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

[KTE]

Seems just like what I've been saying so far... Phenom Repeated.

Undeniably good performance uplift but... IPC uplift indications seem not to be EXC but PD compared ?

I predicted that it will be too big/too power hungry/too low clocks with good base performance. Their FF tech seems problematic. I'm sticking to this.

But this leak isn't enough to conclude any of this on. Just my estimations from the info we've always had.

 

[Shehriazad]

It's no big surprise to me that an 8c/16 thread CPU confined to 95w TDP is a letdown in terms of raw singlecore/multicore perf. Not sure how anyone could expect ANYTHING else.

If you translate this CPU to a cut down 4c/8 thread chip, though...it gets a lot more interesting. Half the physical cores should allow for a way higher base/turbo clock and the rest... is pricing. I'm not saying they have to go super cheap mode again...but Intel is actually giving them enough room to undercut them just right for the product to have a right to exist.

Not to mention, I don't know how exactly the voltage scales on this process, but I feel like they can probably push past 4Ghz a bit and remain at 95w tdp, no? (with 4c/8t of course)

 

[jpiniero]

There is a single Zeppelin die. Harvesting will allow AMD to create parts with 2, 4, 6 core SKUs with 0MB, 8MB or 16MB L3 upon demand. Zeppelin is a multipurpose core, same way Orochi was.

With 16 MB of L3, it can't be a tiny die. That seems like it'd get pretty expensive if they had to sell otherwise good dies at Athlon prices.

 

[The Stilt]

With 16 MB of L3, it can't be a tiny die. That seems like it'd get pretty expensive if they had to sell otherwise good dies at Athlon prices.

If they can sell Polaris 10 based graphics cards for 199$, then they can sell Zeppelin for much cheaper. Realistically, none of the Zeppelin consumer SKUs can be sold for higher than ~350$ (or whatever the i7-5820K, i7-6700K, i7-6800K prices at that moment are). I'd say 179$ for the fastest 4C/8T and 289$ for the fastest 8C/16T SKUs are pretty realistic, unless AMD goes batshi* crazy with the pricing as they did with RX 460 and RX 470 cards.

 

[blublub]

Except that's wasn't the case. The second major prototype stepping (OR-A1, ZD262046W8K43 SKU) already reached 3.6GHz, which is 400MHz lower than the fastest retail part (FX-8120) did at the same TDP.
A1 stepping of Bulldozer was taped out a year (September-October 2010) before the final product was launched. That's why the situation is different to Zeppelin, since obviously we are not talking about a product still year away (hopefully at last).

Yep I was wrong, but the 1st A0 was 3,3GHz, a whooping 700Mhz below final clock speed.

Link:
http://www.overclock.net/t/1291114/coolaler-amd-piledriver-fx-vishera-engineering-sample-benchmarks


So we can expectsome imprevemtn here, maybe a 3,5 Ghz base clock

 

[SAAA]

With 16 MB of L3, it can't be a tiny die. That seems like it'd get pretty expensive if they had to sell otherwise good dies at Athlon prices.

Current estimations put the die at some 160-200mm^2, for reference that's similar to 160mm^2 Xeon-D eight cores that also have 16MB cache on die.
What we don't know yet is if they will die harvest parts for 6 cores from 8, but surely not down to 4. That's probably another die entirely with 4 native cores and a GPU cluster that might get disabled on Athlon parts. I also expect it to be smaller if the APU has only 12 CUs per leaked slide, probably some 120mm^2 or so.

If they can sell Polaris 10 based graphics cards for 199$, then they can sell Zeppelin for much cheaper. Realistically, none of the Zeppelin consumer SKUs can be sold for higher than ~350$ (or whatever the i7-5820K, i7-6700K, i7-6800K prices at that moment are). I'd say 179$ for the fastest 4C/8T and 289$ for the fastest 8C/16T SKUs are pretty realistic, unless AMD goes batshi* crazy with the pricing as they did with RX 460 and RX 470 cards.

That's reasonable but Zen must have costed something being a new architecture and all, if it does perform good I could see the 8 core pushed to +400$ just to get some money back. The die size alone could make it cost 150$ and still be profitable given Polaris size but I doubt it will fall that low unless it's a total fail.

