• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

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

Page 144 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Status
Not open for further replies.
What's the likelihood of Chinese forums getting access to Ryzen before anyone else? I understand it with GPUs since cards are mostly manufactured in China. But CPUs are a bit different. They get fabbed in US, packaged and characterized in Malaysia. The odds of someone getting their hands on a Ryzen sample in China is the same as everywhere else imo. These aren't CPUs going into OEM laptops either, and server parts still have ways to go.
 
So Sweepr, you believe that Ryzen at unknown clock speed is 13% slower than SandyBridge class core. That is not at all odd/strange to you? While at the same time same clocked Ryzen achieves parity with Broadwell-E at the same clock(roughly) running two completely different benchmarks?

We are talking about 35-40% per clock difference here, mind you. Red flag not showing up yet?
 
inf64 said:
So Sweepr, you believe that Ryzen at unknown clock speed is 13% slower than SandyBridge class core. That is not at all odd/strange to you? While at the same time same clocked Ryzen achieves parity with Broadwell-E at the same clock(roughly) running two completely different benchmarks?
Click to expand...

Just leave him alone with his circlejerk. If you don't his posts just ignore them since they are usually worthless.
 
Glo. said:
I thought it was supposed to be around GTX 1070 at that clock speed.
Click to expand...
Yep, that's what the OC.net rumor said.
CatMerc said:
I distinctly remember quite a few Chinese "leaks" that turned out to be bollocks in the RX 480 era. The clockspeed leak was Western in origin, but the R9 Nano perf at stock clocks that was touted by WCCFTech was oriental.
Click to expand...
I don't remember a single one except that fake Lightning rx480 hint. Also, the 3DMark result for rx480 itself was accurate.

Anyways, i wait for any sort of verification now.
 
Intel Core i7-6900K ( This Link )
ST=185 , MT=1578
MT/ST = 8.52
1578/16 threads = 98.625
----------------------------------------------
Intel Core i7-5960X
ST = 177 , MT = 1392
MT/ST = 7.864
1392/16 threads = 87
----------------------------------------------
Intel Core i7-4960x
ST=144 , MT=1078
MT/ST = 7.486
1078/12 threads = 89.83
----------------------------------------------
AMD FX-8350
ST = 96 , MT = 637
MT/ST = 6.635
637/8 threads = 79.625
----------------------------------------------
AMD Phenom 1100T
ST= 93 , MT =501
MT/ST = 5.387
501/6 threads = 83.5
----------------------------------------------
AMD Ryzen
ST = ? , MT = 1188
If Zen's IPC Is higher than 40% Steamroller's IPC ( let's say Steamroller is 10% faster than Piledriver )
then ST= 96*1.50 = 144
MT/ST = 8.25
1188/16 threads = 74.25 (!)

ratio MT / Threads of Ryzen is worse than Previous Gen.Could It fake ? It's up to you.
 
krumme said:
Amd have pointed out they target the cloud server market.
I asume such a market will imply code loads that varies and will expose the branch predictor?
What is important on the cloud server market in ralation to cpu perf?
You dont get a cpu with good branch predictor just because you want it. But neither do you get fpu perf in blender similar to bwe. If anything 3 months ago i would rate the latter as more unlikely. Good perf on fpu heavy stuff is needed but certainly not to the degree we see now when your competitor have such fpu monsters and is expected to take the market where fpu plays the major part.
Click to expand...
It is yet unclear whether the branch predictor is poor. While being Broadwell level is an unlikely feat, I highly doubt it is not far better than Excavator. It is one of the main heels they tried tackling with this design.

I also doubt that either the IEU/FPU are lopsided. It's supposed to be a balanced design, unlike BD.

40% IPC is official, and that is Sandybridge Ivybridge level. I'm sticking with that.

It is EASY to code and compile a bench for a specific uarchs strengths or weakenesses. That's the trouble with closed custom benches. Not sure why people have trouble understanding this.

IF AMD wanted to avoid this lack of scientific validity, they could have ran standardized Cinebench or POVRay benchmarks instead. The fact they didn't, puts many on alert.

There are also specific bottlenecks with every uarch in coding that can make it CHOKE or underperform. Eg. Looping ss fp instr followed by ps fp instr gets close to half the FLOPs compared to 3/4 ss fp followed by 3/4 ps fp on Broadwell. Or using an output to fp calc as input to different precision fp calc on Bobcat. Or mixing different latency instr within the same execution pipe with K10.

