AMD zen pre-release thread!

[Abwx]

They can target whatever on Earth they want...that has no bearing on what they will actually achieve.

The fear i was talking about a few post above...

What does this bring to the discussion.?.

Throughput of an Excavator module would be around that of Broadwell IPC (+SMT).

Throughput of an Excavator module is more than BDW throughput Integer wise, that s why i mentioned it as basis, i didnt said that they would extract this amount..

Personally, I think they have absolutely no chance in hell of reaching that.

Lol, with a 28nm process they bested Intel in perf/Watt up to 25W TDP..

But if you think so it s great since you have no more reasons to post in this thread as your opinion is definitly made, this will let other people give their own estimations without the usual redundant noises that fill all AMD related threads.

 

[jdubs03]

The fear i was talking about a few post above...

Lol, with a 28nm process they bested Intel in perf/Watt up to 25W TDP..

Proof? I find that statement quite suspect.

Edit:
http://www.notebookcheck.net/AMD-Kaveri-A10-7300-Notebook-Processor.117333.0.html
http://www.notebookcheck.net/Acer-Aspire-E5-551-T8X3-Kaveri-A10-7300-Notebook-Review.122063.0.html

Compared against i5-4200U at 22nm:

http://www.notebookcheck.net/Intel-Core-i5-4200U-Notebook-Processor.93563.0.html
http://www.notebookcheck.net/Fujitsu-Lifebook-U574-Ultrabook-Review.133175.0.html

I don't know where you're coming up with better perf/W, based on the power measurements the i5 is lower, yet higher in performance, especially single-core. Even 35W Kaveri FX-7600P barely matches/beats an i5-4200U in CPU perf, and surely not perf/W.

 

[Techhog]

Must be awfully hard to design a CPU for a market that's two or three years in the future.
I could see like a year, but two or three is ages.

You seriously think that CPUs are typically designed in 12 months?

 

[Abwx]

Proof? I find that statement quite suspect.

Edit:
http://www.notebookcheck.net/AMD-Kaveri-A10-7300-Notebook-Processor.117333.0.html
http://www.notebookcheck.net/Acer-Aspire-E5-551-T8X3-Kaveri-A10-7300-Notebook-Review.122063.0.html

Compared against i5-4200U at 22nm:

http://www.notebookcheck.net/Intel-Core-i5-4200U-Notebook-Processor.93563.0.html
http://www.notebookcheck.net/Fujitsu-Lifebook-U574-Ultrabook-Review.133175.0.html

I don't know where you're coming up with better perf/W, based on the power measurements the i5 is lower, yet higher in performance, especially single-core. Even 35W Kaveri FX-7600P barely matches/beats an i5-4200U in CPU perf, and surely not perf/W.

I was talking of Beema/Mullins up to 14W and Carrizo up to 25W.



Now let look at the reviews you linked one after the other starting with this one :

http://www.notebookcheck.net/Acer-Aspire-E5-551-T8X3-Kaveri-A10-7300-Notebook-Review.122063.0.html

Where you didnt notice at the end of the article that :

Follow-up: In the course of further, more detailed Kaveri benchmarks, we installed the latest driver bundle. Only this allows measuring MANTLE effects, among other things. That leads to significantly better results in the single thread performance.

So 15.6" for the Kaveri in a dubbious review on one side and a 13.3" a with a SSD for the Haswell, i guess that it s what you a call a fair and non biaised comparisons according to your "standards"?.

You expect me to comment such an obvious bad faith of yours.?.

Sorry, i have no time to waste.

 

[jdubs03]

Why are you taking this as a personal attack? Sure I didn't read the whole thing, I stopped with the power consumption number; because I have better things to do with my time. Additionally, there is only one actual review of a device with an A10-7300, so don't fault me for AMD not getting more devices, including those with SSDs and new drivers in the hands of reviewers.

Go look at the FX-7600P (35W TDP) benchmarks and you'll see that even on an AMD reference platform it can only match and maybe beat on a couple benches against an i5-4200U in CPU.

I have not seen you once give credit to Intel, its always AMD AMD AMD. Whereas in all of my comments given due credit to ANY vendor, be it Intel, AMD, Apple, Nvidia, TSMC, Samsung, etc.

Why don't you try to be a less partial and less confrontational. Saying what I write is in bad faith just isn't right and it certainly isn't courteous.

