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

[realibrad]

He was speaking about quad cores and 45W, not 8 cores. And he was also saying 4,5 GHz turbo, not base. And he was the first and one to say Zen was an high frequency product when all others spoke doom&gloom with ES at 2,8GHz. Not that everything he (or you, for what's worth) says has or will meet the reality.
And I don't know what's more hilarious, his optimism or your continous AMD bashing.

I think he was saying that Ryzen would use 45W less, not that the chip would be 45W. 140W for Intel vs 95W for AMD would be a difference of 45W.

 

[bjt2]

Chill with the insults...

Your hype was delirious so I couldn't stop laughing.

200MHz more, equal perf at -45W and 50% price.

Then there's the pages about 4.5GHz 95W...

I don't know haha.

Sent from HTC 10
(Opinions are own)

Who was saying that for sure? I was talking of last ES.
Is it true or not that there are ES with 3.6GHz base clock?
Is it true or not that at lower clock Zen beats 6900K in blender and handbrake?
Is it true or not that 3.6 is more than 10% over 3.2?
The only unknown is mean Zen IPC versus BWE.
I said that if Zen IPC is over -10% BWE (and probabily is ) the clock difference will compensate.
What is the problem with these statements?

 

[bjt2]

I think he was saying that Ryzen would use 45W less, not that the chip would be 45W. 140W for Intel vs 95W for AMD would be a difference of 45W.

Yes, but 4.5GHz single core turbo was also one of my forecasts. It's not impossible, since FX8370E, a 95W eight core of many years ago and on the old 32nm SOI, reaches 4.3Ghz in single core turbo, with a 3.3Ghz base clock. All points to 3.6GHz base clock for Ryzen, so i don't see why ryzen could not reach at least 4.3GHz single core turbo as the 8370E, and mabye 4.5-4.6... All this leaving out XFR...

 

[leoneazzurro]

I think he was saying that Ryzen would use 45W less, not that the chip would be 45W. 140W for Intel vs 95W for AMD would be a difference of 45W.

I see, yes, he was talking about -45W on TDP, and that's right, but you have to consider anyway that TDP may be calculated/measured differently and actual power draw of BW-E is generally lower than those 140W.

 

[bjt2]

Meanwhile AMD stocks skyrocket by 7.5%...

 

[realibrad]

I see, yes, he was talking about -45W on TDP, and that's right, but you have to consider anyway that TDP may be calculated/measured differently and actual power draw of BW-E is generally lower than those 140W.

I was just giving insight as to what he meant, and was not commenting beyond that. But...

By that logic, Ryzen will typically use less than 95W. A 6900k with all cores going can easily use 140W. That 140W is quite reasonable as a TDP.

 

[Zucker2k]

I was just giving insight as to what he meant, and was not commenting beyond that. But...

By that logic, Ryzen will typically use less than 95W. A 6900k with all cores going can easily use 140W. That 140W is quite reasonable as a TDP.

So it'll barely hit 140w under torture, but it magically hit 140w under blender and handbrake tests? Hmmm.....

 

[The Stilt]

I was just giving insight as to what he meant, and was not commenting beyond that. But...

By that logic, Ryzen will typically use less than 95W. A 6900k with all cores going can easily use 140W. That 140W is quite reasonable as a TDP.

TDP doesn't account for external VRM losses, only internal (in case of FIVR or similar...) obviously.

http://www.hardware.fr/articles/946-4/overclocking-consommation.html

During non-AVX/2 Prime95 Hardware.fr measured:

6950X = 117.6W
6900K = 124.8W
6800K = 99.6W

These are external VRM input powers (measured from EPS12V).
Since Broadwell-E has FIVR the efficiency of the external (i.e motherboard) VRM will be slightly higher than usual. Probably around ~86% at this kind of currents.
6950X = ~101.136W, 6900K = ~107.328W, 6800K = ~85.656W~.

 

[realibrad]

So it'll barely hit 140w under torture, but it magically hit 140w under blender and handbrake tests? Hmmm.....

CPUs can throttle fyi. So when it reached that power usage at stock clocks, it could throttle. We have seen this type of thing before. Its not magic.

TDP is supposed to be the max at stock, so considering that the AMD chip has 95W, its not unreasonable to think that it would only hit that in specific situations just like how Intel hits its TDP in specific situations.

