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

[The Stilt]

I don't think the actual TDP is the main reason for the low clocks. A vast majority of the potential Zeppelin purchasers would still buy it despite the higher (e.g 125W) TDP, if there was large performance improvement over a 95W TDP model. I think the reason for sticking with the original 95W TDP is that beyond that point the process operates so far beyond it's ideal range, that the actual frequency improvements from the increased power budget are negligible. This brings us back to Polaris 10 (same most likely applies on P11) where the effect can be seen starting from the 1GHz / 860mV mark. I'm pretty certain AMD didn't originally plan to run Polaris at the clocks and voltages they ended up releasing them at (up to 1266MHz / 1.15V). Most likely the launch of Pascals took them by complete surprise and were forced to react in the only way they at that point could. Maybe it is a pure coincidence, but if you look at the clocks and the voltages the ES Polaris cards were running at during the CES 2016 demo, my scenario might not be so far fetched

@ 2:25

 

[AtenRa]

Keeping the same clocks but raising TDP from 95W to 140W will increase performance simple because the CPU/GPU or any IC will operate at the fmax longer and will throttle less.
So, higher TDP is not always correlating to higher clocks only.

Also, comparing a different mArchitecture (Core i7 6800K) on a different process vs ZEN for clocks/TDP is pointless.

 

[Abwx]

This brings us back to Polaris 10

You are aware that a GPU is not the same thing as a CPU..?.

How do you explain that GCN can clock at 1.2GHz within Kaveri while the CPU part reach 4.3GHz using the same transistors, so what is your comparison actually about and are Polaris frequencies of any help to know about Zen..?..

That is, explain us the methodology with sound technical arguments and how your curve could be related to Zen...

 

[SAAA]

6900K from Intel runs at 3.2ghz base and 3.7ghz in a 140w tdp and has far less integrated on the die manufactured on a superior process. Your expectations seem very unreasonable IMO.

Well techincally there's also this:
http://ark.intel.com/products/92992/Intel-Xeon-Processor-E5-1680-v4-20M-Cache-3_40-GHz

Oh and I just discovered the 6900k is listed too as 4GHz turbo on ark, probably a turbo 3.0 feature.
Xeons are definitely better binned, but that miserable increase in clocks we got when going from 22 to 14nm at the same power tells all: the TDP rating is really just a number at this point.

I don't know if the same thing will happen with Zen but given the extremely feeble nature of this rating they might as well push clocks a bit more on good chips and keep all the duds as 6 cores to stay whitin that 95W target. Much easier that way than:
1) increasing TDP, that would hurt both current expectations and motherboard design rules;
2) decreasing clocks on all processors, when some might hit much better speeds from large process variations.

So yeah I'd wait on release and see what they decide, but I hope they push the best model as high as possible at cost of reduced inventory. If they can reach that magical 4.0 number even for single core turbo alone most benchmarks will show all the Zen advantages over old Piledriver, overclock will do the rest rightfully ignoring power tribulations.

 

[The Stilt]

The 14nm Intel HEDT parts actually consume quite significatly less than their TDP.
Hardware.fr measure following values during Prime95 (pre VRM):

- i7-6950X = 117.6W
- i7-6900K = 124.8W
- i7-6800K = 99.6W

With expected motherboard VRM efficiency of 85% and FIVR efficiency of 80% the actual power consumed by the CPU is ~80W (6950X), ~85W (6900K) and ~68W (6800K).

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

 

[Abwx]

The 14nm Intel HEDT parts actually consume quite significatly less than their TDP.
Hardware.fr measure following values during Prime95 (pre VRM):

- i7-6950X = 117.6W
- i7-6900K = 124.8W
- i7-6800K = 99.6W

With expected motherboard VRM efficiency of 85% and FIVR efficiency of 80% the actual power consumed by the CPU is ~80W (6950X), ~85W (6900K) and ~68W (6800K).

