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

[VirtualLarry]

Xeon-D: 16C/32T @ 1.7 GHz OR 12C/24T @ 2.3 GHz at 65W TDP

What is this doing in a Zen / Summit Ridge benchmark thread? That's thead-crapping.

 

[Sweepr]

What is this doing in a Zen / Summit Ridge benchmark thread? That's thead-crapping.

Did you even read the posts above mine? Someone said that 8C/16T at 2.1 GHz base is the best Intel got at similar/lower TDP, which is wrong (and I corrected).

 

[SlowSpyder]

A 4C/8T version of this with faster final clocks (half the cores, less heat to deal with) undercutting current i7 pricing might not be a bad option.

 

[cbn]

If 8C/16T is clocked 2.8/3.2, then the 4C/8T

I don't care about 8c/16t CPU, I was talking about 4c/8t version, which would fit somewhere between i5 and i7. Intel is now selling Skylake i5 for ~200$ (6500 or 6600) and i7-6700 for ~300$. So ST performance of potential 4c/8t Zen based CPU will be lower than i5, and MT performance would probably be between i5 and i7. Plus, those CPUs won't have IGP, so OEMs need to add ~50$ dGPU, if they won't to sell it to advanced home or business users. It will not be competitive product if it costs more than 250$. And last, but not least, there will be no "Intel Inside" on the box 🙂 So therefore I expect it to be positioned where FX-8350 was at launch, around 200$

For advanced home and business users, I think the 35W BGA APUs would work fine.

Anyone needing a board with 95W VRMs is probably DIY or workstation users and for these people having a dGPU will be typical.

P.S. For the 35W BGA SFF desktops and AIO if AMD went purely quad core I don't think anyone would miss the upgradability (re: unlike the 35W Intel "T" CPUs (Celeron thru i7) there would not much spread in the performance). Then for the budget DIY AM4 desktop that could use whatever harvested dual cores APU were left over with a special BIOS that prevents the CPU from throttling when the iGPU is active.

So for budget AM4:

2C/4T with iGPU
4C/8T, no iGPU
6C/12T, no iGPU
8C/16T, no iGPU

All quad core APUs on BGA, leaving no 4C/8T APU option for AM4. (re: 4C/8T AM4 user will most likely use a dGPU and the best quad core APU dies are best left dedicated to BGA IMO. Another reason to not have the Zen quad core APUs on AM4 is that if the BIOS is tuned to not throttle the CPU while the iGPU is active the TDP will likely exceed 95W for quad core APU SKUs. This, unless the quad core APU's CPU and iGPU is clocked low.

 

[Sweepr]

Desktop / Server Zen Specifications (Rumor)

Frequency wall at 3.2 GHz?

 

[cbn]

Desktop / Server Zen Specifications (Rumor)

Frequency wall at 3.2 GHz?

Maybe they are trying segment the processors?

If that is the case maybe AMD needs to release a new AM4 platform with a higher TDP spec? (like they did with the FX chipset for the AM3+ platform)

Maybe 125W to 140W so the 8C/16T would get higher clocks and not be threatened by a 95W 4C/8T?

 

[superstition]

If that is the case maybe AMD needs to release a new AM4 platform with a higher TDP spec? (like they did with the FX chipset for the AM3+ platform)

If AMD has any hopes of regaining a large portion of the large enthusiast community (not just experts) it's a no-brainer to have two levels, at least, for AM4.

And, those levels should have a simple label, like the Nintendo Seal of Approval (and a simplified 80 Plus). Personally, I would have three levels. Bronze, Silver, and Gold.

Gold would be the equivalent of the Sabertooth board for AM3+ in terms of power delivery, etc.

Instead of having board makers purposefully confuse consumers (like with all the claims that the UD3P is an 8 phase board and with the way some boards with a bunch of phases have low-quality MOSFETs) — the minimum specs for compliance with each of the three levels would ensure that consumers would be able to easily match their Zen CPU's requirements with the board at the store.

 

[inf64]

Desktop / Server Zen Specifications (Rumor)

Frequency wall at 3.2 GHz?

That is a summary of clocks that originated from the leak on this very forum (the guy that was talking about ES core/clocks configuration AMD had thus far).

 

[Arachnotronic]

That is a summary of clocks that originated from the leak on this very forum (the guy that was talking about ES core/clocks configuration AMD had thus far).

Aha, right you are!

 

[The Stilt]

That is a summary of clocks that originated from the leak on this very forum (the guy that was talking about ES core/clocks configuration AMD had thus far).

Which were confirmed by the AoTS leak (at least for 95W AM4 part).

 

[Arachnotronic]

Which were confirmed by the AoTS leak (at least for 95W AM4 part).

Well, the 8C/16T one. We'll see if he was right on the others in due time.

 

[F-Rex]

That is a summary of clocks that originated from the leak on this very forum (the guy that was talking about ES core/clocks configuration AMD had thus far).

Exactly
Same datas that we've got several weeks ago.
So far, no leaks saying 2.8/3.2 GHz will be the max clock for Zen.
We only know that 2,8/3,2 is probably the frequency of Zen ES.

I don't know how long it takes to launch a higher frequency part assuming same stepping, but it's probably less than 4 months.

 

[SAAA]

If the others column are indeed right I'd like to share a little speculation:

Scale down both 24 and 32 core parts to 95W TDP with 8 core and clockspeed should be (using quadratic scaling for power)

2.75GHz*(95W/180W*4)^0.5 = 4 GHz
2.9GHz*(95W/180W*3)^0.5 = 4 GHz

Within a MHz, the two values are a bit too close to not mean anything... for a worst case use cubic scaling for power and the quad core could reach that same 4GHz at only 65W.
I hope the process does get better and they can reach those speeds, it smells too much of artificial limitation.