It's just not as simplistic as 'Hey! New unverified benchmark. Run. Result?'

BTW - Enterprise cloud runs every sort of load out there.

Sent from HTC 10
(Opinions are own)
 
behrouz said:
Intel Core i7-6900K ( This Link )
ST=185 , MT=1578
MT/ST = 8.52
1578/16 threads = 98.625
----------------------------------------------
Intel Core i7-5960X
ST = 177 , MT = 1392
MT/ST = 7.864
1392/16 threads = 87
----------------------------------------------
Intel Core i7-4960x
ST=144 , MT=1078
MT/ST = 7.486
1078/12 threads = 89.83
----------------------------------------------
AMD FX-8350
ST = 96 , MT = 637
MT/ST = 6.635
637/8 threads = 79.625
----------------------------------------------
AMD Phenom 1100T
ST= 93 , MT =501
MT/ST = 5.387
501/6 threads = 83.5
----------------------------------------------
AMD Ryzen
ST = ? , MT = 1188
If Zen's IPC Is higher than 40% Steamroller's IPC ( let's say Steamroller is 10% faster than Piledriver )
then ST= 96*1.50 = 144
MT/ST = 8.25
1188/16 threads = 74.25 (!)

ratio MT / Threads of Ryzen is worse than Previous Gen.Could It fake ? It's up to you.
Click to expand...

I don't think AMD would do an event like Horizon to show off its new CPU before launch if the actually made a "ZEN-Dozer".

Additionally the performance discrepancy between the shown Blender/Handbrake and those leaks is just too large for the same CPU
 
inf64 said:
So either the leak is a fake or Ryzen in the leak was running at way below the Broadwell-E MT boost clocks(3.4-3.5Ghz on 6900K). This is of course my opinion on the matter.
Click to expand...

Purely for naming speculation:
uc


EDIT: My pure "guess" (who knows, can hardly make out what characters those are) is 8c/16t Zen @ 3.1 base 3.5 turbo, because 2D2801A2M88E4 had 400 MHz base > turbo clock?

Haswell-E 5960X scores about 1350 points at stock, which would put this result 12% behind the 5960X in Cinebench R15
 
Last edited:
KTE said:
It is yet unclear whether the branch predictor is poor. While being Broadwell level is an unlikely feat, I highly doubt it is not far better than Excavator. It is one of the main heels they tried tackling with this design.

I also doubt that either the IEU/FPU are lopsided. It's supposed to be a balanced design, unlike BD.

40% IPC is official, and that is Sandybridge Ivybridge level. I'm sticking with that.

It is EASY to code and compile a bench for a specific uarchs strengths or weakenesses. That's the trouble with closed custom benches. Not sure why people have trouble understanding this.

IF AMD wanted to avoid this lack of scientific validity, they could have ran standardized Cinebench or POVRay benchmarks instead. The fact they didn't, puts many on alert.

There are also specific bottlenecks with every uarch in coding that can make it CHOKE or underperform. Eg. Looping ss fp instr followed by ps fp instr gets close to half the FLOPs compared to 3/4 ss fp followed by 3/4 ps fp on Broadwell. Or using an output to fp calc as input to different precision fp calc on Bobcat. Or mixing different latency instr within the same execution pipe with K10.

It's just not as simplistic as 'Hey! New unverified benchmark. Run. Result?'

BTW - Enterprise cloud runs every sort of load out there.

Sent from HTC 10
(Opinions are own)
Click to expand...

dude , You're doing 100% wrong If you do it! If you're Producer You can't allow "accurate bench" to be leaked.AMD is in critical stats Ofc they won't reveal many thing.because of a dirty Image , AMD is not going to release more info!
 