 

[realibrad]

TDP doesn't account for external VRM losses, only internal (in case of FIVR or similar...) obviously.

http://www.hardware.fr/articles/946-4/overclocking-consommation.html

During non-AVX/2 Prime95 Hardware.fr measured:

6950X = 117.6W
6900K = 124.8W
6800K = 99.6W

These are external VRM input powers (measured from EPS12V).
Since Broadwell-E has FIVR the efficiency of the external (i.e motherboard) VRM will be slightly higher than usual. Probably around ~86% at this kind of currents.
6950X = ~101.136W, 6900K = ~107.328W, 6800K = ~85.656W~.

It would be the same with an AMD system would it not? VRM efficiency should be a thing on AMD boards as well, and as such should be fairly equal right?

 

[The Stilt]

It would be the same with an AMD system would it not? VRM efficiency should be a thing on AMD boards as well, and as such should be fairly equal right?

The VRM efficiency on AMD (and Intel non-FIVR) boards is somewhat lower (78-83%) in general due the significantly lower output voltage (duty cycle) and higher currents. For example Skylakes run =< 1.2V (AFAIK), while Haswell-E or Broadwell-E have input voltage of >= 1.8V.
Even with the difference in the power draw, the currents are significantly higher on Skylake.

But indeed, the advertized 95W TDP for Ryzen means that at stock the whole package power (PPT) is ceiled to this figure.
When measured from the EPS12V connector (VRM input power) at full load and a proper motherboard, the seen figure should be around < 115W.

 

[Karnak]

https://twitter.com/CPCHardware/status/828035457502048256

Intel freaking out about Ryzen?

Intel prépare la riposte à Ryzen : Core i7 7740K et i5 7640K (avec HT on) - Un soucis avec les Quad-Core Ryzen ?

https://twitter.com/CPCHardware/status/828640373949997057

Article: https://www.cpchardware.com/intel-prepare-la-riposte-a-ryzen/

 

[lolfail9001]

Intel's freak out about Zen leading to Intel releasing a minor refresh of Kaby Lake line-up? I can barely contain my laughter about posts yesterday.

The article does contain a hint: watch out for AMD only sampling the fastest SKU. If they do not, great. If they do, not so great.

 

[KTE]

Intel prépare la riposte à Ryzen : Core i7 7740K et i5 7640K (avec HT on) - Un soucis avec les Quad-Core Ryzen ?

https://twitter.com/CPCHardware/status/828640373949997057

Article: https://www.cpchardware.com/intel-prepare-la-riposte-a-ryzen/

With minimal increase, it means they're not one bit threatened.

Not really good news for Ryzen.

I'm certainly expecting them to know within 1-5% of how Ryzen performs by now..

Sent from HTC 10
(Opinions are own)

 

[Karnak]

With minimal increase, it means they're not one bit threatened.

Not really good news for Ryzen.

Sent from HTC 10
(Opinions are own)

I think HT for an i5 would be more than a minimal increase.

 

[bjt2]

With minimal increase, it means they're not one bit threatened.

Not really good news for Ryzen.

I'm certainly expecting them to know within 1-5% of how Ryzen performs by now..

Sent from HTC 10
(Opinions are own)

There are probabily platform and VRM limits on 1151 MB... So this is the maximum allowed by the safety margins... INTEL does not want a CPU that does not work on some MB...

 

[Zucker2k]

I think HT for an i5 would be more than a minimal increase.

The consumers' gain, or?

 

[Atari2600]

Please bear in mind that if the KL CPUs are currently running up to motherboard socket power limits*, Intel are somewhat restricted in what they can do.

https://ark.intel.com/products/97129/Intel-Core-i7-7700K-Processor-8M-Cache-up-to-4_50-GHz

http://www.anandtech.com/show/10968...-review-the-new-stock-performance-champion/11

If the warranty limit for the socket is 125W, then Intel can clock up to 4.8GHz.

If the warranty limit for the socket is 100W, then that becomes ~4.65GHz.

If the warranty limit is 95W, then its 4.6GHz.

The scope for Intel to improve KL clocks may not be so simple as one would envisage.


*also note that the limit here refers to manufacturers warranty limits. Which of course is not (in 99.999% of cases) equal to the actual limit the motherboard will carry or sustain.

 

[Glo.]

With minimal increase, it means they're not one bit threatened.

Not really good news for Ryzen.

I'm certainly expecting them to know within 1-5% of how Ryzen performs by now..

Sent from HTC 10
(Opinions are own)

May I ask you a question. Why are you not able to draw any sort of conclusions based on what was already presented?

IPC for Ryzen is on Broadwell-Haswell Level. 4C/8T CPU will clock for clock behave on this level. Is it a performance threat to 4C/8T CPUs from Intel? No. Is it a value threat for Intel CPUs? Hell yes!

