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

[The Stilt]

The official 2.78a build AMD used in the demo still seem to lack the new SIMD optimizations. Pulled the newest version from Git and built it with MSVC 2015. 44.4% higher performance than on the official build, on Haswell-EP HCC.

In case anyone wants to try it out: https://1drv.ms/u/s!Ag6oE4SOsCmDhFAm03vWlB3s_qeD (40.5MB)
Password: "ryzen" (without the quotes)

EDIT: According to user reports, > 63% performance improvement over the official 2.78a build on Skylake.

 

[The Stilt]

Huh, it looks like they were running the 6900K with dual-channel memory. Would that make any substantial difference in those benchmarks?

I actually tested this at the time the first 6900K vs. Zen demo came out. Dual channel was MARGINALLY faster 🙂

 

[Glo.]

Right now, before the hype train will suck us, we need to see the real world benchmarks.

 

[UncleCrusty]

Colour me impressed. I do wonder if that Handbrake demo was making use of AVX2; if so, I'm extra impressed.

 

[moonbogg]

Right now, before the hype train will suck us, we need to see the real world benchmarks.

Yeah true, but we can at least rest assured that RYZEN is not another bulldozer. This will be a good CPU. And when I said quads were dead, I said Intel quads were dead, because no one will spend $400 on one with a 6/8 core Zen for around the same price or a little more. $400 for a quad core, yeah, GTFO please.

 

[The Stilt]

Colour me impressed. I do wonder if that Handbrake demo was making use of AVX2; if so, I'm extra impressed.

AVX2 makes almost no difference in X264. They've just started to write the assembly for it.

 

[frozentundra123456]

Yeah true, but we can at least rest assured that RYZEN is not another bulldozer. This will be a good CPU. And when I said quads were dead, I said Intel quads were dead, because no one will spend $400 on one with a 6/8 core Zen for around the same price or a little more. $400 for a quad core, yeah, GTFO please.

Well, lets wait for price and performance figures, huh? Plenty of people still buy intel quads with intel hex cores available at 400.00. Again, it is going to be the traditional trade off of single core performance vs more cores. Personally, four fast cores with hyperthreading is still more than enough for me.

 

[Abwx]

AMD performing better than SKL I'll have to see. From what I remember, the architecture isn't all that exotic.

OoO window: HSW level.
Loads: HSW/SKL.
Stores: HSW.
Scheduler entries: SKL.
Register: SKL.

You are wrong for scheduler entries..

Zen : 84 for INT + 96 for FP , that s a total of 180...

SKL : 97 shared for INT and FP

 

[Hail The Brain Slug]

The official 2.78a build AMD used in the demo still seem to lack the new SIMD optimizations. Pulled the newest version from Git and built it with MSVC 2015. 44.4% higher performance than on the official build, on Haswell-EP HCC.

In case anyone wants to try it out: https://1drv.ms/u/s!Ag6oE4SOsCmDhFAm03vWlB3s_qeD (40.5MB)
Password: "ryzen" (without the quotes)

EDIT: According to user reports, > 63% performance improvement over the official 2.78a build on Skylake.

To be fair, they never said what build or version they used in the demo. The only indication of version was that the download for their benchmark blender file said to go download 2.78a.

I'm having a hard time wrapping my head around that both the stock 6900K and the Ryzen chip rendered it in ~36 seconds, while on the release 2.78a my 4.5 GHz 5960x takes 38 seconds. This doesn't compute unless they are actually using the latest version they compiled themselves, unless someone else has a better explanation.

On your build, I score 26.72 seconds which sounds about right compared to a stock 6900K at 36 seconds.

Edit: For fun, a 4.6 GHz 6700k is at 45.62 seconds in your build.

 

[majord]

I think its great that Rizen performs well.

But I'm guessing they are gonna make money in the OEM bracket ?
I would think those individual customers that wanted 6 or 8 cores have had 6/8 cores for quite a while now. If performance is about equal between the 6/8 core Intel/AMD cpu's, what is the reason to switch to a Risen chip now?
I mean besides the holdout people in the " will only buy AMD" crowd?

I watched the video, I didn't see anything specific that stood out and said to me WOW that's new and much better than what I could have purchased for the past 2 years. Better IPC? no, higher clocks? no.

but overall , a good showing.

The performance in isolation is not the point. It's the idea we may have competition once again is.

 

[CHADBOGA]

If Zen holds up when reviewed by the independent websites, how valuable an employee is Jim Keller?

 

[UncleCrusty]

AVX2 makes almost no difference in X264. They've just started to write the assembly for it.

Ah. Do you know if it makes effective use of ordinary AVX?

 

[happy medium]

Lower tdp ?yes
Lower Price? Yes
Let's the market decide if competition is good enough or not.

Lower tdp at the same performance? we dont know
lower price when It releases? we don't know

 

[sirmo]

To be fair, they never said what build or version they used in the demo. The only indication of version was that the download for their benchmark blender file said to go download 2.78a.

I'm having a hard time wrapping my head around that both the stock 6900K and the Ryzen chip rendered it in ~36 seconds, while on the release 2.78a my 4.5 GHz 5960x takes 38 seconds. This doesn't compute unless they are actually using the latest version they compiled themselves, unless someone else has a better explanation.

On your build, I score 26.72 seconds which sounds about right compared to a stock 6900K at 36 seconds.

Edit: For fun, a 4.6 GHz 6700k is at 45.62 seconds in your build.

Are you talking about the Handbrake demo? It wasn't 36 seconds.

 

[Hail The Brain Slug]

Are you talking about the Handbrake demo? It wasn't 36 seconds.