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

LOL....

The samples sent to the press had basicaly 1% voltage margin for the one sent to HFR while Computerbase.de sample has 6.8%, with normal values thoses TDP figures would be at least 20% higher, no wonder that contrary to their habits HFR didnt test the undervolting, but Computerbase.de did, although not with Prime..

https://www.computerbase.de/2016-05/intel-core-i7-6950x-6800k-test/9/

Not sure that AMD ES used so unprofessional specs, you think they did so with the OEMs..?.

 

[krumme]

You are aware that a GPU is not the same thing as a CPU..?.

How do you explain that GCN can clock at 1.2GHz within Kaveri while the CPU part reach 4.3GHz using the same transistors, so what is your comparison actually about and are Polaris frequencies of any help to know about Zen..?..

That is, explain us the methodology with sound technical arguments and how your curve could be related to Zen...

Look at the curves again as its fantastic information. Its all gpu. Yet the polaris curve is very different from the others in that it stays flat all the way to 900MHz then starts to climb steeply. Its highly probably a process charateristics.

Therefore its reasonable to - asume - the similar behavior can be applied to zen. And that zen will benefit of not going outside the process target area as the cost in efficiency is higher than what we normal experience.

And therefore i also think we can anticipate quite lean mobile gpu from polaris. Nowhere near pascal levels but relatively far better.

 

[Tuna-Fish]

You are aware that a GPU is not the same thing as a CPU..?.

How do you explain that GCN can clock at 1.2GHz within Kaveri while the CPU part reach 4.3GHz using the same transistors, so what is your comparison actually about and are Polaris frequencies of any help to know about Zen?

GPUs clock lower because their pipeline stages are longer. The clock speed of a device is the length of it's longest pipeline stage in FO4 multiplied by the time a single FO4 takes on that process.

CPUs having shorter pipeline stages just means that the curve is shifted directly right by the difference in pipeline length, it does not modify the shape of the curve.

 

[Abwx]

Look at the curves again as its fantastic information. Its all gpu. Yet the polaris curve is very different from the others in that it stays flat all the way to 900MHz then starts to climb steeply. Its highly probably a process charateristics.

There s nothing unusual, it stay flat because the threshold voltage of the fet has fixed value, so below a given voltage you cant no more switch the transistors, as such the only saving possible is to keep the voltage at this minimum value and reduce frequency.

The knees start at the points where the fets behave normaly, that is, voltage must be increased according to a square law slowly morphing to a cubic law and so on.

That being said the curves posted by The Stilt are relevant only for the pipeline whose behaviour they describe, this has no relevannce for a CPU whatever it is or even for another GPU using a different design.

 

[krumme]

There s nothing unusual, it stay flat because the threshold voltage of the fet has fixed value, so below a given voltage you cant no more switch the transistors, as such the only saving possible is to keep the voltage at this minimum value and reduce frequency.

The knees start at the points where the fets behave normaly, that is, voltage must be increased according to a square law slowly morphing to a cubic law and so on.

That being said the curves posted by The Stilt are relevant only for the pipeline whose behaviour they describe, this has no relevannce for a CPU whatever it is or even for another GPU using a different design.

Well tonga is not flat in its charateristics so that imo contradict what you wrote. Look again.

Its all asumptions but its tied to all the leaks of 2.8 3.2 clocks yet a 180w 32c part. Stilt curves correlate perfectly with that and gives until now the best explanation.

It unfortunately seems to me people dont want to see the low clocks because they hope for something else - all that 8c 4ghz boost nonsense. Ofcource we have the usual 4 amd sucks posters but even in this case it seems to me Stilt and Mathias interpret the clocks as its the worst day this year. I think its better to set expectations lower. Look at amd and gf prior execution and go from there. I try to do and then everything looks fine. Lol.

 

[Abwx]

Well tonga is not flat in its charateristics so that imo contradict what you wrote. Look again.

