[Techpowerup] AMD "Zen" CPU Prototypes Tested, "Meet all Expectations"

[Phynaz]

What are you trying to say ??

That you are wrong?

 

[MrTeal]

But the i3/i5 for instance have good ST perf in isolation but disastrous ST perf in multitasking, as much as 60% lower ST perf.

If a CPU need 5s to execute task A and 5s to execute task B then processing the two task simultaneously is assumed as requiring 10s, that s completely wrong, in the case of the i3 it can be as long as 20-25s.

The exemple is Winrar MT + Cinebench ST, that s just 5 threads :

http://www.computerbase.de/2015-10/...gramm-multitasking-test-cinebench-plus-winrar

CB ST score of the i5-6600K is 165, launch Winrar simultaneously and CB ST score collapse to 77, with a i3 it s even worse, so the conclusion is clear, i3/i5 ST perf cant be sustained if there s two softs using 4 and 1 threads respectively.

So much for Intel s ST perf, it s completely rubbish in a multitasking environment.

You ll notice that the so called inferior ST perf of AMD doesnt collapse that much and there s twice the amount of cores, so the FX sustain its throughput and ST perf much better than those ultra hyped CPUs.



The clue is in the link above...

It schedule multiple threads but that s all what it does, once scheduled the threads land in a stalled pipeline.

I don't think you read what I wrote. It was comparing SMT to CMT, so compare the i7-6700k to the FX-8350. Alternately, compare the i3 to an FX-4300.

 

[Abwx]

I don't think you read what I wrote. It was comparing SMT to CMT, so compare the i7-6700k to the FX-8350. Alternately, compare the i3 to an FX-4300.

There s no FX4300 in the test but it s not difficult to deduct that it will perform like half a FX8350 if they are equally clocked since it has half the cores and half the L3 cache.

Hence while the i7 compare favourably to the FX8350 (the former IS faster in FP and the same speed in Integer...) this is not the case for the i3/i5 in respect of the FX43xxx/FX63xx, certainly due to the L3 cache size since this perf penalty affect also the i5 despite its four hardware cores.

In light of thoses results it s obvious that the i7s and FX8350 are in a superior category, a step below there s the FX63xx and the i5s, another step below and we have the 4 cores FXs while the i3s are at the bottom of the barrel, these chips will be outdated at a faster pace because as PCs use more recent apps they will see their perfs collapsing in an exponential fashion.

 

[MrTeal]

There s no FX4300 in the test but it s not difficult to deduct that it will perform like half a FX8350 if they are equally clocked since it has half the cores and half the L3 cache.

Hence while the i7 compare favourably to the FX8350 (the former IS faster in FP and the same speed in Integer...) this is not the case for the i3/i5 in respect of the FX43xxx/FX63xx, certainly due to the L3 cache size since this perf penalty affect also the i5 despite its four hardware cores.

In light of thoses results it s obvious that the i7s and FX8350 are in a superior category, a step below there s the FX63xx and the i5s, another step below and we have the 4 cores FXs while the i3s are at the bottom of the barrel, these chips will be oudated at a faster pace because as PCs use more recent apps they will see their perfs collapsing in an exponential fashion.

Respectfully I think you would need to show a little evidence of that. The E3-1231v3 is 4C/8T and 3.4GHz/8MB and so should be close to double a 2C/4T/3.5GHz/4MB i3-4330.
In CB MT it gets 719 while the i3 gets 350, which is inline with what you'd expect. In CB MT + Winrar the Xeon gets 425 while the i3 gets 109, which if I understand your argument shows that the four threaded i3 isn't suitable for multitasking.
I don't see the logic in assuming that the FX-4xxx will be superior to the i3 in this use case. If four-thread i3 is loaded with more than four threads and the performance in CB tanks for this specific test, why would you think that the same would not happen when running it on the four thread FX-4xxx?

 

[SlowSpyder]

600+ replies? Zen must be pretty good to ruffle this many feathers. Carry on.