behrouz said:
Intel Core i7-6900K ( This Link )
ST=185 , MT=1578
MT/ST = 8.52
1578/16 threads = 98.625
----------------------------------------------
Intel Core i7-5960X
ST = 177 , MT = 1392
MT/ST = 7.864
1392/16 threads = 87
----------------------------------------------
Intel Core i7-4960x
ST=144 , MT=1078
MT/ST = 7.486
1078/12 threads = 89.83
----------------------------------------------
AMD FX-8350
ST = 96 , MT = 637
MT/ST = 6.635
637/8 threads = 79.625
----------------------------------------------
AMD Phenom 1100T
ST= 93 , MT =501
MT/ST = 5.387
501/6 threads = 83.5
----------------------------------------------
AMD Ryzen
ST = ? , MT = 1188
If Zen's IPC Is higher than 40% Steamroller's IPC ( let's say Steamroller is 10% faster than Piledriver )
then ST= 96*1.50 = 144
MT/ST = 8.25
1188/16 threads = 74.25 (!)

ratio MT / Threads of Ryzen is worse than Previous Gen.Could It fake ? It's up to you.
Click to expand...

You're expecting ~4.2GHz boost for 8C/16T Ryzen?
Excavator at 3.4GHz scores around 82 points in R15 ST, without CMT.
 
Markfw said:

I am posting as a moderator, when I say, this is constructive input, however, I don't like the term "clickbait"
Markfw
Anandtech Moderator.
Click to expand...
I will like to be specific and mention that by clickbait i mean the article on the site i will not mention by name. Not sweepr post. Just to be clear.
I say it because building articles like that is a common business practice that happens more and more and imo does nothing good except for a short time for their owner or the hordes of seo and quick content builders.

I just used the term because thats what people working with the stuff calls it because it frames what is done and what metology is used in writing, keyword density, headlines and so on.

The problem is invading us.
When i seek "blender cinebench r15" on google now i get the article bottom page 1. Its effective stuff. Google know people is clicking it and can measure they actually read it.
 
sirmo said:
What's the likelihood of Chinese forums getting access to Ryzen before anyone else? I understand it with GPUs since cards are mostly manufactured in China. But CPUs are a bit different. They get fabbed in US, packaged and characterized in Malaysia. The odds of someone getting their hands on a Ryzen sample in China is the same as everywhere else imo. These aren't CPUs going into OEM laptops either, and server parts still have ways to go.
Click to expand...
The motherboard makers need chips to test their products. The most likely source of leaks at this point is MB maker employees leaking things.

They have ES chips, but one thing of note is that the chips they have don't necessarily have to be competitively clocked.
 
Is there any evidence this is even linked to Summit Ridge other than from a forum poster saying so? Where is this forum post? Has anyone translated it? This reeks of trolling, leaving out so much information, unexpectedly lower scores...

Sure did stoke the fire of conversation though! The Intel fans seem to be more talkative suddenly about Zen now too 😀
 
Doom2pro said:
Is there any evidence this is even linked to Summit Ridge other than from a forum poster saying so? Where is this forum post? Has anyone translated it? This reeks of trolling, leaving out so much information, unexpectedly lower scores...

Sure did stoke the fire of conversation though! The Intel fans seem to be more talkative suddenly about Zen now too 😀
Click to expand...
Like I said, it seems like the CPU name matches AMD's Engineering Sample scheme.
Whether the result is real or fake I'm not going to speculate on.
But, the score is 12% slower than the 5960X (~1350) and about 23% slower than 6900K (~1550)

Nobody has any idea what speed the CPU was actually running or anything else, but it's possible 3.1/3.5 or 3.4/3.5
 
Tuna-Fish said:
The motherboard makers need chips to test their products. The most likely source of leaks at this point is MB maker employees leaking things.

They have ES chips, but one thing of note is that the chips they have don't necessarily have to be competitively clocked.
Click to expand...
That's true.. duh didn't think of that.
 
We won't know the CPU's worth until a retail sample has been OC'd and benched in watchdogs 2 and BF1 by a review site who doesn't just LOVE trolling people with GPU limited CPU tests.
I can't wait to see all the Zen CPU reviews. Benchmarks will look like this:

Zen vs "others"
1440p & 4K
Max details
GTX 1060 3gb
 
behrouz said:
Intel Core i7-6900K ( This Link )
ST=185 , MT=1578
MT/ST = 8.52
1578/16 threads = 98.625
----------------------------------------------
Intel Core i7-5960X
ST = 177 , MT = 1392
MT/ST = 7.864
1392/16 threads = 87
----------------------------------------------
Intel Core i7-4960x
ST=144 , MT=1078
MT/ST = 7.486
1078/12 threads = 89.83
----------------------------------------------
AMD FX-8350
ST = 96 , MT = 637
MT/ST = 6.635
637/8 threads = 79.625
----------------------------------------------
AMD Phenom 1100T
ST= 93 , MT =501
MT/ST = 5.387
501/6 threads = 83.5
----------------------------------------------
AMD Ryzen
ST = ? , MT = 1188
If Zen's IPC Is higher than 40% Steamroller's IPC ( let's say Steamroller is 10% faster than Piledriver )
then ST= 96*1.50 = 144
MT/ST = 8.25
1188/16 threads = 74.25 (!)

