised it makes sense superpi is real singlethreaded benchmark otherwise why do intel need half the amount of cores to beat amd across the board.

ä That's as funny as assuming just because AMD wins some gaming benchmarks that the Intel CPUs must be "crippled" in some manner. Even though there were CPU-Z/GPU-Z validations to show the settings on those benchmarks...

äI don't know why Intel fanboys feel the need to go on and on about power consumption when no one cares about it at all. Why do they also feel the need to make comparisons to inferior Intel product when the AMD product is clearly on par with much newer product in nearly everything? I wish someone could explain to me why they feel the need to put down AMD product, while all the AMD crowd does is talk about AMD product...no Intel bashing in here...just AMD discussion. Maybe we should start a thread called "Intel fanboys go here" and you all can talk about power consumption and BS comparisons that mock reality at best and are a complete farce at the worst...

What do you think?

DO YOU JUST NOT GET IT?

 

[juanrga]

I agree with you. The i5-2550k ~ i5-3350p. I am considering a difference in performance of about a 5%, although this intel enthusiast claims a 6%

http://www.tomshardware.co.uk/forum/349920-28-3350p-2500k

That is for the 2500K Dat lack of IGP on the 2550K and 100MHz added :3 Anyways, it should be safe to estimate Kaveri being around 2550K performance. inb4hajifursaysitwillbelessthanbloomfield

Yes my mistake now I would correct my estimation by a 1-2%

 

[sarinaide]

FX 9590

Reading through this it seems that the higher clock rates are having profound effects on mitigating IPC lose inherent in the modular design, it would be interesting if since we know the SR cores will evolve a lot iro i/o, cache, IMC and FPO it may be true that a 4ghz SR FX 8XXX part may very well exhibit the same performance as the higher clocked 9590 on significantly less power and lower clock speed. Since the fX9590 is competitive against i7 and Intel Extreme depending on the benches run, carried through all generations could see a very competitive line up against Intel's haswell.

 

[rmpumper]

I have never understood why people care so much about Cinebench numbers. If you want real word rendering performance comparison, just have a look at Tom's 3ds max test results, i.e. from the 4960X review:

 

[szatkus]

Fact 1. i3-3220 is 8% faster in SINGLE THREAD.
Source: http://cpuboss.com/cpus/Intel-Core-i3-3220-vs-AMD-A10-6800K

Yeah, because two synthetic benchmarks are more reliable than some real world benchmarks.

Fact 1b. AMD claims Steamroller is 15-20% faster.

True.

Fact 2. Haswell i5 is about a 5% faster than Ivy Bridge

Ok.

Fact 2b. Several reviews are reporting performance regressions. E.g. this one

http://www.pcgameshardware.de/screenshots/original/2013/05/Test-Haswell-4770K-4670K-4570-Crysis-3-v2.png

**it happens.

Fact 3. i5 lacks HT

Fact 3b. Benchmarks used for i5 already consider turbo.

Captain obvious.

Fact 4. There is not anything as true or false cores, because neither Intel nor you have the monopoly of the definition of core.

Fact 4b. Again, this semantics is all irrelevant, because we are using benchmarks for A10, i3, and i5 and benchmarks don't care about cores/modules definitions neither which has what.

Semantics is very revelant. One module isn't the same thing as two normal cores. Even if you find quad core i5 which single thread performance is on par with some 2 module CPU then i5 will be faster in multithread work. Buy FX-8xxx, lock some threads, do some benchmarks 4m/4t vs. 2m/4t.

Conclusion: Kaveri fights i5 regarding CPU alone. With HSA software, kaveri will [strike]fight[/strike] destroy i7s.

My wrong. With HSA Kaveri is the most powerful CPU in whole universe.

http://www.youtube.com/watch?v=m9kCCLwEga8

 

[szatkus]

I was responded to the i5 3350p vs amd a10-6800k being about 80% faster.
i3 3220 being on par with the a10's which it is looking at the benchmarks you gave out overall. Also cpuboss is a load of BS. Most of there benchmarks are not accurate.

'Conclusion: Kaveri fights i5 regarding CPU alone. With HSA software, kaveri will fight destroy i7s.'
lol using your logic the kaveri cpu will destroy the fx9590 a low power amd cpu destroying there top of the line 220w tdp in your own words if you enable hsa lol. Whats the point of the fx9590 might as well get a kaveri cpu and destroy it with hsa lol.

A4 competes against celeron, a6 competes against pentium, a8 and a10 compete and fx4000 series compete against i3. fx6000 series competes against low end i5's and fx8000/9000 series competes against 4c 8t i7's leaving the i7 extreme 6 core cpus all alone.

