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

[RussianSensation]

Beating Broadwell-E clock for clock even if it's just Blender is still an amazing achievement. To design an architecture from scratch and leapfrog decades of Intel's CPU development (on an inferior process no less, with a fraction of the R&D budget) is nothing short of amazing. Many of us hoped for near Haswell IPC. Remains to be seen how it clocks and how well it does in other disciplines, but this is looking really promising.

True, esp. if 3rd party independent sources can validate AMD's benchmarks. We've seen too many times AMD's over-hyped marketing PowerPoint slides not translating into real world performance. If Zen can reach BW-E IPC in heavily multi-threaded apps, it's a huge win. It cannot be understated given Intel's R&D advantage, human capital advantage, at least 3 full/major architectural leads over Vishera in terms of IPC (Nehalem -> Sandy/Ivy -> Haswell/Broadwell) and having world's best fabs. I have no doubt that Intel will continue to have the performance lead since in 2017 they will introduce KB-X and SKL-X. Zen seems to form a nice foundation for Zen+ for AMD.

It'll be interesting to see what AMD's final CPU clock speeds will be, overclocking headroom and pricing given the 8-core competitor's products cost almost $1,100. Even if AMD doesn't outright win against the competition, the gap in pricing between a 4-core HT $350 processor and 8-core $1100 one is so vast that AMD should be able to find an attractive price/performance market segment. Hopefully the added cash flow from Zen sales allows AMD to hire even better engineers and net another 10-15% IPC improvement by 2020, while increasing CPU clocks to 4.0-4.2Ghz over the next 3 years. Competition could ensure the blue team may just be forced to bring out a 6-core HT mainstream platform offering under $400. I will still wait for final reviews and benchmarks as I have been too disappointed by BD. I still think for gamers Zen won't be a game changer but for servers/workstations and power users, it could finally be a new alternative.

Personally, I have no regrets buying a 6700K though. By the time Zen and SKL-X/KBL-X roll around in 2017, Skylake Z170 6700K platform will be almost 2 years old. In hindsight, imho, it wasn't worth waiting for either of these platforms when the alternative was enjoying cutting edge performance (and most likely for less $$$ too) all this time. I think Zen went after the lucrative server market anyway. It makes sense since that's where AMD could stand to gain the most design wins/market share.

 

[KTE]

Unfortunately the maximum clock is in many ways controlled by the manufacturing process. If GloFo can't deliver the clocks then AMD can't do much about it. And I bet that's the real issue right now. The Broadwell Blender video and the leaked ES sample results we've been dissecting shows that IPC improvements are good. Everything now rests on the final clock speeds and price.

I wonder if the delay until 2017 is in order to give GloFo more time to improve their still immature process? Processes get better with time, with the biggest efficiency improvements usually coming early in its lifespan. A delay of only a few months could result in some relatively big gains in efficiency (or usable chips). Maybe 3.2ghz is the max clock possible today, but in a few months 3.5ghz might be obtainable. If that's the case than AMD needs to delay. Too much is riding on a succesful Zen launch. They need to put their best foot forward from the beginning.

Sent from HTC 10
(Opinions are own)

 

[DrMrLordX]

They exactly said IPC.

Right, but remember that many of us interpreted their statement as "one Zen core w/out SMT would be 40% faster than an XV module running a single thread of software". I made the leap of logic some time ago that they also meant that eight Zen cores with SMT would be 40% faster than 8 XV modules while handling 16 threads.

If I was right - if AMD really pulled that off - then that's huge, and, in their own vague way, they told us it would happen some time ago.

 

[dark zero]

Seems that Samsung is clearly behind this chip. I see clearly the mark of efficiency of that chip from them. Is their personal fight against Intel and not AMD themselves.

Sadly I can't see OCing capabilities, but still is a magnificent return of AMD if this is confirmed. If is even 10% less, the winnings are so massive that impress even the haters...

PS: If Raven ridge manages to come before the 2nd movie of Star Wars, I want to see HP doing an SS promotion like this:
2 models, 2 sides, the Force with Intel and nVIDIA combo and the Dark Side with AMD CPU and dGPU.

 

[VirtualLarry]

PS: If Raven ridge manages to come before the 2nd movie of Star Wars, I want to see HP doing an SS promotion like this:
2 models, 2 sides, the Force with Intel and nVIDIA combo and the Dark Side with AMD CPU and dGPU.

I think that you've got that backwards. Intel and NV would be "the empire", and AMD would be "the rebels".

 

[The Stilt]

Right, but remember that many of us interpreted their statement as "one Zen core w/out SMT would be 40% faster than an XV module running a single thread of software"

That assumption is incorrect.
AMD has stated 40% IPC improvement over a Excavator core from the day one, and according to AMD a single module is two cores on every 15h design.

 

[Abwx]

That assumption is incorrect.
AMD has stated 40% IPC improvement over a Excavator core from the day one, and according to AMD a single module is two cores on every 15h design.