Edit: I'll throw in Geekbench just for kicks:
A10-7300: http://browser.primatelabs.com/geekbench3/search?dir=desc&q=a10-7300&sort=score
FX-7600P: http://browser.primatelabs.com/geekbench3/search?dir=desc&q=fx-7600p&sort=score

 

[AtenRa]

Kaveri single Module throughput can reach Haswell single Core throughput when the application is not Intel optimized (x264, Pov-Ray, 7Zip etc)

Adding an L3 cache would make it have even higher Throughput. If you just port Excavator from 28nm to 14nm FF (without any architectural changes) it would have the same/higher throughput as Broadwell and reach the same perf/watt.

So even if they will not even change the architecture they could have an extremely powerful Core at 14nm FF. Now add that they are focusing more on efficiency than when they were designing the Bulldozer Architecture and we can come to a conclusion that ZEN could produce highly competitive SKUs against Intels 14nm products in 2016.

Those that they believe that ZEN is a small core design are mistaken, you dont need 4+ years and your best/biggest team for a small core design like Jaguar/Puma.

 

[Abwx]

Why are you taking this as a personal attack? Sure I didn't read the whole thing, I stopped with the power consumption number; because I have better things to do with my time. Additionally, there is only one actual review of a device with an A10-7300, so don't fault me for AMD not getting more devices, including those with SSDs and new drivers in the hands of reviewers.

It s not like equivalent models using Intel CPUs are not available, actualy they are a majority and few use a SSD for instance, why increase the difficulty of a comparison by selecting non ressembling devices.?.

Go look at the FX-7600P (35W TDP) benchmarks and you'll see that even on an AMD reference platform it can only match and maybe beat on a couple benches against an i5-4200U in CPU.

I didnt say that it beat an Haswell CPU wise but it beat it in CPU + GPU efficency, wich is not displayed by the CPU + GPU stress test.

I have not seen you once give credit to Intel, its always AMD AMD AMD. Whereas in all of my comments given due credit to ANY vendor, be it Intel, AMD, Apple, Nvidia, TSMC, Samsung, etc.

I give credit where it is due, Intel has been given more than their fair share in the mobile department with efficency exagerated and the belief that their superiority perf/Watt wise in the DTs would extend to low power solutions.
That s not the case and a 19W Kaveri throttled at 15W level still manage to achieve the CB 11.5 score of genuines 15W Haswells, so AMD is not late when it comes to mobile even with Kaveri.

A 13.3" Kaveri, the CPU run at barely 15W in the benches :

http://www.notebookcheck.com/Test-HP-EliteBook-725-G2-Notebook-J0H65AW.126960.0.html

Why don't you try to be a less partial and less confrontational. Saying what I write is in bad faith just isn't right and it certainly isn't courteous.

Edit: I'll throw in Geekbench just for kicks:
A10-7300: http://browser.primatelabs.com/geekbench3/search?dir=desc&q=a10-7300&sort=score
FX-7600P: http://browser.primatelabs.com/geekbench3/search?dir=desc&q=fx-7600p&sort=score

Sorry if you took it as confrontational, it wasnt my intent, as for Geekbench i dont pay much attention to this bench when it comes to Windows dedicated items.

 

[CruciallyBad]

This tar ball was posted to the mailing list, the tar ball has the following flags:{ "CPU_ZNVER1_FLAGS",+"Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuSYSCALL|CpuRdtscp|Cpu387|Cpu687|CpuFISTTP|CpuNop|CpuMMX|CpuSSE|CpuSSE2|CpuSSE3|CpuSSE4a|CpuABM|CpuLM|CpuFMA|CpuFMA4|CpuBMI|CpuF16C|CpuCX16|CpuClflush|CpuSSSE3|CpuSVME|CpuSSE4_1|CpuSSE4_2|CpuAES|CpuAVX|CpuPCLMUL|CpuLZCNT|CpuPRFCHW|CpuXsave|CpuXsaveopt|CpuFSGSBase|CpuAVX2|CpuMovbe|CpuBMI2|CpuRdRnd|CpuADX|CpuRdSeed|CpuSMAP|CpuSHA|CpuXSAVEC|CpuXSAVES|CpuClflushOpt|CpuCLZERO" }

So FMA4 is still in there, and they also added AVX2. Phoronix didn't actually bother to fact check.

 

[monstercameron]

http://www.kitguru.net/components/c...aster-than-current-x86-cores-zen-is-incoming/

good jump!
products availability? approximately late 2016 early 2017 [a la kaveri]

 

[toyota]

I dont think 40% is enough. there are a few games where it takes 4.5 on the 8350 just to match a 4770k at 2.5. and of course Intel will have Skylake out probably 9 months to a year before the Zen FX cpus even get launched.

