AMD Ryzen (Summit Ridge) Benchmarks Thread (use new thread)

[Toettoetdaan]

Pentiums with HTT and a k model i3,that's about as much pressure as intel felt from ZEN.

And a 6700k that is allowed to consume 5% more power called the 7700k.

Intel knows that AMD probably can't match the IPC and the clock, so with most games still favouring quad-cores, the hello effect is already countered.

 

[TheELF]

And a 6700k that is allowed to consume 5% more power called the 7700k.

The whole line up gets better clocks,this would have happened regardless.

As would the trickle down effect of pentiums getting HTT,like back in ~2010-11 where i5 still had 2c/4t before finally getting 4c/4t and 2c/4t being trickled down to i3 exclusively.
We just don't know,and never will,if this happened now because of ZEN or if it was time for it to happen anyway.

 

[KTE]

If the 8 core is >$400-460 I can't see it being a bad performer. It won't be far off the 6800K/5960X then.

But I find it difficult to see anyone doing that at 14nm at below < 125W, simply due to the clocks required.

It will still not be able to challenge the high-end, not even for the desktops.

If their top-end chip is challenging 6850K/6900K... We will have a winner all-round. But it will be priced around the $599 mark at the very least.

The performance bands trickle down. That's why high IPC+Clocks is always the main design aim for x86. You can then segment that appropriately.

Sent from HTC 10
(Opinions are own)

 

[bjt2]

If the 8 core is >$400-460 I can't see it being a bad performer. It won't be far off the 6800K/5960X then.

But I find it difficult to see anyone doing that at 14nm at below < 125W, simply due to the clocks required.

It will still not be able to challenge the high-end, not even for the desktops.

If their top-end chip is challenging 6850K/6900K... We will have a winner all-round. But it will be priced around the $599 mark at the very least.

The performance bands trickle down. That's why high IPC+Clocks is always the main design aim for x86. You can then segment that appropriately.

Sent from HTC 10
(Opinions are own)

Bristolridge in OC reach 4.9GHz at 1.43V on the 28nm BULK with HDL libraries.
On hardocp we are told that Kabylake OC'ed at 4.8GHz with 1.3V (some specimens arrive also at 5.1GHz with 1.37V), but with an HP 14nm FF plus probabily HPL libraries...

This is only possible if Bristolridge has a very low FO4.

Zen should have similar FO4, but has a new process that has:
1) lower Vth (i was told that the difference is negligible, but a bit it's better than none)
2) higher transconductance (AFAIK > +20%)
3) lower parasitic capacitance (AFAIK > -20%)
4) lower leakage (AFAIK ~1/6)

This, at same FO4, translated, at 4.9GHz, at least -200mV of Vcore, plus lower power consumption. And we can also lie in stable conditions... Anyway probabily 4.5GHz (at least of single core max turbo) are easily feasible even for an 8 core Zen.

 

[jpiniero]

As would the trickle down effect of pentiums getting HTT,like back in ~2010-11 where i5 still had 2c/4t before finally getting 4c/4t and 2c/4t being trickled down to i3 exclusively.

That's not actually true, the original i5 (Lynnfield) was quad core. That was when the enthusiast processors started at $284, heh.

 

[guskline]

The highest end Summit Ridge (Zen) 8c/16t overclockable cpu will not be a $400 chip. especially since the 6950 is @$1000 and AMD was touting the ZEN as matching a Broadwell E 10 core, a $1,700 chip, in one test. My guess is at least $699 if not $799. For goodness sake the 9590 had a very large release price.

I really hope that the AMD event on the 13th gives some detail on the specs of the ZEN they are running, especially when they show off gaming. I would really like to see the comparison with an AMD 9570?, Intel 6700k and Intel 6950.

 

[cytg111]

The highest end Summit Ridge (Zen) 8c/16t overclockable cpu will not be a $400 chip. especially since the 6950 is @$1000 and AMD was touting the ZEN as matching a Broadwell E 10 core, a $1,700 chip, in one test. My guess is at least $699 if not $799. For goodness sake the 9590 had a very large release price.