No.

To be fair, they never said what build or version they used in the demo. The only indication of version was that the download for their benchmark blender file said to go download 2.78a.

They never displayed the Blender results but if you time the video and/or watch very carefully, it appears to take about 36 seconds and Blender looks to show a high 35.xx time on both systems.

 

[rainy]

Lol.
Intel is the biggest company and, not surprisingly, they are the most innovative one. Or more practically, they simply have the best products. So you might try again, but you really humiliate yourself if you point at Intel's enormous R&D budget. That's their asset, not a liability 😉.

Intel isn't the biggest company of IT industry and quite obviously you didn't understand my point, which was huge disparity of R&D budget between them and AMD - circa 13B vs just 1B.
Somehow AMD was still able to create an competitive CPU architecture.

 

[scannall]

Goddamn, it wasn't smoke and mirrors!

AMD actually is back!



Man, this is exciting. I think my 2500k's end is near next year...

Goddamn, it wasn't smoke and mirrors!

AMD actually is back!



Man, this is exciting. I think my 2500k's end is near next year...

Maybe my Westmere needs a vacation.... I sure do. ;-)

 

[DrMrLordX]

Well, Broadwell is not quite Intel's greatest as far as IPC, is it?

No. But Skylake isn't *that* much better. And at the present, it's the best uarch Intel has on HEDT. That'll remain the case until Q3 2017.

Those two tests make me think that their SMT is working as it should and per Core throughput is way higher vs CMT XCV designs.

Yes, and I am very happy to see that! All the estimations that people were doing based on the 40% ST IPC commentary lead to the conclusion that Zen's throughput would not beat XV's at all. Now we can see that it is the throughput monster it should be given its core width.

The only reason im not cheer-leading now is because Broadwell-E will almost be one year old when first ZEN will launch.
But AMD have made a hugeeeeeeeeeee leap in the right direction with ZEN and everyone should acknowledge that and be happy, because we will have some competition back in the Desktop/Laptop that we all here care about.

Well I have to say that AMD did what was necessary to win my support for the CPU. Once we see the actual SKUs, I'm going to line up and buy one. FM2+, I love ya, buuuut history is calling.

Makes me wish I had my 1.4 GHz Tbird still, just so it could watch.

Well, lets wait for price and performance figures, huh? Plenty of people still buy intel quads with intel hex cores available at 400.00. Again, it is going to be the traditional trade off of single core performance vs more cores. Personally, four fast cores with hyperthreading is still more than enough for me.

If it's more than enough for you, then great. Be happy with your purchase. But bear in mind that the 5960x still beats Skylake (yes, including the 6700k) in some gaming benchmarks just on the strength of its l3 alone.

And Ryzen has a whopping 20 MB l3 cache. This cache ain't BD/PD-style slowball L3 either.

Ryzen @ 4.0 GHz or higher will give the 7700k fits.

If Zen holds up when reviewed by the independent websites, how valuable an employee is Jim Keller?

Ask Tesla, maybe they will tell us?

 

[inf64]

PcPer shows a video of Handrake test.
6900K @ stock (with Turbo) : 66.5fps
Ryzen @ 3.4Ghz (fixed clock) 69.5fps

By the time the whole test finished(seen in live stream) zen took 54s Vs 59s for 6900K, so around 10% less time roughly. With dynamic boost enabled Ryzen will be very potent 16T chip for MT tasks, even at "stock".

 

[Maverick177]

No.



They never displayed the Blender results but if you time the video and/or watch very carefully, it appears to take about 36 seconds and Blender looks to show a high 35.xx time on both systems.

If I remember correctly, AMD showed both CPUs running at 3Ghz and Zen was actually faster by 2s.

 

[Hail The Brain Slug]

If I remember correctly, AMD showed both CPUs running at 3Ghz and Zen was actually faster by 2s.

We're looking at today's Ryzen event benchmarks, not the old blender benchmark they showed months ago.

 

[sirmo]

Really, kudos to the AMD guys. To not only design a grounds up architecture, but also do a tick and tock all at once, with fraction of the resources of their competitors, and deliver a processor with seemingly better perf/watt (at least in blender) on an arguably inferior process it's nothing short of amazing.

This is like Leicester winning the league. Or what Cubs did this year if you're into baseball.

 

[DrMrLordX]

Or what Cubs did this year if you're into baseball.

Holy cow!

 

[Edgemeal]

The official 2.78a build AMD used in the demo still seem to lack the new SIMD optimizations. Pulled the newest version from Git and built it with MSVC 2015. 44.4% higher performance than on the official build, on Haswell-EP HCC.

In case anyone wants to try it out: https://1drv.ms/u/s!Ag6oE4SOsCmDhFAm03vWlB3s_qeD (40.5MB)
Password: "ryzen" (without the quotes)

EDIT: According to user reports, > 63% performance improvement over the official 2.78a build on Skylake.

FWIW,

i7-3770 @ 4GHz (100 x 40)
blender-2.78a-windows64.zip........101.16 seconds
Blender_278_1412GIT_AVX2_MSVC.7z... 75.43 seconds

 

[witeken]

Intel isn't the biggest company of IT industry and quite obviously you didn't understand my point, which was huge disparity of R&D budget between them and AMD - circa 13B vs just 1B.
Somehow AMD was still able to create an competitive CPU architecture.

Intel doesn't spend $13B on CPU architecture development. ARM is also able to produce decent architectures with even less money.

The reason AMD is able to do this is because you can't make CPUs faster indefinitely. The high-hanging fruit will give you what, maybe 10% IPC? That's a lot of effort for not so much increase in performance.