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

[looncraz]

Well, don't we sort of know that Zen's die is quite a bit larger than just 2 CCXs?

My bet is on 192mm^2

Making a statements based on the binned top tier rx480 does not help your cause, though. What we really know is that at least first batches of rx480 had ridiculous variance, ranging from bearable to madly throttling.

The GTR is not top-binned, it's basically stock. My sample, in fact, isn't very good compared to what many others get - particularly with newer ones (the process has improved).

XFX RX 480 Black looks the same, but is higher binned, and the 'GTR' or 'Black' is often missing from the descriptions online, so the confusion is easily understandable.

Should fit about 10CUs and 8 RoPs... In line with last RR rumors, too.

Yes, but memory bandwidth is that by which graphics performance lives or dies. 50GB/s would be a major step up, but 10CUs at 1Ghz will saturate that with a thirst for much more.

Polaris 10 has 36 CUs and 256GB/s bandwidth - that's 7.11GB/s of bandwidth per CU. Polaris 10 is VERY bandwidth sensitive (though Vega may be less so given its new management and cache layout).

Assuming Vega is less sensitive, such that 7GB/s is enough to prevent a performance loss greater than the addition of a new CU, 8 CUs would be the sweet spot (albeit above the curve). Of course, if they are going to clock the thing at 850Mhz or so... 10 CUs would make sense.

Then again, AMD usually provides more CUs than the memory can handle on the APUs...

 

[CatMerc]

Well, don't we sort of know that Zen's die is quite a bit larger than just 2 CCXs?

Making a statements based on the binned top tier rx480 does not help your cause, though. What we really know is that at least first batches of rx480 had ridiculous variance, ranging from bearable to madly throttling.

Okay, that is the viable explanation, i am dumping my pessimist version.

Should fit about 10CUs and 8 RoPs... In line with last RR rumors, too.

I made this a while ago by frankensteining a Polaris 10 die shot onto a Zen die shot lol
11 CU's and 8 ROPS as you said.

Spoiler: Large Image Ahead

 

[Veradun]

Exactly. I have currently no need for an 8-core cpu. Main thing I do is browsing and gaming. High clocked quad still rules in this aspect. I agree that 7700k is kind of a disappointment. Still, without to much tinkering (eg. delid) a 4.8Ghz OC should be possible easily. Can the 6 or 8-core Zen get close to 7700k 4.8Ghz ***gaming*** performance?

Fixed for you. You don't actually need ST performance, as there is no current ST game

 

[formulav8]

Some extra/updated cpu, mobo, skus and such.

 

[Glo.]

I made this a while ago by frankensteining a Polaris 10 die shot onto a Zen die shot lol
11 CU's and 8 ROPS as you said.

Spoiler: Large Image Ahead

11CU design is most likely cut down from 12 CUs, MOBILE APU.
There are two designs for Raven Ridge NPUs. Smaller and bigger. Smaller one is Mobile design, without HBM2, bigger one is desktop design with HBM2.

You will ask why. HBM2 is important for Professional and HPC markets. Also consumer high-end NPU, will get huge traction.

 

[CatMerc]

11CU design is most likely cut down from 12 CUs, MOBILE APU.
There are two designs for Raven Ridge NPUs. Smaller and bigger. Smaller one is Mobile design, without HBM2, bigger one is desktop design with HBM2.

You will ask why. HBM2 is important for Professional and HPC markets. Also consumer high-end NPU, will get huge traction.

I was under the assumption Raven Ridge has just one variant, and it's Horned Owl/Banded Kestrel that will come as two separate variants.

 

[blublub]

Is there some conclusion now what the standard Turbo acutally means? - I mean is it summoned t SCT or ACT

 

[jpiniero]

Is there some conclusion now what the standard Turbo acutally means? - I mean is it summoned t SCT or ACT

Full core Turbo, albeit not guaranteed (although you could OC to keep it at FCT of course)

 

[lolfail9001]

Is there some conclusion now what the standard Turbo acutally means? - I mean is it summoned t SCT or ACT

You mean, the 4.0 in case of 1800X? SCT.

 

[formulav8]

That looks to be true, 4ghz = 1 core only. Also expected.

 

[CentroX]

Here is a question for you all. Has the benchmark leaks been controlled by amd or have they been independent? Why havent there been any gaming benchmark leaks.

 

[coercitiv]

Here is a question for you all. Has the benchmark leaks been controlled by amd or have they been independent? Why havent there been any gaming benchmark leaks.

You mean... like the gaming benchmark CPC Hardware did on the Zen ES?

 

[CentroX]

You mean... like the gaming benchmark CPC Hardware did on the Zen ES?

I missed that one. Link?

 

[formulav8]

Here is a question for you all. Has the benchmark leaks been controlled by amd or have they been independent? Why havent there been any gaming benchmark leaks.