I think that is a very hard call, several conjectures spring to mind.
1. Market penetration. May want to establish your self in a segment at a cost.
2. Yields are bad, you sell every last chip you make at a high as possible premium.
3. Yields are good, chips are spitting out of the factory, you may have to lower prices to sell them all (inventory is a silent killer)
4. Performance and clocks may not be what we expect them to be, may not be competitive with said products, thus going into the old "moar coars vs singlethread perf" scenario.

Calling a "right" price right now is being brave IMO.

ie. "For goodness sake the 9590 had a very large release price"
- Maybe they binned 2 pr day? Would almost go at any price to *some* nut .

 

[KTE]

Anyway probabily 4.5GHz (at least of single core max turbo) are easily feasible even for an 8 core Zen.

I think you are seriously being irrational.

4.5GHz at least??

Mate, it's either Clocks or IPC, and the third to balance is Power.

You here are saying we will get the best of all 3!!

Sent from HTC 10
(Opinions are own)

 

[TheELF]

That's not actually true, the original i5 (Lynnfield) was quad core. That was when the enthusiast processors started at $284, heh.

Huh?
https://web.archive.org/web/2016112...E60-F6A50F491DFB/July_18_10_1ku_Price.xls.pdf
No matter,there where i5s with 2c/4t back in the day and then there where none,that was my point and I don't think that it changes if there was a quad version i5 first or if there wasn't.

$284 would get you a normal 4c/8t i7.
i7-860 (8M L3 cache, 4 Cores, 8 Threads, 2.80 GHz 45nm) $284

 

[bjt2]

I think you are seriously being irrational.

4.5GHz at least??

Mate, it's either Clocks or IPC, and the third to balance is Power.

You here are saying we will get the best of all 3!!

Sent from HTC 10
(Opinions are own)

It's matter of FO4 and clock. Not IPC and clock. It is not impossible to do an high IPC design with low FO4.

Just design a very high FO4, high IPC CPU, then break the stages with flip flops and increase pipeline stage number. You lose something for branch miss penality, but if branch prediction is good, it's negligible. So you have an high IPC low FO4 design.

We know that Zen will be a low FO4 design because of the OFFICIAL 19 stage integer pipeline, that is in the same ballpark than Bulldozer (15-20). We know that will be an high IPC design because the declared IPC is 40% more than Bulldozer.

Let me be more clear:
Let's suppose BD/XV FO4 is 17 (the number is unimportant). It reaches 4.9GHz at 1.43V on the 28nm BULK and HDL libraries, with all 4 cores, 2 modules.
Now we have a Zen core, that has the same (moreless) FO4 and the same (moreless) transistors than a XV module (2 cores), half than the overclocked XV depicted above.
With the same FO4 and half transistors, a zen core is on the 14nm FF that has:
1) Less Vth
2) Less leakage
3) Less parasitic capacitance
4) More transconductance (transistor strength)

And do you think that a Zen core can not arrive at least at 4.9GHz of turbo core on one core?

 

[Nothingness]

Just design a very high FO4, high IPC CPU, then break the stages with flip flops and increase pipeline stage number. You lose something for branch miss penality, but if branch prediction is good, it's negligible. So you have an high IPC low FO4 design.

Increasing the number of pipe stages due to lower logic depths has impacts beyond branches: if you have too few logic gates at each stage then you don't have time to rename, or time to forward data, al of which will have impacts all around the place.

 

[bjt2]

Increasing the number of pipe stages due to lower logic depths has impacts beyond branches: if you have too few logic gates at each stage then you don't have time to rename, or time to forward data, al of which will have impacts all around the place.

Obvously there is a limit. I don't think that you can go much lower than BD's FO4.
But with it's low FO4 BD anyway manages to have 1 4way int scheduler and 1 4 way fp scheduler.
Zen went to 6 1way schedulers for integer pipeline and a comparable scheduler for FP.
Why they do that? This can impair the IPC. 6 1 way scheduler have less transistors than a 6 way scheduler and more importantly a low FO4. Lower than a 4 way scheduler, obviously. Why lower the FO4? The only reason is that the rest of the design has lower FO4...

 

[Toettoetdaan]

The whole line up gets better clocks,this would have happened regardless.

Yes that is the point, Intel is confident in what they have. Wouldn't they do more if they were afraid of Zen? Like a 120watt version or something.