 

[Enigmoid]

There s no FX4300 in the test but it s not difficult to deduct that it will perform like half a FX8350 if they are equally clocked since it has half the cores and half the L3 cache.

But an i3 is half an i7 and the two perform quite differently.

 

[Abwx]

Respectfully I think you would need to show a little evidence of that. The E3-1231v3 is 4C/8T and 3.4GHz/8MB and so should be close to double a 2C/4T/3.5GHz/4MB i3-4330.
In CB MT it gets 719 while the i3 gets 350, which is inline with what you'd expect. In CB MT + Winrar the Xeon gets 425 while the i3 gets 109, which if I understand your argument shows that the four threaded i3 isn't suitable for multitasking.

You did understand the logic but there s another test wich use Winrar MT with likely 4 threads and a single thread from Cinebench, this single thread score is halved as well despite being theoricaly only a marginal increase in CPU loading, yet these are system that are loaded with only thoses apps, the generic apps that we have in PCs are not implemented.

I don't see the logic in assuming that the FX-4xxx will be superior to the i3 in this use case. If four-thread i3 is loaded with more than four threads and the performance in CB tanks for this specific test, why would you think that the same would not happen when running it on the four thread FX-4xxx?

The origin of the problem is the L3 cache, actualy i wouldnt have thought that it could have such an influence but on second thoughts the more threads you have the lower the cache blocks dedicated to each thread, hence the pipeline will stall because the read/writes necessitated by the computations are made with smaller datas blocks wich will increase the number of interuptions.

On benches made with a single soft, even MT, this cant be detected as a single application has the full cache available; of course Intel is aware of this detail and they use the cache size for better segmentation of their products, this way a i7 is much better than a i5 even if on paper this latter is only 25% below.

But an i3 is half an i7 and the two perform quite differently.

Of course, otherwise the price difference wouldnt be justified...

In our case the i3 is significantly less than half a i7.

 

[Abwx]

600+ replies? Zen must be pretty good to ruffle this many feathers. Carry on.

Well, thanks for the reminder, i guess that i was somewhat OT..

As for Zen, and in light of thoses rare benches i have no doubt that it will be more than competitive, rumour is that Zen target is better throughput than a Piledriver module at half the power comsumption, we ll see if they manage to get there but from what we know of GF process it s largely within the realm of possibilities.

 

[MrTeal]

The origin of the problem is the L3 cache, actualy i wouldnt have thought that it could have such an influence but on second thoughts the more threads you have the lower the cache blocks dedicated to each thread, hence the pipeline will stall because the read/writes necessitated by the computations are made with smaller datas blocks wich will increase the number of interuptions.

On benches made with a single soft, even MT, this cant be detected as a single application has the full cache available; of course Intel is aware of this detail and they use the cache size for better segmentation of their products, this way a i7 is much better than a i5 even if on paper this latter is only 25% below.

If it is the case of the application having to go outside the L3 to main memory, wouldn't the having of memory also effect the FX-4XXX chips with 4MB L3 vs the 8MB in the 8XXX, which is the same drop as an 8MB 4790k to a 4MB i3-4330?

 

[Enigmoid]

You did understand the logic but there s another test wich use Winrar MT with likely 4 threads and a single thread from Cinebench, this single thread score is halved as well despite being theoricaly only a marginal increase in CPU loading, yet these are system that are loaded with only thoses apps, the generic apps that we have in PCs are not implemented.




The origin of the problem is the L3 cache, actualy i wouldnt have thought that it could have such an influence but on second thoughts the more threads you have the lower the cache blocks dedicated to each thread, hence the pipeline will stall because the read/writes necessitated by the computations are made with smaller datas blocks wich will increase the number of interuptions.

On benches made with a single soft, even MT, this cant be detected as a single application has the full cache available; of course Intel is aware of this detail and they use the cache size for better segmentation of their products, this way a i7 is much better than a i5 even if on paper this latter is only 25% below.



Of course, otherwise the price difference wouldnt be justified...

In our case the i3 is significantly less than half a i7.