Apus have crippled cpu functions compared to the normal cpu line like cache I believe. Also about single threaded performance a 2.4ghz c2d beats a 4ghz fx8350 at single threaded performance. My i5 3317u is about 60% faster then the fx8350 at single threaded benchmark and that was intels slowest i5 ivy bridge ulv at 17w.

http://www.legitreviews.com/article/2055/11/

Superpi is the real benchmark that shows proper single threaded performance. I expect the i3 3220 on superpi to be 2x faster on single threaded benchmarks. It only makes sense as the amd apu scores similar with all 4 cores untilised vs 2 of the intels. Anyway have fun trolling. Steamroller could improve 100% over piledriver or kaveri singlethreaded performance and still lose to a similar clocked intel cpu on single threaded performance. 8.03ghz fx8350 still would lose to a stock 3770k on single-threaded benchmark. Theres a reason why intel compete wit half as many cores as amd as they are just over 2x faster per core.
http://www.youtube.com/watch?v=7meOK_8nnRE

'Super PI is a single threaded benchmark that calculates pi to a specific number of digits. It uses the Gauss-Legendre algorithm and is a Windows port of a program used by Yasumasa Kanada in 1995 to compute pi to 232 digits.'

I understand why cinebench single threaded cores are close as they really use 2 threads for intel and amd on it.

SuperPI? Srsly?

http://24.media.tumblr.com/tumblr_m4inl8uoVC1qazgp3o1_400.png

 

[juanrga]

'Conclusion: Kaveri fights i5 regarding CPU alone. With HSA software, kaveri will fight destroy i7s.'
lol using your logic the kaveri cpu will destroy the fx9590 a low power amd cpu destroying there top of the line 220w tdp in your own words if you enable hsa lol. Whats the point of the fx9590 might as well get a kaveri cpu and destroy it with hsa lol.

Why do you believe that AMD is replacing the 8 piledriver cores Opteron series by the new Berlin APU with 4 steamroller cores? Besides the increase in IPS of the CPU part, Berlin APUs are fully HSA:

CPU+GPU >> CPU

I have never understood why people care so much about Cinebench numbers.

After spending time in forums, I got the conclusion that people uses CB, SuperPi, and similar benchs to say themselves, friends, or other posters how good their processor looks under those benchs.

No owner of a 3930k would be proud if his $600 cpu is only 31 seconds faster than a cheap 8350 in 3ds Max.

 

[juanrga]

Fact 1. i3-3220 is 8% faster in SINGLE THREAD.
Source: http://cpuboss.com/cpus/Intel-Core-i3-3220-vs-AMD-A10-6800K

Yeah, because two synthetic benchmarks are more reliable than some real world benchmarks.

There are good and bad synthetic benchmarks, there are good and bad real world benchmarks. Moreover, not all non-synthetic benchmarks are real world benchmarks.

Fact 1b. AMD claims Steamroller is 15-20% faster.

True.

Fact 2. Haswell i5 is about a 5% faster than Ivy Bridge

Ok.

Fact 2b. Several reviews are reporting performance regressions. E.g. this one

http://www.pcgameshardware.de/screenshots/original/2013/05/Test-Haswell-4770K-4670K-4570-Crysis-3-v2.png

**it happens.

Fact 3. i5 lacks HT

Fact 3b. Benchmarks used for i5 already consider turbo.

Captain obvious.

It was obvious that you took an i3 as baseline and mentioned that an i5 has turbo, but forgot to mention that the i5 lacks HT

Fact 4. There is not anything as true or false cores, because neither Intel nor you have the monopoly of the definition of core.

Fact 4b. Again, this semantics is all irrelevant, because we are using benchmarks for A10, i3, and i5 and benchmarks don't care about cores/modules definitions neither which has what.

Semantics is very revelant. One module isn't the same thing as two normal cores. Even if you find quad core i5 which single thread performance is on par with some 2 module CPU then i5 will be faster in multithread work. Buy FX-8xxx, lock some threads, do some benchmarks 4m/4t vs. 2m/4t.

Semantic is irrelevant for benchmarks. If I give you i5, i3, A10, A4 benchmarks, the results of those benchmarks are independent of if you consider that modules have "true cores" or "normal cores" or otherwise. The benchmark score is the same with independence of your opinion.

That is the reason why I give i3 benchmarks without mentioning if threads are running in two real cores plus two virtual cores or what.

Conclusion: Kaveri fights i5 regarding CPU alone. With HSA software, kaveri will [strike]fight[/strike] destroy i7s.