2.5x the frequency for 10% higher voltage, that s about flat, for such a voltage delta the frequency delta should be 1.21x....

Its all asumptions but its tied to all the leaks of 2.8 3.2 clocks yet a 180w 32c part. Stilt curves correlate perfectly with that and gives until now the best explanation.

Ofcource we have the usual 4 amd sucks posters but even in this case it seems to me Stilt and Mathias interpret the clocks as its the worst day this year. I think its better to set expectations lower. Look at amd and gf prior execution and go from there. I try to do and then everything looks fine. Lol.

As said Kaveri s GCN GPU clock at 1.2GHz while Steamroller run at 4GHz+ with the same process, so why shouldnt Zen be capable to run at comparable frequency given that GCN clock higher with Finfets, whatever the exact flavour..?.

 

[krumme]

Zen looks like a really nice product to me from a business perapective. Who wouldnt want a smallish strong core fit for servers consoles mobile workstations and fanless notebooks?. In the hand of nv this product would be carried to the end of the universe. In the hand of amd and sweepr we start by playing tf2 on the desktop

 

[Sweepr]

Zen looks like a really nice product to me from a business perapective. Who wouldnt want a smallish strong core fit for servers consoles mobile workstations and fanless notebooks?. In the hand of nv this product would be carried to the end of the universe. In the hand of amd and sweepr we start by playing tf2 on the desktop

Not my fault if the first benchmarks do not match your expectations, but I wouldn't describe ~2.8-3.2 GHz cores with the IPC increase that AMD promised as 'strong cores' - and that's if they deliver. I'm sure you will still praise this product and carry it to the end of the universe even if it's a huge letdown, but some of us have been burned by their promises before, so we're more cautious this time.

 

[The Stilt]

Let's look at the bright side: Zeppelin can't be any worse than the products AMD is currently trying to sell

As a side note: Based on my "extensive" (less than two weeks) experience of a rather modern Intel CPU (Haswell-EP), I must admit that AMD has quite superior power management in their most recent designs (read Carrizo and newer). A large part of the Intel's power management of course operates under the radar (PCU), however still I can't say I would be terribly impressed with it. AMD CPUs and APUs have loaned much of the power management related stuff from the GPUs and much of the development is no doubt driven by the deficiencies in other areas (manufacturing process, technology). The power management on Intel appears to still operate based on fused / pre-calculated patterns, much like AMD designs prior to Carrizo. It might use some of the real time measured physical variables to adjust it's operation, but not certainly all of them.

If only RTG and Intel would make babies...

<picture removed>

How did that picture have anything to do with being on topic.
If i see this sort of nonsense in this thread again, you will be given an infraction for drifting off topic

Moderator Aigo

 

[os2wiz]

Looks very slow compared to Skylake. How does Ashes of the Singularity scale over many CPU cores/threads? How fast is an FX-8350 in this benchmark?

Hey, it is an engineering sample, no doubt an early engineering sample, not optimized and the release version will be somewhere between 3.2 to 3.5 GHZ for default frequency. So these benchmarjs are considerably less than the finished package wuill offer. The porimitive motherboard used for the ES sample wuill be refined as well. No way can any serious conclusion be drawn from this benchmark. My expectatiion is that the finished product will exceed a Haswell I7 and be close to a Broadwell I7

 

[Abwx]

Not my fault if the first benchmarks do not match your expectations, but I wouldn't describe ~2.8-3.2 GHz cores with the IPC increase that AMD promised as 'strong cores' - and that's if they deliver. I'm sure you will still praise this product and carry it to the end of the universe even if it's a huge letdown, but some of us have been burned by their promises before, so we're more cautious this time.