The leaks look nice when it comes to Intel's E line. I personally don't know all of the specifics, but for myself only, AMD is so far looking to provide a much better bang for the buck than Intel has in years. So I personally do not care if AMD fully topples Intel. I care more about what I can get for the price.

There looks to be others expecting a KL toppling miracle. I do Not expect nor care that much if they reach that feat.

 

[IllogicalGlory]

I missed that one. Link?

It was one of the first, and it's an aggregate score using somewhat unusual games to test, but here it is:

I think the reason that there haven't been others is that it's not that easy to start up a game, find a good testing location, test the same place with other CPUs and get reliable data. It's much easier to run passmark, firestrike physics, cinebench or CPU-Z, as the scores are easily comparable to other systems, and they have the tools on hand.

 

[lolfail9001]

Here is a question for you all. Has the benchmark leaks been controlled by amd or have they been independent? Why havent there been any gaming benchmark leaks.

Almost positive the first Passmark leak was not controlled by AMD. Second Passmark leak was probably not controlled as well, even though it uses AMD's ref board. Cinebenches are wide open.
What we are left with is why there is no gaming leaks and only canned benches posted? Well, because gaming leaks require videos.

 

[Glo.]

I was under the assumption Raven Ridge has just one variant, and it's Horned Owl/Banded Kestrel that will come as two separate variants.

http://www.bitsandchips.it/english/...two-versions-of-raven-ridge-under-development

Yes, they can be wrong. But 16 CU+HBM2 is more believable than 24 CU APU+ HBM2.

 

[Ryzenenthusiast]

http://www.bitsandchips.it/english/...two-versions-of-raven-ridge-under-development

Yes, they can be wrong. But 16 CU+HBM2 is more believable than 24 CU APU+ HBM2.

16 Vega CU+HBM2 is excellent enough for an APU. People who wants 24CU can buy an RX 470. It is out of the range of APU's. 7nm APU's may have 24 CU's+HBM2

 

[Arachnotronic]

16 Vega CU+HBM2 is excellent enough for an APU. People who wants 24CU can buy an RX 470. It is out of the range of APU's. 7nm APU's may have 24 CU's+HBM2

HBM2 on a mainstream APU isn't happening anytime soon.

 

[Ryzenenthusiast]

HBM2 on a mainstream APU isn't happening anytime soon.

It is actually happening according to Fottemberg(Bitsandchips guy). I am talking solely based on his rumor...

 

[looncraz]

It is actually happening according to Fottemberg(Bitsandchips guy). I am talking solely based on his rumor...

If it does, I'm going to be building one heck of a beautiful HTPC system for a client (probably around Christmas, so hopefully it's on the market by then).

 

[looncraz]

Is there some conclusion now what the standard Turbo acutally means? - I mean is it summoned t SCT or ACT

There's still vast disagreement.

I don't believe there is a set-in-stone single-core turbo.. or all all core turbo. That would be really strange way to use the capabilities of the chip. I think it will be TDP and temperature based exclusively and that it just happens to be that the top turbo frequency is the frequency you are only likely to hit on a single, fully loaded, core using the stock cooler provided with the CPU prior to heat-soak.

I think better cooling, even without XFR or overclocking, will allow you to hit higher turbos on more cores. Keeping the CPU at ambient temperatures may enable achieving new maximum turbo on four+ cores, if that's the case... we will see

 

[JDG1980]

Making a statements based on the binned top tier rx480 does not help your cause, though. What we really know is that at least first batches of rx480 had ridiculous variance, ranging from bearable to madly throttling.

This is true. However, keep in mind that Polaris was basically a pipe cleaner for the GloFo 14LPP process. There was a lot of variance in the first RX 480 cards because GloFo still was working out the kinks. AMD decided it was better to launch then with what they had (since there wasn't much else new for 2016) than to wait for slightly better yields and improved perf/watt.

This shouldn't be an issue with Zen, since both AMD and GloFo now have much more experience with 14LPP. Not only that, but what we've heard so far indicates that Samsung will also be fabricating Zen as well. This means if GloFo is still not on par, we might wind up seeing Samsung chips for the top (best binned) SKUs and GloFo for the lower ones. If nothing else, Zen's release should give us an opportunity to analyze GloFo vs. Samsung on 14LPP, which hasn't really been done yet in an "apples-to-apples" comparison.

 

[lolfail9001]

Not only that, but what we've heard so far indicates that Samsung will also be fabricating Zen as well.

Never seen any conclusive evidence on this one.

 

[Doom2pro]

Never seen any conclusive evidence on this one.

AMD did say they ran product at Samsung and that they can use Samsung to increase production if necessary.

Source: http://www.forbes.com/forbes/welcom...-diversifies-14nm-manufacturing-with-samsung/