 

[krumme]

That is a summary of clocks that originated from the leak on this very forum (the guy that was talking about ES core/clocks configuration AMD had thus far).

Where did he say eg 2.750 all cores for a 32c part?? 2900 for 24c part?
Doesnt make sense. Way to high.
I do remember him though saying haswell like ipc sans avx2 parts.

 

[inf64]

Which were confirmed by the AoTS leak (at least for 95W AM4 part).

True but I honestly doubt this is all AMD can get and will launch the retail SKUs at the same clock as the max ES clock. They gotta have 10-15% headroom within the same TDP and unless there is some uarchitecture clock wall there is no reason they shouldn't clock at 3.2/3.7Ghz when they launch.

 

[lefty2]

I was looking at those AotS benchmarks. A 3.00GHz 8 core i7-5960X scores 5700 ( on high setting), while a 4Ghz 4 core i7-6700K scores 6000. That is telling me that clockspeed is more important than number of cores in that particular benchmark. So, the Zen engineering sample has 8 cores, but is only clocked at 2.8Ghz and it scores 5000 - probably due more to the low clock speed.

 

[DrMrLordX]

Early 8 core prototypes (OR-A1) reached up to 3.6GHz at 95W. The fastest retail part at the same TDP operated 400MHz higher, three major and several minor revisions later (OR-B2G). The absolute fastest retail part, regardless of the TDP (125W) operated 600MHz higher than that (4.2GHz).

32nm SOI was a process geared towards high clocks, though. It's hard to compare Bulldozer/Zambezi to Zen/Summit Ridge.

Also Con cores up through Vishera have (generally) been designed for high clockspeed potential.

I was looking at those AotS benchmarks. A 3.00GHz 8 core i7-5960X scores 5700 ( on high setting), while a 4Ghz 4 core i7-6700K scores 6000. That is telling me that clockspeed is more important than number of cores in that particular benchmark. So, the Zen engineering sample has 8 cores, but is only clocked at 2.8Ghz and it scores 5000 - probably due more to the low clock speed.

I would tend to agree. It also puts Summit Ridge pretty close to the performance of a 5960X, which isn't too bad.

 

[jpiniero]

I would tend to agree. It also puts Summit Ridge pretty close to the performance of a 5960X, which isn't too bad.

Except you can't really use the regular framerate as a guide since the 480 is mostly GPU limited, even on High. 3930K for instance gets roughly 5600-6100 in general; While the 5960X gets into the 9000s and the 6700K gets into the high 8000s. Even the regular 6500 tends to get about 5800-6000. Granted you don't know how high those unlocked processors are overclocked....

The 9590 gets roughly 5000-5400 btw.

 

[DrMrLordX]

Guess we'll have to wait until someone does more extensive benchmarking before we can draw any real conclusions, then?

 

[Azuma Hazuki]

Guess we'll have to wait until someone does more extensive benchmarking before we can draw any real conclusions, then?

Indeed we do. Or that's the rational thing anyway. For certain values of "we" (notably Phynaz and Arachnatronic...) bashing AMD seems to take priority. I can't imagine why; what do they think their precious Intel chips would cost without AMD around?

 

[Insert_Nickname]

Did you even read the posts above mine? Someone said that 8C/16T at 2.1 GHz base is the best Intel got at similar/lower TDP, which is wrong (and I corrected).

I mentioned the E5-2620v4 because its the most directly comparable socketed Intel CPU to this ES sample. That'd be 8C/16T in a socketed form factor @ around 90W. Yes, Intel has better. But the Xeon-D is BGA only. Remaining with socketed CPUs, one could also mention the 55W 10 core 2630Lv4. But its still only 1.8/2.9GHz. Or the 65W 14 core 2650Lv4 @ 1.7/2.5GHz

 

[The Stilt]

If the others column are indeed right I'd like to share a little speculation:

Scale down both 24 and 32 core parts to 95W TDP with 8 core and clockspeed should be (using quadratic scaling for power)

2.75GHz*(95W/180W*4)^0.5 = 4 GHz
2.9GHz*(95W/180W*3)^0.5 = 4 GHz

Within a MHz, the two values are a bit too close to not mean anything... for a worst case use cubic scaling for power and the quad core could reach that same 4GHz at only 65W.
I hope the process does get better and they can reach those speeds, it smells too much of artificial limitation.

Based on a rumor I heard, the 32C/64T part has base clock of around 1.5GHz. Those turbo frequencies (single core) should be accurate, based on the same rumor.

True but I honestly doubt this is all AMD can get and will launch the retail SKUs at the same clock as the max ES clock. They gotta have 10-15% headroom within the same TDP and unless there is some uarchitecture clock wall there is no reason they shouldn't clock at 3.2/3.7Ghz when they launch.

Based on what?
The partners made their AM4 motherboards using Bristol Ridge SKUs, meaning they didn't have the need to have access to extremely early prototypes. Since the leaks obviously originate from China (AotS), it means that AMD has now shipped Zeppelins to the motherboard / system manufacturers. And as I said, they didn't and still have no need to do that unless it is a silicon representing the final product. There will definitely be some improvements, but I would expect them to come purely from the maturing process.

 

[cytg111]

Haswell ES was 2.8 I believe ..(reason i pick haswell is only cause it is the only chip i remember had extensive ES leaks before launch).. and some of those suckers OC'd to 5GHz+

 

[blublub]

Has anyone an idea what 1D/2D stands for? - could it mean "deactivated", because that would explain why D1 is faster than D2 (more features deactivated), if true it would imply a D0 which could be faster... Just guessing here

 

[The Stilt]

The first character has separated differently designated (internal / external use) samples in the past. "Z" starting parts generally represent the final, fully validated silicon. No idea if that's still the case with Zeppelin, but most likely it is.