My wrong. With HSA Kaveri is the most powerful CPU in whole universe.

http://www.youtube.com/watch?v=m9kCCLwEga8

Wrong again. With HSA software, Kaveri is working as an APU (CPU+GPU) not as a CPU. As said before, the CPU alone will be at the level of an i5 CPU

 

[The Q6660 Inside]

FX 9590

Reading through this it seems that the higher clock rates are having profound effects on mitigating IPC lose inherent in the modular design, it would be interesting if since we know the SR cores will evolve a lot iro i/o, cache, IMC and FPO it may be true that a 4ghz SR FX 8XXX part may very well exhibit the same performance as the higher clocked 9590 on significantly less power and lower clock speed. Since the fX9590 is competitive against i7 and Intel Extreme depending on the benches run, carried through all generations could see a very competitive line up against Intel's haswell.

Well I hope steamroller comes out soon and see if amd actually deliver or will it be still way off sandy bridge.

Piledriver has already delivered and in an Apples2Apples comparison, can already best Ivy Bridge performance per dollar wise.

 

[8350rocks]

http://translate.google.co.uk/translate?client=firefox-a&hl=en&ie=UTF8&oe=UTF8&u=http://www.computerbase.de/artikel/grafikkarten/2013/intel-iris-pro-5200-grafik-im-test/

In opencl the irispro spanks the a10 apu and is extremely power efficient.

Most of the power is the cpu as crysis 3 is cpu demanding.
http://translate.google.co.uk/translate?client=firefox-a&hl=en&ie=UTF8&oe=UTF8&u=http://www.computerbase.de/artikel/grafikkarten/2013/intel-iris-pro-5200-grafik-im-test/

AMD are under threat in graphics and cpu as intel are considerably ahead performance per watt in both and features.

^^ good review, especially the 1100 euro price tag for both the iris pro and the gtx 750m+ haswell that performs 25% faster. so ... why aren't oems rushing to ship every laptop with iris pro?

and sure, being 5 watts more efficient than the A10 6700 (and a whopping 7.5% faster) is really something to brag about. at that cost ($130 for the a10 to $440 for the iris pro) it will take 64 years (24/7 @ $0.11/kwh) to pay the $310 difference in initial cost.

+1 LOL...he just doesn't get it.

Perf/Watt is irrelevant here.

Perf/$ is entirely relevant here.

Logic: If I save enough money up front over what the marginally better perf/watt CPU costs (i.e. $100+), then it takes so long to recover the difference in initial cost...that the perf/watt savings are entirely "blue sky" as it won't matter.

Anything that takes longer than 6-12 months to recover in efficiency savings is overstated. If the difference was something like $20 between the 8350 and 3770k, it would be a conversation worth having, but it's not...so the conversation is moot.

Perf/Watt == useless when the initial cost difference is greater than you can recover in 12 months or less.

 

[The Q6660 Inside]

http://translate.google.co.uk/translate?client=firefox-a&hl=en&ie=UTF8&oe=UTF8&u=http://www.computerbase.de/artikel/grafikkarten/2013/intel-iris-pro-5200-grafik-im-test/

In opencl the irispro spanks the a10 apu and is extremely power efficient.

Most of the power is the cpu as crysis 3 is cpu demanding.
http://translate.google.co.uk/translate?client=firefox-a&hl=en&ie=UTF8&oe=UTF8&u=http://www.computerbase.de/artikel/grafikkarten/2013/intel-iris-pro-5200-grafik-im-test/

AMD are under threat in graphics and cpu as intel are considerably ahead performance per watt in both and features.

^^ good review, especially the 1100 euro price tag for both the iris pro and the gtx 750m+ haswell that performs 25% faster. so ... why aren't oems rushing to ship every laptop with iris pro?

and sure, being 5 watts more efficient than the A10 6700 (and a whopping 7.5% faster) is really something to brag about. at that cost ($130 for the a10 to $440 for the iris pro) it will take 64 years (24/7 @ $0.11/kwh) to pay the $310 difference in initial cost.

+1 LOL...he just doesn't get it.

Perf/Watt is irrelevant here.

Perf/$ is entirely relevant here.

Logic: If I save enough money up front over what the marginally better perf/watt CPU costs (i.e. $100+), then it takes so long to recover the difference in initial cost...that the perf/watt savings are entirely "blue sky" as it won't matter.

Anything that takes longer than 6-12 months to recover in efficiency savings is overstated. If the difference was something like $20 between the 8350 and 3770k, it would be a conversation worth having, but it's not...so the conversation is moot.

Perf/Watt == useless when the initial cost difference is greater than you can recover in 12 months or less.

+9001

 

[Cazalan]

Perf/Watt is irrelevant here.

Perf/Watt == useless when the initial cost difference is greater than you can recover in 12 months or less.

+9001

Perf/Watt is always relevant. There may be cases where you choose to ignore it (desktop) but spanning the whole market for CPUs it's very important. Also that recovery period is more like 3-5 years when you're dealing with corporate purchases.

Perf/Watt is single-handedly responsible for AMDs loss of 15% market share of the server market. From around 20% to 5% over the last few years. That has hurt AMD's bottom line significantly. Enough that they're practically ignoring the high end server market now. This leading to delays in Steamroller FX as well.