We just don't know,and never will,if this happened now because of ZEN or if it was time for it to happen anyway.

Unfortunately, that is true

 

[LTC8K6]

It would be scary for the competition if Intel could react quickly to Zen, if necessary.

My guess is that what Intel is releasing has nothing to do with Zen. It all was already coming down the pipeline.

 

[cytg111]

It would be scary for the competition if Intel could react quickly to Zen, if necessary.

My guess is that what Intel is releasing has nothing to do with Zen. It all was already coming down the pipeline.

Coffeelake vs. Zen+ ? I can see that being related.
Outside of watchdogs2(anomaly?) I cant see where software should go to evolve beyond the quads we have today.. *shrugs*

 

[frozentundra123456]

The highest end Summit Ridge (Zen) 8c/16t overclockable cpu will not be a $400 chip. especially since the 6950 is @$1000 and AMD was touting the ZEN as matching a Broadwell E 10 core, a $1,700 chip, in one test. My guess is at least $699 if not $799. For goodness sake the 9590 had a very large release price.

I really hope that the AMD event on the 13th gives some detail on the specs of the ZEN they are running, especially when they show off gaming. I would really like to see the comparison with an AMD 9570?, Intel 6700k and Intel 6950.

Yea, and I still remember the AMD PR for Bulldozer showing it to be as fast as a thousand dollar EE intel chip in gaming. Of course, that was because they used a gpu limited scenario. So I dont really put much credence to the Blender test. We dont really know the limitations of the test, or how much clockspeed Zen had left on the table. We do know that they had to underclock BW-E to get the results they wanted, when in the real world, nearly every HEDT chip will be highly overclocked.

I do agree with you on the pricing. There is a lot of hype, self-delusion, wishful thinking, whatever from AMD fans (not you, obviously, but the usual suspects) thinking Zen is going to provide eight core intel performance at half or less of the price. If they can match BW-E 8 core, then they have a lot of leeway to sell cheaper and still make a nice profit. Even in that best case scenario though, I dont see them selling it for less than 7 or 8 hundred dollars. More realistic to me is that in a wide range of uses it will compete more closely with hex core hedt, perhaps winning most multi-threaded tests and losing in single threaded. AMD is lucky that BW-E was basically a flop for Intel, except for the availability of the 10core chip, so that gives them an easier target. However, not too long after Zen becomes available, they will have to compete with Skylake HEDT, which I expect to bring both IPC gains and much better overclocking than BW-E.

 

[tential]

The highest end Summit Ridge (Zen) 8c/16t overclockable cpu will not be a $400 chip. especially since the 6950 is @$1000 and AMD was touting the ZEN as matching a Broadwell E 10 core, a $1,700 chip, in one test. My guess is at least $699 if not $799. For goodness sake the 9590 had a very large release price.

I really hope that the AMD event on the 13th gives some detail on the specs of the ZEN they are running, especially when they show off gaming. I would really like to see the comparison with an AMD 9570?, Intel 6700k and Intel 6950.

The main benefit of Zen is that it can be a lower price per core option for high end users who want lots of threads. I really don't think it will be anything spectacular at gaming at all. I think it will be good for people who want cheaper highly threaded CPU options.
For me, that means using it in my UNRAID box.
I imagine there will be gamers willing to take a single threaded perf loss for more cores as well.

I just don't think we should expect anything more from Zen other than more cores/threads per dollar than intel with less single threaded perf per dollar compared to intel.

If there is a chip close to $400 it's a 6 core option. If I buy it will be the 8 core option. I want to see the perf, but I may wait til Zen+. No rush to upgrade my server personally, but I believe that's where it will shine. People who REALLY want threads and don't want to pay the huge costs it takes to get them from intel.

 

[Abwx]

2) higher transconductance (AFAIK > +20%)

Other numbers are right but for transconductance the improvement is 52.5% according to the dynamic
characteristics of the two ARM CPUs used as test vehicles at 28nm HPP and 14nm LPP.

 

[blublub]

Bristolridge in OC reach 4.9GHz at 1.43V on the 28nm BULK with HDL libraries.
On hardocp we are told that Kabylake OC'ed at 4.8GHz with 1.3V (some specimens arrive also at 5.1GHz with 1.37V), but with an HP 14nm FF plus probabily HPL libraries...