ratio MT / Threads of Ryzen is worse than Previous Gen.Could It fake ? It's up to you.
Click to expand...

No it's not worse, actually it lines up perfectly with 40% increase over excavator: 1188/8.25(*)=144, well over any phenom or -dozer in IPC if it was running at ~3GHz.
*This assuming single/multi ratio the same as broadwell 8 core, phenom isn't valid as you use it because it's 6 physical cores so 6 real threads=~6 ratio, SMT gains are less than a real core worth here.
 
KTE said:
It is yet unclear whether the branch predictor is poor. While being Broadwell level is an unlikely feat, I highly doubt it is not far better than Excavator. It is one of the main heels they tried tackling with this design.

I also doubt that either the IEU/FPU are lopsided. It's supposed to be a balanced design, unlike BD.

40% IPC is official, and that is Sandybridge Ivybridge level. I'm sticking with that.

It is EASY to code and compile a bench for a specific uarchs strengths or weakenesses. That's the trouble with closed custom benches. Not sure why people have trouble understanding this.

IF AMD wanted to avoid this lack of scientific validity, they could have ran standardized Cinebench or POVRay benchmarks instead. The fact they didn't, puts many on alert.

There are also specific bottlenecks with every uarch in coding that can make it CHOKE or underperform. Eg. Looping ss fp instr followed by ps fp instr gets close to half the FLOPs compared to 3/4 ss fp followed by 3/4 ps fp on Broadwell. Or using an output to fp calc as input to different precision fp calc on Bobcat. Or mixing different latency instr within the same execution pipe with K10.

It's just not as simplistic as 'Hey! New unverified benchmark. Run. Result?'

BTW - Enterprise cloud runs every sort of load out there.

Sent from HTC 10
(Opinions are own)
Click to expand...

Isnt it a difficult task to code and compile for bd so it gets fast in blender and handbrake without also beeing fast on bwe ? 😉 lol - but yeaa i know where you are going and i understand it.

As for the IPC its now official they have exceeded the 40%. And no then its not 42% but i would say a meaningfull number is 45% up.
Unfortunately we dont know what workload they use and thats probably what gives me most concern.
Seconly they have hyped specifically the branch predictor. I would gladly give a beer if its not a good deal better than xv. I think it will be just near bwe levels because of all the hype.
And because they need it for the cloud market.

Now as you say there is weak points and imo it would be far more interesting speculating what that is instead of asuming the branch predictor is bad. Its simply far more likely there is other weak parts.
Besides from the power consumption that didnt look so impressive to me the gaming benchmark was also trailing the bwe by a good deal. If we let the branchpredictor sleep for a moment what could be causes of that?
 
krumme said:
...Besides from the power consumption that didnt look so impressive to me...
Click to expand...

http://www.bitsandchips.it/english/...her-than-usual-vcore-during-new-horizon-event

"According to our sources, in order to avoid a crash during RyZen presentation, AMD has set up a higher than usual vCore. This explain the high full load power consumption of RyZen (95W TDP) compared to full load power consumption of the i7-6900K (Also Turbo features were disabled in RyZen).

However, despite the power consumption recorded, RyZen was abundantly under the TDP limit of 95W. This means that RyZen could be clocked higher, just like Lisa Su said."
 
BeepBeep2 said:
While I can't speculate about the CB R15 cropped screen score or its legitimacy or what CPU, the ES naming convention on quick glance seems to fit the naming convention (in terms of length) for the leaked Geekbench 32 core Zeppelin chip. There isn't enough there to be able to tell what any of the characters are though.
Click to expand...
My guess: 2D3101A2M88E4_35/31_N

Edit: Now with pic:
w7JUNXD.jpg
 
Last edited:
Status
Not open for further replies.

TRENDING THREADS

Back
Top