And he stated :

an XV module running a single thread of software"

So he s saying a single core without CMT penalty.

 

[DrMrLordX]

That assumption is incorrect.
AMD has stated 40% IPC improvement over a Excavator core from the day one, and according to AMD a single module is two cores on every 15h design.

But if that were what they really meant, there's no way Summit Ridge ever could have matched Broadwell-E in throughput at the same clockspeed. It wouldn't even have touched Haswell.

 

[majord]

But if that were what they really meant, there's no way Summit Ridge ever could have matched Broadwell-E in throughput at the same clockspeed. It wouldn't even have touched Haswell.

Correct.. And there's nothing to say it will match haswell/broadwell on average

This is now being over-hyped based on a single benchmark, designed to display that it can likely throw some good blows at BW/HW, but in other areas no doubt it will fall well behind. It's still impressive nonetheless.


Forget about throughput for now.. Now that both architectures are wide, SMT based deisgns, we don't have to seperate single thread "IPC" and total throughput "IPC" as dramatically as we had to with Construction cores

 

[YBS1]

I'm not comfortable with the assumption that Zen will be a lot more afforable than Intel offerings. If you guys were around during the Athlon XP and Athlon 64 era, AMD charged according for them. AMD is not running a charity.

I see the same mistaken assumption before the launch of RX 480. People were actually expecting it to cost $199 and beat a GTX 1070. If it were capable of beating a 1070 it would be priced accordingly. AMD charges what they can, people. Let's not make that mistake.

People weren't expecting that and AMD wasn't promising that, what led to some of those feelings right before release was there were a lot of (we have to assume now) completely B.S. leaks of "easy 15-1600MHz OC's on air". I'm with you on the pricing somewhat, if by some miracle AMD trounced Intel they would rape us in a heartbeat, they aren't a charity, anyone who thinks otherwise are foolish. However, if they are competitive but not dominant, then as long as they are pulling down some decent profit I would expect them to well undercut Intel simply to establish market share. High volume with decent profit will do them better than high margins and crap volume.

 

[Enigmoid]

While this looks promising I will hold any applause. Lets not forget the last "live demo" debacle of polaris vs. the 950,

And these numbers were also demoed live. This claim turned out to be grossly misleading with polaris.

https://www.computerbase.de/2016-08...nahme-des-gesamtsystems-star-wars-battlefront
https://www.computerbase.de/2016-08/radeon-rx-460-test/3/#diagramm-star-wars-battlefront-1920-1080_2

I'm sorry AMD but you do not have the benefit of the doubt with me any longer. I will not believe any performance claims you make, even live demos, without seeing how the hardware performs by third parties.

Edit: I am not trying to threadcrap but simply saying that even live demos can be presented in a way that is not at all representative of what you will get in the real world. I think the blender test is nice but it is by all means inconclusive. Generally the safe bet, proved time and time again by Nvidia, Intel, and AMD is that the actual product never lives up to the marketing.

 

[YBS1]

While this looks promising I will hold any applause. Lets not forget the last "live demo" debacle of polaris vs. the 950,

And these numbers were also demoed live. This claim turned out to be grossly misleading with polaris.

https://www.computerbase.de/2016-08...nahme-des-gesamtsystems-star-wars-battlefront
https://www.computerbase.de/2016-08/radeon-rx-460-test/3/#diagramm-star-wars-battlefront-1920-1080_2

I'm sorry AMD but you do not have the benefit of the doubt with me any longer. I will not believe any performance claims you make, even live demos, without seeing how the hardware performs by third parties.

Edit: I am not trying to threadcrap but simply saying that even live demos can be presented in a way that is not at all representative of what you will get in the real world. I think the blender test is nice but it is by all means inconclusive. Generally the safe bet, proved time and time again by Nvidia, Intel, and AMD is that the actual product never lives up to the marketing.

Just a guess as I'm not at all familiar with Battlefront or the GTX 950. I could totally see this result being "legit" as it's obviously done with vsync on as they both scored 60fps. Knowing that the 480 is significantly faster than a 950, I imagine the settings were chosen as they were because the 950 is likely screaming it's guts out to barely achieve 60fps, and the 480 with vsync on is probably coasting and running well below it's max boost. Turning vsync off would likely skyrocket the 480's power usage as well as it's performance though.

 

[Enigmoid]

Just a guess as I'm not at all familiar with Battlefront or the GTX 950. I could totally see this result being "legit" as it's obviously done with vsync on as they both scored 60fps. Knowing that the 480 is significantly faster than a 950, I imagine the settings were chosen as they were because the 950 is likely screaming it's guts out to barely achieve 60fps, and the 480 with vsync on is probably coasting and running well below it's max boost. Turning vsync off would likely skyrocket the 480's power usage as well as it's performance though.