 

[The Stilt]

Yep I was wrong, but the 1st A0 was 3,3GHz, a whooping 700Mhz below final clock speed.

Link:
http://www.overclock.net/t/1291114/coolaler-amd-piledriver-fx-vishera-engineering-sample-benchmarks


So we can expectsome imprevemtn here, maybe a 3,5 Ghz base clock

Single core turbo @ 3.5GHz might be possible if the process improves enough, but there is no power budget available for 3.5GHz base even if it was otherwise possible.
Remember that the CCXs alone do not have the full power budget in their disposal. On 95W TDP parts I would expect that the NB and the FCH consume at least 15W combined.

 

[AtenRa]

On 95W TDP parts I would expect that the NB and the FCH consume at least 15W combined.

Since 990FX with 42x PCIe Gen 2.0 lanes at 65nm was ~20W, ZENs NB and FCH at 14nm will not even use 5W

 

[Erenhardt]

I hope AM3 motherboard manufacturers make it easier to disable TDP limit in bios. FM2+ was a nightmare

 

[The Stilt]

Since 990FX with 42x PCIe Gen 2.0 lanes at 65nm was ~20W, ZENs NB and FCH at 14nm would not even use 5W

This is a face palm of such a magnitide that I have no other option but to put you into ignore
I'm pretty certain I got cancer of some sort from reading this.

If you really think you can run the northbridge, the memory controllers, GMI links, PCIe links, SATA & USB controller, a system management processor, an encryption processor, IOMMU and various clock generators at less than five watts you are pretty delusional.

ps. The 18W for SR5690 (990FX) does not include the southbridge (SBxxx), which includes most of the chipset functionality outside the PCIe links.


Having a family die from cancer and watching him pass away, trust me, you don't want cancer nor should it be wished upon to anyone.

I dont find jokes like this funny, or smart.
So lets keep these figure of speeches like this to a minimum.

Moderator Aigo

 

[blublub]

Single core turbo @ 3.5GHz might be possible if the process improves enough, but there is no power budget available for 3.5GHz base even if it was otherwise possible.
Remember that the CCXs alone do not have the full power budget in their disposal. On 95W TDP parts I would expect that the NB and the FCH consume at least 15W combined.

I don't think that NB will draw that much power.

However what I had forgotten about was that Intel 6900K has a TDP of 140W - which is 22,9Mhz/Watt and ZEN has 29,5Mhz/watt - this is pretty impressive and might come down to the same or maybe slightly better performance/watt compared to Intel Skylake

 

[SpaceBeer]

And yet i7-6700 has TDP of 65W, with 3.4/4 Ghz clocks. Or maybe to compare it with 6700T with 35W TDP and 2.8/3.8 Ghz?

AMD (Zen) can not be as efficient as Skylake (or Kaby lake), not after Intel has invested billions of $ in power efficiency. So I hope they will give us unlocked CPUs, and we will care about power consumption and electirc bills

 

[blublub]

And yet i7-6700 has TDP of 65W, with 3.4/4 Ghz clocks. Or maybe to compare it with 6700T with 35W TDP and 2.8/3.8 Ghz?

AMD (Zen) can not be as efficient as Skylake (or Kaby lake), not after Intel has invested billions of $ in power efficiency. So I hope they will give us unlocked CPUs, and we will care about power consumption and electirc bills

Yes but the leak shows the 8c so we will have to wait and see what ZEN 4c will be like with 65W TDP.....

 

[AtenRa]

This is a face palm of such a magnitide that I have no other option but to put you into ignore
I'm pretty certain I got cancer of some sort from reading this.

If you really think you can run the northbridge, the memory controllers, GMI links, PCIe links, SATA & USB controller, a system management processor, an encryption processor, IOMMU and various clock generators at less than five watts you are pretty delusional.

ps. The 18W for SR5690 (990FX) does not include the southbridge (SBxxx), which includes most of the chipset functionality outside the PCIe links.

SB950 (southBridge) that has 6x SATA 6 + 14x USB Gen 2.0 + PCIe lanes for connecting Gigabit Ethernet + HD Audio + other devices , at 65nm is only 6W.

FCH A88X Bolton for the FM2+ socket with 8x SATA-6 ports, 4x USB 3 + 8x USB 2.0 at 65nm is only 7.8W

Also, the memory controller is not part of the FCH.

