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

[SpaceBeer]

So make presentation in January, say the sales starts in February, it will actually be able available in March, due to high demand, prices will go up and will stabilize in August :|

 

[KTE]

March pushout would be a bigger delay... Meaning silicon/platform issues are still prevalent.

I did say at the very beginning that I could see them buying as much time as possible for better clocks@TDP.

Sent from HTC 10
(Opinions are own)

 

[mikk]

In the meantime Skylake-E is not far away after all these delays. They have to deal with Skylake-E for the most part during the lifetime of Zen v1, the window where it could compete against Broadwell-E becomes smaller and smaller.

 

[krumme]

What clocks will 8c zen+ have when it is clocked low enough to fit in the consoles tdp budget?

edit : and here comes the bug creep it seems .. here is to hoping that nothing serious turns up .

(the way AMD has slow played the hand so far, crippling smt and tracecache could be another iteration of that slowplay .. who knows.)

Yeaa. We have a 180w tdp 32c server variant comming this q2 = 45w 8c. So 30w for a 8c zen+ at 7nm seems realistically to me? Seems the same to me as tdp for 8c jaguar at 1.7GHz 28nm

 

[jpiniero]

In the meantime Skylake-E is not far away after all these delays. They have to deal with Skylake-E for the most part during the lifetime of Zen v1, the window where it could compete against Broadwell-E becomes smaller and smaller.

Well, Skylake-X is August. Even if real availability is not until end of March-April, that's still 4-5 months.

 

[zinfamous]

So make presentation in January, say the sales starts in February, it will actually be able available in March, due to high demand, prices will go up and will stabilize in August :|

Yep. This also puts them on the timetable to see some Zen sales show up Q1 reports for AMD, though maybe not so significant. Q1 report will show mobo and CPU orders for sure, if consumers are able to buy these things off of shelves by March.

CES next week should have more solid details for Zen with a real release schedule. Interesting times.

 

[krumme]

In the meantime Skylake-E is not far away after all these delays. They have to deal with Skylake-E for the most part during the lifetime of Zen v1, the window where it could compete against Broadwell-E becomes smaller and smaller.

So what. Amd is going for the i3 i5 and i7 market not 1000 usd cpu on an expensive mb. This is build for AM4 not x99. Different segments. Perhaps highest prices will be into lowend hedt segment.
What matters now is cpu efficiency, yield and gf capacity.
Bristol Ridge is pentium/celeron/atom competitor.
That article you linked showed the oem was impressed by perf/ $ and expected a rush and even an effect to polaris and vega sold. Clearly this is not a niche product.

 

[cytg111]

Yeaa. We have a 180w tdp 32c server variant comming this q2 = 45w 8c. So 30w for a 8c zen+ at 7nm seems realistically to me? Seems the same to me as tdp for 8c jaguar at 1.7GHz 28nm

Okay, but at what clocks? We do know that clocks and watts are not proportional, theres diminishing returns both on the upper and the lower end of that graph. I also think a new generation of consoles will be on 14nm, consoles are iterating faster and 14nm is gonna be a long node for AMD.. My crystal 8-ball says : too volatile, try again tomorrow.

 

[mikk]

So what. Amd is going for the i3 i5 and i7 market not 1000 usd cpu on an expensive mb.

They don't have to go for the 1000 USD market. Broadwell-E 6C is listed for less than 450 USD. Zen starts with the biggest SKUs and don't expect prices in a range of 200-300 USD. It wouldn't make sense with the poor availability. Zen 4C and 6C is a different matter but they are further away. It's too early to say something about this.

 

[krumme]

Okay, but at what clocks? We do know that clocks and watts are not proportional, theres diminishing returns both on the upper and the lower end of that graph. I also think a new generation of consoles will be on 14nm, consoles are iterating faster and 14nm is gonna be a long node for AMD.. My crystal 8-ball says : too volatile, try again tomorrow.

We dont know but if its made on 14lpp and a server uses it its probably optimal for tdp/perf as that have the largest impact on tco. Should be the same for a console then. So as it is on 14lpp we can have a 45w tdp already waiting for us. 300w tdp for a console and there is aprox 200w for some vega gpu part. Then its cpu heavy but then they can upgrade gpu midway.

But hey i mean they just launched ps4 pro. Is 3 years out not reasonably. Meaning eg q4 2019 for ps5? Thats fine for 7nm. Or perhaps some cheaper 10nm.