Winrar uses more than 4 threads as the 5930k is significantly faster than the 4790k.

The point with the i3 is that the i7 is half an i7 yet ST performance collapses. Its not due to the nature of the core, its due to the nature of the test (thread count matters). Thus there is no support for the claim that a FX 4300 is half a 83xx CPU so it will perform similarly.

The other thing that you seem unwilling to believe is that multitasking performance is crashing on some CPUs because CB is simply not being allocated any CPU time.

Look at the i5-5675C.

Winrar: 2:44
Winrar + CB: 2:46

CB: 561
CB + Winrar: 175

Looks like the CPU is simply allocating most of its throughput toward Winrar.
Scaling looks like >100%; this is likely because Winrar is limited in some other fashion (latency limited I would guess).

 

[Abwx]

If it is the case of the application having to go outside the L3 to main memory, wouldn't the having of memory also effect the FX-4XXX chips with 4MB L3 vs the 8MB in the 8XXX, which is the same drop as an 8MB 4790k to a 4MB i3-4330?

Intel has inclusive caches wich reduce the total cache size more than exclusive cache designs when part of the L3 is fused off.

I benefit from this thread to point that the FX4350 has 8MB cache and should be comparatively better than the FX8XXX in multitasking, i.e, it s more than half a FX83xx.

That said i think that AMD s much criticised write/through implementation wich was branded as the culprit for ST perfs was choosen because of heavily multithreaded environments in servers pplications wich was the target, Intel on the other hand rely onhuge caches for server CPUs, i guess that now everybody knows why....

 

[Abwx]

Winrar uses more than 4 threads as the 5930k is significantly faster than the 4790k.

Winrar is core count aware, it will use 4 threads in a 4T CPU....

Look at the i5-5675C.

Winrar: 2:44
Winrar + CB: 2:46

CB: 561
CB + Winrar: 175

Looks like the CPU is simply allocating most of its throughput toward Winrar.
Scaling looks like >100%; this is likely because Winrar is limited in some other fashion (latency limited I would guess).

I noticed this detail too, the chip prioritize integer code over FP code, same with the i5, what is obvious is that theses are not computing dedicated CPUs.

Intel somewhat managed to segment their offering to make the i7 the CPU of choice for heavy computing while negating this possibility to the mainstream CPUs, my theory is that it s thanks to the cache, but who knows..?..

Edit : We can eventualy transfer this discussion in a relevant thread and get back on topic in this one....

 

[Dresdenboy]

Because it obviously would be impossible move threads between cores.

How about renaming the logical cores?

 

[tential]

600+ replies? Zen must be pretty good to ruffle this many feathers. Carry on.

Same people drumming up replies. I mean, I just got a performance estimate of Zen being 90% of Broadwell-E performance at 95 TDP vs Broadwell-E's 140 TDP but will be able to overclock to the 140 TDP of Broadwell-E.... I mean, the performance estimates of Zen no know ceilings. Then, it'll come out a lot less than expected, and we'll be hearing the same hype on Zen+

 

[scannall]

Same people drumming up replies. I mean, I just got a performance estimate of Zen being 90% of Broadwell-E performance at 95 TDP vs Broadwell-E's 140 TDP but will be able to overclock to the 140 TDP of Broadwell-E.... I mean, the performance estimates of Zen no know ceilings. Then, it'll come out a lot less than expected, and we'll be hearing the same hype on Zen+

Just going by the promotional stuff, and assorted leaks my own guess is that it will be somewhere between Ivy Bridge and Haswell in IPC. If it overclocks well, and has more than 4 cores I'll at least give it a look when I build a new system next year.

But yeah, some of these guesses are pretty crazy. And that's all anyone here is doing, just guessing.

 

[DrMrLordX]

Edit : We can eventualy transfer this discussion in a relevant thread and get back on topic in this one....