At 14nm FF(integrated to the die), the FCH and the PCIe Gen 3.0 should not be more than 5-6W.

AMD managed to integrate the FCH in to a 65W TDP APU at 28nm (only 8x PCIe Gen 3.0 lanes) and you really believe the FCH + PCIe Gen 3.0 lanes integrated in to the die and produced at 14nm FF (more than 4x less power than 65nm) will use 15W ??

 

[cbn]

Current estimations put the die at some 160-200mm^2, for reference that's similar to 160mm^2 Xeon-D eight cores that also have 16MB cache on die.

Xeon-D actually has 12MB L3 cache (in addition to 256KB L2 cache per core). It also has 24 PCIe lanes, dual channel memory controller and usb 3.0 on the same die.

.

What we don't know yet is if they will die harvest parts for 6 cores from 8, but surely not down to 4. That's probably another die entirely with 4 native cores and a GPU cluster that might get disabled on Athlon parts.

Do we know if Zen APU will have L3 cache? If doesn't then I think AMD could harvest quad core from octocore. And even if the APU does have L3 cache, it might be the octocore die is smaller than an APU with 12CUs and L3 cache.

 

[Phynaz]

This is a face palm of such a magnitide that I have no other option but to put you into ignore
I'm pretty certain I got cancer of some sort from reading this.

If you really think you can run the northbridge, the memory controllers, GMI links, PCIe links, SATA & USB controller, a system management processor, an encryption processor, IOMMU and various clock generators at less than five watts you are pretty delusional.

ps. The 18W for SR5690 (990FX) does not include the southbridge (SBxxx), which includes most of the chipset functionality outside the PCIe links.

The new ignore is awesome. You don't even see when somebody else quotes the person. It's as if the person you put on ignore doesn't even exist. I just had to do it a few minutes ago m

 

[aigomorla]

You guys i have received more reports from this thread alone since it was created then what cpu and overclocking has seen in the past few months combined.

If a member is bothering you, or just getting under your rugs, use the ignore feature, don't let that member bait you.

If you drift off Topic... well your in the wrong subform... Off Topic is That way ->
You will be given infractions for off topic.

Moderator Aigo

 

[Erenhardt]

If you drift off Topic... well your in the wrong subform... Off Topic is That way ->
You will be given infractions for off topic.

Moderator Aigo

Speaking of it. How to get my forum navigation sitebar back?

 

[SpaceBeer]

Yes but the leak shows the 8c so we will have to wait and see what ZEN 4c will be like with 65W TDP.....

According to this leak
https://techreport.com/news/30514/rumor-amd-zen-engineering-samples-leaked-and-benchmarked
4c/8t Zen has TDP of 65W and 2.8/3.2 clocks. Skylake i7-6700 has 25% higher clocks with same TDP. I'm sure Intel's 14nm process is better than GF 14LPP, but don't know how much

 

[cbn]

Regarding the discussion of where 4C/8T Zen enthusiast AM4 desktops chips would come from (cut down 8C/16T Zen or 4C/8TZen APU with iGPU disabled ) Ideally there would not be many of these. (re: under ideal circumstances all 8C/16T dies would go to server and be fully utilized and all 4C/8T APU dies would go to BGA mobile and be fully utilized with iGPU).

And if AMD is going to make a desktop 4C/8T with whatever scant amount of harvested 4C/8T I don't think it should be 65W. 65W is just too in between a 35W/45W BGA mobile and a workstation level 95W chip to be meaningful or purposeful IMO. 4C/8T AM4 really needs to be 95W. Same goes for any harvested dual core APU, it should also be 95W.

As far as the OEMs go, I think AMD should point them at BGA for desktop and let AM4 be strictly an enthusiast DIY platform (using the handful of chips AMD has leftover from the Server and mobile efforts).

 

[blublub]

According to this leak
https://techreport.com/news/30514/rumor-amd-zen-engineering-samples-leaked-and-benchmarked
4c/8t Zen has TDP of 65W and 2.8/3.2 clocks. Skylake i7-6700 has 25% higher clocks with same TDP. I'm sure Intel's 14nm process is better than GF 14LPP, but don't know how much

Yeah that doesn't add up then ....so all speculation is probably BS anyways

 

[cbn]

Another strike against 65W is all the effort AMD has made for quiet 95W coolers.