Why not a comparison with a 32GB 10 cores 140W 6950X..??

http://www.ashesofthesingularity.co...-details/197f5df8-db37-43ee-a597-ddda5c7956f2

And Zen ES :

 

[The Stilt]

Hey, it is an engineering sample, no doubt an early engineering sample, not optimized and the release version will be somewhere between 3.2 to 3.5 GHZ for default frequency. So these benchmarjs are considerably less than the finished package wuill offer. The porimitive motherboard used for the ES sample wuill be refined as well. No way can any serious conclusion be drawn from this benchmark. My expectatiion is that the finished product will exceed a Haswell I7 and be close to a Broadwell I7

"An early engineering sample" of a product, which was supposed to launch in retail within ~ three months? "A primitive motherboard", which happens to be the exactly same as the one for a product already launched (on paper, Bristol Ridge AM4)?

 

[Glo.]

"An early engineering sample" of a product, which was supposed to launch in retail within ~ three months? "A primitive motherboard", which happens to be the exactly same as the one for a product already launched (on paper, Bristol Ridge AM4)?

Both can be exact case here. Even if they look completely ridiculous. You cannot exclude that this really is A0 version of the Zen, and is mounted on currently available product. Finished products can have much higher clocks.

 

[jpiniero]

I don't think the actual TDP is the main reason for the low clocks. A vast majority of the potential Zeppelin purchasers would still buy it despite the higher (e.g 125W) TDP, if there was large performance improvement over a 95W TDP model.

It may not be worth the effort/validation time/money since presumably people would just overclock. Now if Zen can't go much above 3.2 without the power draw spiraling out of control (or even at all), AMD's got bigger problems.

 

[cbn]

It may not be worth the effort/validation time/money since presumably people would just overclock. Now if Zen can't go much above 3.2 without the power draw spiraling out of control (or even at all), AMD's got bigger problems.

If the goal is high core count server and mobile (ie, BGA APU) then maybe having this restriction is not so bad even if they have to sacrifice enthusiast desktop a bit. I guess it just depends on how well the frequency vs. power scales at lower data points.

With that mentioned, I am wondering how well the interconnect for multiple die scaling will work for high core count server. How much gain going from 8C/16T to 16C/32T and beyond via whatever interconnect they will be using? (Will be very interesting to find this out).

 

[cbn]

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.

I do remember reading about 970 chipset AM3+ boards that listed support for 220W FX AM3+ CPUs but ended up with reported throttling. Yeah, that was not a good situation.

Perhaps AMD could offer two TDP levels (65W and 95W cTDP) for the 4C/8T Zen SKUs? Then have a 95W motherboard certification of some kind that would support the 95W cTDP up on 4C/8T if necessary. This depending on how badly the frequency and voltage curve were? (ie, How much more does a 95W 4C/8T need in the way of VRMs vs. 95W 8C/16T? higher voltage and lower amps vs. lower voltage and higher amps....how much impact on VRM design?

 

[superstition]

Perhaps AMD could offer two TDP levels (65W and 95W cTDP) for the 4C/8T Zen SKUs? Then have a 95W motherboard certification of some kind that would support the 95W cTDP up on 4C/8T if necessary.

They should do better than that for enthusiasts.

Not only is 95 watts much too little for a top-end enthusiast spec, it's time to end the overclocking Wild West where board makers sell consumers all sorts of specs of dubious relevance (20X copper!, bison-style chokes!) and obfuscate important ones like the actual VRM/power delivery specs.

1) The use of doublers should have to be explicitly stated (and not in tiny print).

2) There should be a big print box that gives the rating for the VRMs (including the efficiency of their heatsink at a specific standard amount of airflow and practical case temperature) at different temperatures and power draws, right on the packaging — so consumers will see what the power delivery is capable of.

Both of these changes should be required to obtain an AM4 certification, along with a sticker consumers can use to match their processor with the board (e.g. 95W CPU gets bronze or whatever). The sticker could have a color and also the maximum wattage it supports with a standard level of air and practical ambient and case temps.