AMD mentioned a slight up-tick in server sales but didn't give any breakdown. I suspect these are from SeaMicro division. But what exactly was the mix of kit?

SM15000-OP(Piledriver), SM15000-XN(Ivy), SM15000-XE(Sandy), SM15000-64(Atom).

Their up-tick in server sales was likely from selling Intel Kit. Ironic.

It also explains the shift of focus from the 12/16 core CPUs to 4C APUs. The SeaMicro blades are primarily 4C and below. So introducing new APU versions of SeaMicro kit would help gain some market share back.

http://www.seamicro.com/SM15000

 

[noob2222]

always get a laugh when the theory is if you run more, you recover faster.

$310*1000 cpus = $310,000. 5w @1000 cpus = 5,000 watts

its still the same 64 years to recoup the $310,000 cost difference in ppw.

The big issue with corporate world is time = money. PPW is irrelevant, miniscule in comparison. Say that $310k is 7.5% faster. Lets reduce the math back to just one computer and one employee.

lets take for example a good tech employee @$30/hr. 7.5% faster means that you reduced your employee cost by $2.25/hr, provided they are working at 100% efficiency and 100% of all the work they do is dependent on cpu speed. Your PPW cost is 5Wh @$0.11/kWh is $0.00055/hr. PPW is near irrelevant.

your cost is recouped in 17.2 work days in employee labor, or if they are waiting on the cpu more realisticly only 10% of the time, 172 work days. Or in the case of where I work, mabye 1% waiting on the computer, 1720 work days or 6.6 years. <--- this is why we buy cheap comptuers.

obviously on minimum wage labor, the cost vs time are quite different, but the same concept applies. How much time are they waiting on their computer to finish a task.

 

[juanrga]

@szatkus,

it is simpler. Take Richland A10 CPU scores and add a 20% performance.

I obtain ~ i5-2550k ~ i5-3350p.

I also obtain 4C SR ~ Opteron 3000 series. This coincides with AMD replacing Opteron 3000 series with Berlin.

Perf/Watt is always relevant. There may be cases where you choose to ignore it (desktop) but spanning the whole market for CPUs it's very important. Also that recovery period is more like 3-5 years when you're dealing with corporate purchases.

Perf/Watt is single-handedly responsible for AMDs loss of 15% market share of the server market. From around 20% to 5% over the last few years. That has hurt AMD's bottom line significantly. Enough that they're practically ignoring the high end server market now. This leading to delays in Steamroller FX as well.

AMD mentioned a slight up-tick in server sales but didn't give any breakdown. I suspect these are from SeaMicro division. But what exactly was the mix of kit?

SM15000-OP(Piledriver), SM15000-XN(Ivy), SM15000-XE(Sandy), SM15000-64(Atom).

Their up-tick in server sales was likely from selling Intel Kit. Ironic.

It also explains the shift of focus from the 12/16 core CPUs to 4C APUs. The SeaMicro blades are primarily 4C and below. So introducing new APU versions of SeaMicro kit would help gain some market share back.

http://www.seamicro.com/SM15000

There are several reasons why AMD lost server market share and this includes Intel bribes for strong arming OEMs and distributors into buying its lower performing CPUs and not to buy, nor sell AMD based systems and also delays on the launch of Opteron 6000 series processors due to fab problem at GlobalFoundries.

Also AMD is replacing 4C/8C PD CPUs for 4C SR APU/CPU in the 3000 series. In the 6000 series, AMD will be launching 12C/16C PD Warsaw CPU.

 

[The Q6660 Inside]

Intel server cpus destroy amd server cpus like xeon destroys opteron.

Just a baseline performance you can see the higher end xeon cpus are over 2x faster then the opteron 6234 and the 2.6ghz xeon has same tdp.
http://www.cpubenchmark.net/high_end_cpus.html

Now ivy bridge ep suppose to add another 4 cores to make it 12 cores so its like making it 3x faster then amd fastest opteron cpu. This is the significant advantage of having a power efficient architecture as you can have amazing performance if you crank the power up slightly which intel are reluctant to do until 8 core haswell mainstream cpu is out. Its obvious that amd are not much of a threat on the server market as thats when intel can really flex there muscles as companies pay top money there.

http://www.theregister.co.uk/2013/07/18/intel_q2_2013_server_drilldown/

That is per core.
>implying not having massive clusters where companies pay top money
>hajifur
>implying PD arch is not extremely good at multithreaded tasks

 

[The Q6660 Inside]

In actual news, http://www.cpu-world.com/news_2013/2013062401_Athlon_X2_370K_now_available_in_US.html http://www.tomshardware.com/news/AMD-Richland-Athlon-FM2-cpu,23116.html something that has slipped right past us.