 

[SAAA]

Regarding the discussion of where 4C/8T Zen enthusiast AM4 desktops chips would come from (cut down 8C/16T Zen or 4C/8TZen APU with iGPU disabled ) Ideally there would not be many of these. (re: under ideal circumstances all 8C/16T dies would go to server and be fully utilized and all 4C/8T APU dies would go to BGA mobile and be fully utilized with iGPU).

And if AMD is going to make a desktop 4C/8T with whatever scant amount of harvested 4C/8T I don't think it should be 65W. 65W is just too in between a 35W/45W BGA mobile and a workstation level 95W chip to be meaningful or purposeful IMO. 4C/8T AM4 really needs to be 95W. Same goes for any harvested dual core APU, it should also be 95W.

As far as the OEMs go, I think AMD should point them at BGA for desktop and let AM4 be strictly an enthusiast DIY platform (using the handful of chips AMD has leftover from the Server and mobile efforts).

About the whole Zen lineup: I have the feeling that AMD is trying to go after Intel segments and will offer a comparable or better speced part at every price.
Quad cores with IGP? That's mainstream Skylake equivalent, if they price it right a good GPU and 8 thread might be a decent offer against locked i5, pheraps even the unlocked parts.
Eight cores? That's aiming at enthusiasts and will battle mostly the 6 cores, it's what existed when Zen was planned anyway, so current extreme 8-10 cores might be a little out of reach.
For those we might see some 16 cores released at some point, or higher clocked stock parts like the FX-9000.

They are surely in a much better condition to compete with the planned IPC increase, regardless of frequency, because speed can be adjusted with new process/stepping but microarchitecture is the foundation for a whole generation of products. I'm probably more interested in AMD finally getting good laptops/tablets with Zen and their APUs than high end: 4W TDP sounds low enough to finally catch Core-M and all the mobility crowd.

 

[cbn]

Regarding the discussion of where 4C/8T Zen enthusiast AM4 desktops chips would come from (cut down 8C/16T Zen or 4C/8TZen APU with iGPU disabled ) Ideally there would not be many of these. (re: under ideal circumstances all 8C/16T dies would go to server and be fully utilized and all 4C/8T APU dies would go to BGA mobile and be fully utilized with iGPU).

And if AMD is going to make a desktop 4C/8T with whatever scant amount of harvested 4C/8T I don't think it should be 65W. 65W is just too in between a 35W/45W BGA mobile and a workstation level 95W chip to be meaningful or purposeful IMO. 4C/8T AM4 really needs to be 95W. Same goes for any harvested dual core APU, it should also be 95W.

As far as the OEMs go, I think AMD should point them at BGA for desktop and let AM4 be strictly an enthusiast DIY platform (using the handful of chips AMD has leftover from the Server and mobile efforts).

About the whole Zen lineup: I have the feeling that AMD is trying to go after Intel segments and will offer a comparable or better speced part at every price.
Quad cores with IGP? That's mainstream Skylake equivalent, if they price it right a good GPU and 8 thread might be a decent offer against locked i5, pheraps even the unlocked parts.
Eight cores? That's aiming at enthusiasts and will battle mostly the 6 cores, it's what existed when Zen was planned anyway, so current extreme 8-10 cores might be a little out of reach.
For those we might see some 16 cores released at some point, or higher clocked stock parts like the FX-9000.

I understand the rationale about going after Intel segments (example: OEMs like 65W or lower for socketed desktops), but couldn't a 45W BGA APU work for that?

Why does the APU need to be socketed like Intel for OEM desktops? Remember Intel has a very wide range of socketed processors ranging from Celeron to i7 (with corresponding wide swing in capability), but the APU range would much more narrow. In fact, if all BGA APUs were quad core there wouldn't be much swing in capability and thus I wonder why a socket even needs to exist?

 

[AtenRa]

Does anyone has a pic of the ZEN AoTS benchmark run used in the OP ??

Just run the benchmark with my Core i7 3770K @ 4.44GHz + 16GB ram at 2154MHz + R9 390 default with Crimson 16.7.3 Win 10 64bit