Look at the links. The 950 uses the same amount of power and is actually slightly faster. Needless to say whether those benchmarks are accurate or not they are misleading. Just like this blender test can be made misleading (ie optimized for L1/L2 size or something like that)

 

[YBS1]

The links are in German, and I'll be honest....I'm lazy, yes even too lazy for a poorly translated Google translation. I assume the above is the RX 480 though, which I don't see in those benches you linked, are you looking at the 460? I can't imagine the 950 beating the 480 at anything performance wise.

 

[Fileas Corrales]

Interesting discussion of CPU benchmark, I like it!

 

[FIVR]

It won't beat the GTX 1060 in Time Spy

 

[YBS1]

It won't beat the GTX 1060 in Time Spy

Uh...What won't?
We were talking about the 950 above though.

 

[FIVR]

Oh. I got confused... I though this was the technical free-for-all forum where we just post about whatever technical thing we are thinking of at the moment.

 

[superstition]

Just like this blender test can be made misleading (ie optimized for L1/L2 size or something like that)

My understanding, based on skimming recent the posts here is that Blender has optimized code paths. That's not unusual. Prime95, for instance, uses SSE instead of AVX for Bulldozer to get better performance from it. XOP is used by DX12 as far as I know, too, when it's run on an AMD CPU that has it.

If a program is optimized for one architecture and not another that's much more misleading than having separate code to maximize performance from all important architectures — if you're talking about a real-world test like Blender.

 

[YBS1]

Oh. I got confused... I though this was the technical free-for-all forum where we just post about whatever technical thing we are thinking of at the moment.

It kinda went that way for a moment it seems.....haha

 

[DrMrLordX]

Correct.. And there's nothing to say it will match haswell/broadwell on average

. . . and the AotS benchmark seems to imply that it won't. Take that with a grain of salt.

Still an interesting result, though. Typically I would expect an AMD product to get hammered in something like Blender.

 

[Enigmoid]

The links are in German, and I'll be honest....I'm lazy, yes even too lazy for a poorly translated Google translation. I assume the above is the RX 480 though, which I don't see in those benches you linked, are you looking at the 460? I can't imagine the 950 beating the 480 at anything performance wise.

Yeah, I'm comparing it to the 460.

I'm not trying to crap on AMD, I like them and their products (will definitely buy a 460/470 when I upgrade my desktop) but this benchmark shenanigans has been going on far too long. I simply do not trust anything their marketing has to say after the polaris debacle. (I'm also amazed at how these forums jump up every time AMD markets something and predictably never comes to pass - look at some of the completely wild polaris predictions).

 

[DaveSimmons]

Yeah, I'm comparing it to the 460.

I'm not trying to crap on AMD, I like them and their products (will definitely buy a 460/470 when I upgrade my desktop) but this benchmark shenanigans has been going on far too long. I simply do not trust anything their marketing has to say after the polaris debacle. (I'm also amazed at how these forums jump up every time AMD markets something and predictably never comes to pass - look at some of the completely wild polaris predictions).

Agreed. I really want AMD to start competing again, since it hasn't made sense for me to even consider their CPUs since the Core 2 was released almost a decade ago. Before that, my AMD A64 X2s were great CPUs.

But I'm not going to trust pre-release teaser benchmarks and rumors from any of AMD, intel or nvidia. All of them have over-promised too much in the past.

 

[dark zero]

I think that you've got that backwards. Intel and NV would be "the empire", and AMD would be "the rebels".

Lol, but I was meaning by the color of the sabers... red are Sith and blue and green are the Force.

Seems that AMD is back on the game, but this time seems that Samsung decided to step up to not to see GloFo screwing their own process by them.

 

[itsmydamnation]

While this looks promising I will hold any applause. Lets not forget the last "live demo" debacle of polaris vs. the 950,

And these numbers were also demoed live. This claim turned out to be grossly misleading with polaris.

https://www.computerbase.de/2016-08...nahme-des-gesamtsystems-star-wars-battlefront
https://www.computerbase.de/2016-08/radeon-rx-460-test/3/#diagramm-star-wars-battlefront-1920-1080_2

I'm sorry AMD but you do not have the benefit of the doubt with me any longer. I will not believe any performance claims you make, even live demos, without seeing how the hardware performs by third parties.

Edit: I am not trying to threadcrap but simply saying that even live demos can be presented in a way that is not at all representative of what you will get in the real world. I think the blender test is nice but it is by all means inconclusive. Generally the safe bet, proved time and time again by Nvidia, Intel, and AMD is that the actual product never lives up to the marketing.

The big bit your choosing to ignore is that the 460 isn't a full P11, That comparison likely was. Infact your very own picture says "Polaris Architecture" please show me were it says 460. Yes it kind of sucks we dont have a full P11 to compare to given the target for ZEN APU we might never see a desktop version.

Now we know we are getting 8 core parts, the only question is clock speed. They demo'd @3ghz ( higher then the ES's given out) and have hinted at an even higher release clock. As i have said many times the key one im waiting for is spec int 403.gcc .