The bigger problem Intel has is 6 and 8 core CPU, with Haswell/Broadwell level of performance, per clock. Because according to rumors, that CPU can cost 400$, with 3.4 GHz base clock, and 3.7 GHz turbo. What this means? That you will get twice the amount of cores, for similar amount of money, that are clock, for clock within 5% of latest Kaby Lake architecture from Intel, and use similar amount of power.

6 core CPU, with 12 threads would be in 250-300$ price range completely unmatched, for Intel.

Guys, look at the leaves are falling. Intel was right, that Zen is not a threat in performance. But its extreme threat from value perspective. No matter what happens, with Skylake, and how fast it is. If AMD prices even one 8 core/16T CPU at 399$, Intel cannot charge more than 450-499$ for Any of their own 8 Core CPUs, that are currently on market, or will be in future.

 

[SAAA]

I think HT for an i5 would be more than a minimal increase.

Indeed, if price range stays the same too we are looking at literally 100$+ slash for unlocked 4/8 cpus. Zen quads would compete against similar priced i7s now called i5 and the actual i7 get just little more cache and clock boost/better binning.

It was pretty obvious after HT enabled on Pentiums that next we'd see something similar, maybe they'll release Kabylake i3 with 4core/4thread later this year too, so the whole "i" line would sit at 4 cores.
Considering HT was disabled only for segmentation reasons (a defect in the actual die is likely 0%) turning it on costs nothing design wise, revenue might get impacted for a while but consider next year there will be 6 cores mainstream at i7 prices again and that could make a lot of users upgrade.

 

[LTC8K6]

First we find the Core i7 7740K. It has 8 MB of L3, with a base frequency of 4.30 GHz (100 MHz more than the Core i7 7700K) at the price of a significantly higher TDP (> 100W against 91W). The Turbo frequency is not known, but should logically be 4.60 GHz.

One of the other sample takes the name of Core i5 7640K. At first glance, this is also a speed bump of the Core i5 7600K: Quad Core, 6MB L3, 4.0GHz base frequency for the 7640K against 3.8GHz for the 7600K. On the TDP side, it would also climb above the 100W. But there is much more interesting: the Core i5 7640K would have active SMT (Hyper-threading), a first for a Core i5 Desktop. This modification profoundly upsets the segmentation of the different "Core". Until now, the Core i5 were distinguished from the Core i7 by the absence of Hyper-threading.

Google translation...

 

[LTC8K6]

Seems like the performance of the 7700 and the 7640K would be similar.

 

[LTC8K6]

Why would the chips with fewer cores clock lower?

As we wrote in our article, the frequency of Ryzen Quad Core should be between 3.8 GHz and 4.2 GHz base to compete effectively with the Intel range. But we are far from it. The current quasi-final samples are limited to 3.2 GHz in base frequency. This is very insufficient and especially surprising since the two 8-core models reach 3.4 / 3.8 and 3.6 / 4.0 GHz.

Why a frequency so low on the Quad Core while the Octo Core go much higher? One can first imagine that a Turbo mode particularly swift will come to save the furniture, but currently, we have no information in this sense. For now, we only have access to the 6- and 8-core versions of the Ryzen in stepping B and we do not know the Turbo frequencies of the Quad Core B-Step. We can, however, risk a comparison. Between stepping A0 and B, the turbo mode of 8-core models increased by 400 MHz at best from 3.6 to 4 GHz. In stepping A0, Quad Core were clocked at best at 3.4 GHz (for 2.9 GHz base). In stepping B, the Quad Core Ryzen should therefore be at 3.8 GHz. It's still far too little to compete with the next Core i5 7640K for example. This situation seems all the more strange since AMD seems determined to sell Ryzen with 8 cores clocked faster, which proves that the architecture can reach more frequencies. So what?

 

[Zucker2k]

What about the gist on quadcore ryzen debuting at 3.2ghz? What the actual......, if true, that is.

 

[SAAA]

Please bear in mind that if the KL CPUs are currently running up to motherboard socket power limits*, Intel are somewhat restricted in what they can do.

Well thanks for the info, that makes Kaby Lake X much more interesting considered it will use X299 boards that should have a greater power headroom... 125W enough for a 4.8GHz stock chip? Nice little beast even if it's still a quad!

Besides I still don't buy the rumor that 6 core Zen could reach as low as 250$ (if it even exists and it's not just quad and eight cores): it's way too much performance/$ even for an AMD part, where would they fit simple quad cores, all APUs skus and so on then? If extra 100-50$ buy you 50% more cores there's not much to think, unless the quads clock much higher...