This is only possible if Bristolridge has a very low FO4.

Zen should have similar FO4, but has a new process that has:
1) lower Vth (i was told that the difference is negligible, but a bit it's better than none)
2) higher transconductance (AFAIK > +20%)
3) lower parasitic capacitance (AFAIK > -20%)
4) lower leakage (AFAIK ~1/6)

This, at same FO4, translated, at 4.9GHz, at least -200mV of Vcore, plus lower power consumption. And we can also lie in stable conditions... Anyway probabily 4.5GHz (at least of single core max turbo) are easily feasible even for an 8 core Zen.

If that would be the case why are all rumors pointing to an All core base clock of 3.2 or 3.3ghz + yes I know TDP is 95w, but for the sake of performance and kicking Intel's but one could release the a black edition wit 4.0ghz all core base clock - since there are no hints pointing that way I doubt Zen will OC higher than 3.5-3.6ghz regardless how high the TDP will be.

 

[Arachnotronic]

And do you think that a Zen core can not arrive at least at 4.9GHz of turbo core on one core?

4.9GHz on air/all-in-one cooler? Or on LN2 for a suicide run?

 

[guskline]

tundra123456, excellent points. I really hope the top end 8 core ZEN puts pressure on Intel to reduce the price of 8 core and even 10 core chips but that will only happen if it is a good overclocker.

ZEN is 95W vs 140W for my 5960x. My custom water-cooled (EK Supremacy EVO block and 3 480mm rads) 5960x is at stock 3Ghz 8core/16 thread but OC's solidly to 4.4 Ghz 1.34v running Asus Real Bench all day at that clock and voltage with temps never exceeding the high 60s MAX.

I cannot expect a ZEN 8 core that equals the 5960x at stock and is a decent OCer to be much less than $700 no matter how much I wish for it.

 

[guskline]

I really dislike those one core suicide runs UNLESS the cpu is a dual core. If all 8 zen cores can run at 4 Ghz then we are talking.

 

[Arachnotronic]

Yea, and I still remember the AMD PR for Bulldozer showing it to be as fast as a thousand dollar EE intel chip in gaming. Of course, that was because they used a gpu limited scenario. So I dont really put much credence to the Blender test. We dont really know the limitations of the test, or how much clockspeed Zen had left on the table. We do know that they had to underclock BW-E to get the results they wanted, when in the real world, nearly every HEDT chip will be highly overclocked.

I do agree with you on the pricing. There is a lot of hype, self-delusion, wishful thinking, whatever from AMD fans (not you, obviously, but the usual suspects) thinking Zen is going to provide eight core intel performance at half or less of the price. If they can match BW-E 8 core, then they have a lot of leeway to sell cheaper and still make a nice profit. Even in that best case scenario though, I dont see them selling it for less than 7 or 8 hundred dollars. More realistic to me is that in a wide range of uses it will compete more closely with hex core hedt, perhaps winning most multi-threaded tests and losing in single threaded. AMD is lucky that BW-E was basically a flop for Intel, except for the availability of the 10core chip, so that gives them an easier target. However, not too long after Zen becomes available, they will have to compete with Skylake HEDT, which I expect to bring both IPC gains and much better overclocking than BW-E.

Here's what AMD's CFO recently said at an investor conference:

So there was a time in the AMD world where chased share for the sake of share. And you can say, you have 15% to 20% market share from a unit standpoint, that's important. But I think, going back to what we just talked about in the desktop space, that is higher ASP, higher-margin product, higher mix. And from my standpoint, that's more important. So it’s more revenue share and the gross margin dollar that we can get.

"Intel performance for half the price" is wishful thinking.

 

[guskline]

Arachnotronic: Your last post just raised the highest end 8 core ZEN to $799!

 

[bjt2]

Other numbers are right but for transconductance the improvement is 52.5% according to the dynamic
characteristics of the two ARM CPUs used as test vehicles at 28nm HPP and 14nm LPP.

I didn't remember and didn't want to say something exaggerated, but nice to know... Even better...
So the odds to have 4GHz base and 4.8-5.0GHz turbo are increasing...