This discussion is relevant, in that AMD is making the transition from CMT-based modules to SMT-capable cores by moving to Zen. There is an open question as to what exactly AMD means by a 40% increase in IPC. We also must wonder how Zen's SMT implementation compares to Intel's (namely: does SMT only add 30% to the "throughput" of the core?). XV is quite capable as a design. It is the implementation that leaves much to be desired. For Zen to be 40% faster than that would appear to make Zen very, very powerful. Perhaps "40% higher throughput" is not what AMD intended to say. That is how I personally interpret it, since the alternative - considering Zen to be 40% faster when running a single, isolated thread per core vs. XV running an isolated thread per module - would make Zen lower in overall throughput than XV. Zen would only have the potential advantage of requiring fewer transistors per core vs an XV module, maybe.

And we still have no idea how clockspeeds will look, nor do we know anything reliable about voltage scaling of the 14nm LPP process. But it is fodder for discussion of interest.

 

[Abwx]

This discussion is relevant, in that AMD is making the transition from CMT-based modules to SMT-capable cores by moving to Zen. There is an open question as to what exactly AMD means by a 40% increase in IPC. We also must wonder how Zen's SMT implementation compares to Intel's (namely: does SMT only add 30% to the "throughput" of the core?). XV is quite capable as a design. It is the implementation that leaves much to be desired. For Zen to be 40% faster than that would appear to make Zen very, very powerful. Perhaps "40% higher throughput" is not what AMD intended to say. That is how I personally interpret it, since the alternative - considering Zen to be 40% faster when running a single, isolated thread per core vs. XV running an isolated thread per module - would make Zen lower in overall throughput than XV. Zen would only have the potential advantage of requiring fewer transistors per core vs an XV module, maybe.

And we still have no idea how clockspeeds will look, nor do we know anything reliable about voltage scaling of the 14nm LPP process. But it is fodder for discussion of interest.

The 40% IPC is undoubtly related to the ST perf as 40% including HT would put a Zen core throughput at 70% of an EXV module, and at 14nm this latter would be much more relevant in such a case.

Besides this 40% is a combined number that include FP and Integer code, what is sure is that FP will benefit from a bigger uplift than Integer given the massive increase of the FPU width and capabilities wich is theoricaly as much as twice a EXV module throughput, FTR an EXV module has about the FP throughput of an Intel core + HT, so they could had recycled EXV FPU as it is if it wasnt for new instructions support.

Integer wise the four ALUs allow theoricaly twice the integer throughput of an EXV core and likely close to an EXV module but the real perf will be closer to what they announced since they lag much less in Integer ST perf than in FP.

Now perhaps that we should ask their competitor what they think about this 8C/16T future competitor, of course they will say nothing but a hint is that they are projecting to compete better by increasing the core count to 10...

 

[AtenRa]

This discussion is relevant, in that AMD is making the transition from CMT-based modules to SMT-capable cores by moving to Zen. There is an open question as to what exactly AMD means by a 40% increase in IPC. We also must wonder how Zen's SMT implementation compares to Intel's (namely: does SMT only add 30% to the "throughput" of the core?). XV is quite capable as a design. It is the implementation that leaves much to be desired. For Zen to be 40% faster than that would appear to make Zen very, very powerful. Perhaps "40% higher throughput" is not what AMD intended to say. That is how I personally interpret it, since the alternative - considering Zen to be 40% faster when running a single, isolated thread per core vs. XV running an isolated thread per module - would make Zen lower in overall throughput than XV. Zen would only have the potential advantage of requiring fewer transistors per core vs an XV module, maybe.

And we still have no idea how clockspeeds will look, nor do we know anything reliable about voltage scaling of the 14nm LPP process. But it is fodder for discussion of interest.

Yea now that you mentioned it,

If ZEN Single Thread performance (IPC) is 40% higher than Excavator, the ZEN SMT scaling should be at 60% just to reach the same Throughput as the CMT Excavator Module at the same clock.

Are we missing something here ??

 

[AtenRa]

Same people drumming up replies. I mean, I just got a performance estimate of Zen being 90% of Broadwell-E performance at 95 TDP vs Broadwell-E's 140 TDP but will be able to overclock to the 140 TDP of Broadwell-E.... I mean, the performance estimates of Zen no know ceilings. Then, it'll come out a lot less than expected, and we'll be hearing the same hype on Zen+

That 90% performance at 95W TDP was an example, not a prediction. Read more carefully next time.