 

[monstercameron]

I dont think 40% is enough. there are a few games where it takes 4.5 on the 8350 just to match a 4770k at 2.5. and of course Intel will have Skylake out probably six months to a year before the Zen FX cpus even get launched.

we don't know the clock speeds yet and this has [allegedly] has double the fpu compared to the fx which might have other effects on performance.

 

[StinkyPinky]

I dont think 40% is enough. there are a few games where it takes 4.5 on the 8350 just to match a 4770k at 2.5. and of course Intel will have Skylake out probably 9 months to a year before the Zen FX cpus even get launched.

Yeah but you have no idea at the clock speeds. Although I find it hard to believe they can make an 8 core CPU clocked higher than 3.5. In saying that, if there's no onboard graphics....

I guess the question is...

Do you want an 8 core CPU at IB/Haswell IPC?
Or a 4 core with HT at skylake IPC?

 

[Abwx]

http://www.kitguru.net/components/c...aster-than-current-x86-cores-zen-is-incoming/



good jump!
products availability? approximately late 2016 early 2017 [a la kaveri]

Look like it desesperate some people who are flooding the other thread, it s all deffamatory statements that it s a best case, when everyone with an ounce of knowledge on the matter know that AMD always use averages...

 

[Abwx]

I dont think 40% is enough. there are a few games where it takes 4.5 on the 8350 just to match a 4770k at 2.5. and of course Intel will have Skylake out probably 9 months to a year before the Zen FX cpus even get launched.

That has nothing to do with IPC but rather cache managements, in integer code, like in games, 40% will be more than enough to match an Haswell and even beat it in some instances.

 

[CHADBOGA]

Look like it desesperate some people who are flooding the other thread, it s all deffamatory statements that it s a best case, when everyone with an ounce of knowledge on the matter know that AMD always use averages...

Must have been one hell of an average to predict that Bulldozer wouldn't suffer IPC loss.

 

[Sequences]

http://www.kitguru.net/components/c...aster-than-current-x86-cores-zen-is-incoming/

good jump!
products availability? approximately late 2016 early 2017 [a la kaveri]

That's one hell of a jump for 40%. The chart makes you think it is several times the multiple.

 

[Arachnotronic]

That's one hell of a jump for 40%. The chart makes you think it is several times the multiple.

Marketing -- everybody does it

 

[Abwx]

Must have been one hell of an average to predict that Bulldozer wouldn't suffer IPC loss.

And a hell of a temptation when thread crapping AMD threads with irrelevant posts...

Did they state a %age for BD like they did for Zen, or eventualy Llano s 6%, or SR s 10% on average, or Excavator 5%..?..

AMD numbers have been historically accurate, whatever the nay sayers perpetual deffamations.

 

[toyota]

That has nothing to do with IPC but rather cache managements, in integer code, like in games, 40% will be more than enough to match an Haswell and even beat it in some instances.

how does that make any sense? the improvement in IPC will partly come from addressing those things you just mentioned...

 

[Asterox]

Ok now is officially confirmed AMD Zen will have SMT. This is certainly very good news, especially for future Desktop Zen APU or more precisely 4 Cores / 8 Threads.

 

[crashtech]

40% improvement gets them caught up, more or less. Giving up on CMT is also a very good sign. They can't release this soon enough!

 

[Abwx]

how does that make any sense? the improvement in IPC will partly come from addressing those things you just mentioned...

IPC wise there s not that much difference in Integer code, best cases are softs like Fritzchess with 41.5%, and that s comparatively to the FX8350, so with 40% over excavator they should largely match the competition, in games the FX will often waste IPC by loading full modules, this will decrease the IPC by 15% or so, hence the varying numbers in games.

 

[Phynaz]

Puma+ and A57 besed Skybridges will be released in 2015 since they displayed the slide with this year at the japanese conference

I guess not.

 

[CHADBOGA]

And a hell of a temptation when thread crapping AMD threads with irrelevant posts...

Did they state a %age for BD like they did for Zen, or eventualy Llano s 6%, or SR s 10% on average, or Excavator 5%..?..

AMD numbers have been historically accurate, whatever the nay sayers perpetual deffamations.

Defamation only has one "f".

At this stage, I'm a lot more hopeful that AMD will produce a half decent CPU than I was at the same stage with Bulldozer, but a healthy dose of skepticism is a must, with any so called performance % increases quoted by any Semi-conductor entity.

 

[ZGR]

I hope Zen is a success. If it fails to compete AMD will be in a bit of trouble. I want Zen to make my i5 look like terrible value!