AMD has the opportunity to force board makers to clean up their acts somewhat with AM4/Zen. There will always be at least one company that is willing to play ball. If the rest say no to the new regs then AMD can say "Fine, they'll get all the business". It's not like AMD would deny them the ability to sell cheap boards with cheap components. They would just have to be up-front about the quality/ability of the components in their boards.

There are too many stories of people being burned by seemingly capable boards that are nowhere close to being capable enough. People shouldn't have to scour forums in the hopes of finding this information.

And, really, it hasn't even always been an overclocking Wild West. Some of these boards have issues even at stock and not even with a 9000 series chip. It's a mess and AMD has a good opportunity now to step in and correct it.

 

[The Stilt]

Finished products can have much higher clocks.

Unless AMD never had a genuine intention or the real ability to ship any Zeppelins (regardless of the segment) in 2016, then they've had the final silicon available for months now. A newer die revision might technically exists even if AMD had a genuine intention to release them in 2016, however there is so far exactly zero indication about that. All of the three Zeppelin's we've seen so far have been ZP-A0 revision: The one displayed by Lisa Su at Computex, and both of the two SKUs benchmarked with AotS. "Michael Yuan Feng" who benchmarked and leakaged the AotS results appears to be a Chinese or Taiwanese person. Given the timing of these leaks, most likely this person either works for a system manufacturer (e.g. HP) or a OEM. In either case there is no need for AMD to send early prototypes (i.e non-qualification samples) to anyone outside the company for several reasons:

- To contain the leaks (impossible once the parts have entered China or Taiwan)
- Zeppelin is not required in order to design and validate the AM4 motherboards, initially. At least two of the biggest motherboard manufacturers used Bristol Ridge AM4 parts for the purpose. Same way the initial work on AM3+ was performed with AMD K10 chips. FM2 platforms were an exception and TN-A0 silicon was handed out, since there was no compatible CPUs / APUs which could be used until the final silicon arrived
- If newer than A0 silicon revision existed, why would AMD hand out superceded revision to a partner for qualification? By the time the leak occured the "newer revision" (whatever it would be, non A0) must have existed for ~ three months (to meet the requirement of launch capability in 2016).

Only AMD knows the truth, however I believe that they might have ended up to the same conclusion as I have: A newer die revision might not yield significant improvements in any of the critical areas and there are larger gains to be had from the process maturation itself. If that's the case then the obvious solution would be sit back and wait (postpone the launch and hope for the best). Based on the characteristics of Zeppelin and the issues AMD has had with all of their past designs (CPUs since K7 and GCN GPUs), I would expect that the first thing to limit the Fmax of the design itself would be the L2 caches. If that's the case then a newer die revision wouldn't make much of a difference, since there is no way you can change the latency of the caches with a simple respin. Despite the cache latency on Zeppelin can be considered as aggressive, I still believe Zeppelin will at least initially be limited by the manufacturing process and not by the design itself. If that's the case and there is a clear trend for the process improving to be seen, then postponing the launch would be the only right thing to do.

 

[krumme]

Not my fault if the first benchmarks do not match your expectations, but I wouldn't describe ~2.8-3.2 GHz cores with the IPC increase that AMD promised as 'strong cores' - and that's if they deliver. I'm sure you will still praise this product and carry it to the end of the universe even if it's a huge letdown, but some of us have been burned by their promises before, so we're more cautious this time.

What is wrong with you?
As i wrote, i havnt looked at the fps numbers you posted as they say nothing of importance to me.
I expected 2.5 3.2 even before the first leaks arived, so there is not much to it imo.
I dont beliewe how you expect us to beliewe you have been "burned by their promise before" and "were more cautions this time". lol.

 

[guskline]

Perhaps AMD wants to see Intel's Kabylake release before it releases Zen? I give Dr. Su credit for holding fast on her timelines.

There seems to be a fair amount of pentup "demand" for Zen so it should sell well when released.

AMD needs a solid release that tightens the gap with Intel. The difference can come in price.