 

[Phynaz]

"We expect across a wide variety of workloads for Barcelona to outperform Clovertown by 40 percent."

-AMD

 

[tential]

That 90% performance at 95W TDP was an example, not a prediction. Read more carefully next time.

An example of what.... what was your point? To give a scenario where I might pick Zen? I will pick Zen if it's a decent chip. But as you've pointed out, it won't beat Broadwell-E. So I won't buy it for performance. If it's a cheap way to get a lot of cores on one chip, sure I'll get Zen. But at the end of the day, Intel still takes the big dollars, and possibly the small ones too if Zen isn't cheap enough.

Just going by the promotional stuff, and assorted leaks my own guess is that it will be somewhere between Ivy Bridge and Haswell in IPC. If it overclocks well, and has more than 4 cores I'll at least give it a look when I build a new system next year.

But yeah, some of these guesses are pretty crazy. And that's all anyone here is doing, just guessing.

I mean, if it's between Ivybridge/Haswell, then it's useless to me. I don't need a slower chip, I need something faster. So for me, I'll look at it sure, but it really has to beat out Broadwell-E, and I just don't see it beating out an intel processor for my workloads. Now if they make a chip that's in that IPC range, but has 2-4 more cores than the same intel processor in the same price bracket? Great!

If they can get a i5-i7 priced processor with 8 cores 16 threads? (~$250). Sure. Then I'll look at it for my server.

 

[frozentundra123456]

600+ replies? Zen must be pretty good to ruffle this many feathers. Carry on.

Either that or a lot of AMD supporters are desperate to keep the hype train rolling at full speed.

 

[TurtleCrusher]

Either that or a lot of AMD supporters are desperate to keep the hype train rolling at full speed.

There's no hype train. People are interested since it's the only interesting thing coming for PCs. How many posts in this thread are just Intel shills spouting the same bullshit over and over and over..?

"No ridiculous R&D money spent, no way it'll be good"

"Cant match Intel's Process Node"

"It won't matter if it's comparable anyway"

"It was cancelled anyway"

How many useless replies are there in this thread by people who will never even consider an AMD chip? This is not what the enthusiast community is about. There's a clear line between enthusiast/consumers and shills/fanboys, and some of you make it all too obvious. Why even bother posting in this thread? It seems like all of you shareholders should start thinking about short selling since you seem to care so much about Zen, knocking it when there isn't hardly any concrete information out there. You wouldn't bother unless you were afraid.

 

[bystander36]

As far as the poll goes, there is a lot of room between Haswell-E 8 core and bulldozer status. I'd consider it a success even if it isn't quite that fast.

 

[frozentundra123456]

There's no hype train. People are interested since it's the only interesting thing coming for PCs. How many posts in this thread are just Intel shills spouting the same bullshit over and over and over..?

"No ridiculous R&D money spent, no way it'll be good"

"Cant match Intel's Process Node"

"It won't matter if it's comparable anyway"

"It was cancelled anyway"

How many useless replies are there in this thread by people who will never even consider an AMD chip? This is not what the enthusiast community is about. There's a clear line between enthusiast/consumers and shills/fanboys, and some of you make it all too obvious. Why even bother posting in this thread? It seems like all of you shareholders should start thinking about short selling since you seem to care so much about Zen, knocking it when there isn't hardly any concrete information out there. You wouldn't bother unless you were afraid.

Why would I be "afraid" of Zen being a good chip? I just get tired of the endless, ever inflating predictions from the AMD fans in these forums. And BTW, I own no Intel stock, and have absolutely no financial interest in Intel. Just because someone does not buy into endless hype does not mean they have a financial interest in a competitor. You might also be surprised to know that I used to be a fan of AMD cpus, and I am currently using an AMD gpu. So if they make a product that meets my needs I will buy it.