 

[USER8000]

If it performs like that with the SMT and uop cashe nerfed what will happen when it is fully functional?

The productivity tests were done with nerfed HT?? Wait,wut??

 

[krumme]

They don't have to go for the 1000 USD market. Broadwell-E 6C is listed for less than 450 USD. Zen starts with the biggest SKUs and don't expect prices in a range of 200-300 USD. It wouldn't make sense with the poor availability. Zen 4C and 6C is a different matter but they are further away. It's too early to say something about this.

Availability will play a big part. They need to ramp up and even if amd fan base is miniscule the firstmovers here will have an effect on price too. But when it settles after a few months the market for zen is where Intel have their core line. Its not a matter of performance but price. They can have an xxx black skull edition that sells for 700usd but it really doesnt matter except for brand development and mindshare. Most cost is fixed, capacity is there to be used, dev cost is sunk, so its all just about maximizing profit by units sold at profit per unit. And Intel shows and knows what price segments there is volume. Amd is so small they cant change that a bit.

 

[Abwx]

The productivity tests were done with nerfed HT?? Wait,wut??

The article say that there s a fix but that it wasnt available set apart for a happy fews, all CanardPC had was a bios that unlocked all Zen features (despite said limitations that could result in instabilities according to CPC) as if it was a final chip.

 

[cytg111]

The productivity tests were done with nerfed HT?? Wait,wut??

Depending on how severe the bugs are I am guessing we may see a refresh at some point with those things fixed. Who knows.
edit : Oh, allready fixed it seems. Good times .

 

[Glo.]

The leak from African forum from previous months has claimed exactly same thing, that there is a bug in SMT. Interesting that leak, publicly claimed by many to be fake, have turned out to be accurate, at least in this occasion.

 

[The Stilt]

The leak from African forum from previous months has claimed exactly same thing, that there is a bug in SMT. Interesting that leak, publicly claimed by many to be fake, have turned out to be accurate, at least in this occasion.

The "leak" said exactly nothing about a bug in SMT. It claimed that there is a bug, similar to the TLB issue in Phenom which causes a performance penalty of 30-40%. Based on the leaked CPC results (alleged), does it appear to you that there is 30-40% of performance missing?

 

[bjt2]

The "leak" said exactly nothing about a bug in SMT. It claimed that there is a bug, similar to the TLB issue in Phenom which causes a performance penalty of 30-40%. Based on the leaked CPC results (alleged), does it appear to you that there is 30-40% of performance missing?

Probabily 30%-40% will be missing if you disable uop cache and/or SMT to avoid the bugs. But it seems that french magazine keep SMT and uop cache activated, risking the bug, but not losing much performance...

Anyway i think that a MB in which NVIDIA GPUs and SSDs does not work can not be classified as retail class and so i think that probabily some other few percent of performance was lost...

 

[PhonakV30]

Hell no , 30% means Zen at 3.5Ghz above Skylake at 4.2Ghz.No way

 

[Maverick177]

Ok here's the big fat question: Will Ryzen beat 6900K ?

No nonsense, Yes or No.

 

[bjt2]

Ok here's the big fat question: Will Ryzen beat 6900K ?

No nonsense, Yes or No.

If it will clock >= 3.6GHz base and >=4GHz turbo, probabily yes...

 

[witeken]

Ok here's the big fat question: Will Ryzen beat 6900K ?

No nonsense, Yes or No.

Will depend on benchmark.

 

[rvborgh]

Anyone know if Zen will support transactional memory?

https://en.wikipedia.org/wiki/Advanced_Synchronization_Facility

 

[lolfail9001]

Ok here's the big fat question: Will Ryzen beat 6900K ?

No nonsense, Yes or No.

Nobody actually can tell you that, and even if they can, it will highly depend on task most likely.

I will make a bet that at equal clocks 6900k will be better in gaming, though.

 

[DrMrLordX]

http://www.digitimes.com/news/a20161226PD208.html

That's pretty disappointing, if true. We had been told before that it was the lower-end chips shipping in March, not the entire set. I guess if they have SMT bugs they haven't been able to sort out, they need more delays . . .

CES next week should have more solid details for Zen with a real release schedule. Interesting times.

I sincerely wish they would keep to their earlier timeline of selling their high-end chip in January! How are they going to have a Ryzen giveaway on Jan 18th when the chips don't even ship until March? Do you get a voucher for a future chip or what?

 

[blublub]

Huh march would certainly suck since on the 13th Su said "soon" and " on track" - march is at least not "soon"