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

[OrangeKhrush]

http://wccftech.com/amd-ryzen-5-1600x-cpu-benchmark-leak/

I tested my htpc, i5 4460 with the same clocks and Ryzen gets 200pts more in single thread.

ipc is better than Haswell, Passmark my 4460 and 3ghz locked 5960x are 1800-2000 to Ryzen 2050@3.4 it is more and more commonly showing that. A good source who tests for Asus gives it about 5% to broadwell which is very likely. He also told me about the IMC limitations 4 weeks ago

 

[OrangeKhrush]

Has anyone posted this link before?
http://www.overclock.net/t/1623292/lets-talk-about-a-ryzen-es

The guy seems to have Ryzen (check pics of his setup in the topic).
In his own words:
"
- IPC is at the least Ivy Bridge-E and higher
- SMT for Ryzen is more efficient then Intel's HyperThreading
- Ryzen has no cold Bug
- Cinebench R15 hits 145 single thread @ 3.4GHz on ES, earlier models hit 130-140, retails should hit 140-150
- Most Ryzen ES samples hit 4.3-4.5GHz MAX on Air with all core enabled
- Intel is testing out Skylake-X, and beats out current 6950X with 8C/16T because it can hit higher clocks."

So lower end of the IPC scale is IVB-E, upper end is SKL.
No cold bug, OCing on par with BDW-E if not better.
Cinebench R15 score at 3.4Ghz seems a bit high at 145pts. SKL gets 37% boost from SMT, if Zen gets the same(similar) boost then 3.6Ghz base clock model(top SKU) should score 1682pts provided clocks do not go above 3.6Ghz. That is 9% higher than stock 6900K that runs @ 3.5Ghz in this benchmark.

Hype train has been shifted in the next gear. Nobody can get aboard since it is going too damn fast.


edit:
Same user posted this also: "The 179.99 Dollar R3 1300 is going to be 5-8% slower than a 7700K and be 150 dollars cheaper". <- I assume he is talking about OCing on air this sucker to 4.3Ghz? Has to be since base and Turbo of this poor chip are very low.

Hype level over 9000!!!

I got 150 off the ol 1188 leak so its all so close and IVY is not that at 3.4, it is at 3.9 though

 

[CentroX]

So intel is planning to take the crown with skylake-X but it is 6 months away right? So amd will have time to take marketshare.

 

[PhonakV30]

No.
It just indicates that the C6 states and CPB (turbo) are enabled.

wow , Is that even Possible? How about SKL/KB ? Do they have ?

 

[OrangeKhrush]

i7 3770K (3.5-3.9) 143/708

In single thread Cinebench always takes max turbo which is 3.9Ghz

143 * 3.4 / 3.9 = 124.6.

what i love about cinebench is that it is perfectly linear, every 100mhz increases single thread by the same ratio, and likewise with multi threaded scores all goes up by the exact same ratio

 

[The Stilt]

Well, considering that presumably Dolphin benefits from FMA, that could the explanation.

No FMA instructions at least in the public 5.0 binaries.

Does dolphin use avx(2)?

5.0 has some 128-bit AVX code in it.

 

[JoeRambo]

Jesus, that 6 core from AMD is tasty. Generous 16MB of L3 and ST perf is nice. If they can somehow hit ~4.4Ghz when OCed they will sell truckload of them if leaked prices are correct.

 

[The Stilt]

wow , Is that even Possible? How about SKL/KB ? Do they have ?

I meant that these features are enabled. Obviously they are not active at the same time, on the same cores.
C6 = power gating.

C6 is a hard requirement for CPB to work.

 

[CatMerc]

about that...

Why would it *require* tight timings?

 

[dark zero]

Oh boy... So Zen is Broadwell or even Skylake levels?

Intel will get finally a taste of their medicine (IPC improvement) in desktop at last! Hope that AMD gets some marketshare in order to make Intel react and stop sleeps on their glory days.

 

[Valantar]

Jesus, that 6 core from AMD is tasty. Generous 16MB of L3 and ST perf is nice. If they can somehow hit ~4.4Ghz when OCed they will sell truckload of them if leaked prices are correct.

My thoughts exactly. Even if they average around ~4-4.1GHz on non-golden samples, I'll still probably get one. But if they reach ~4.4-4.6? My, oh my. They should sell like hotcakes. Let's just hope we don't get a mining craze-style demand-outstripping-supply situation with chips selling at 2x MSRP. That would destroy this, and hurt AMD badly.

 

[coercitiv]

Why would it *require* tight timings?

In the context of the thread, it would allegedly require tight timings to get optimal performance. Tight timings are not a mandatory requirement ofc.

 

[CatMerc]

In the context of the thread, it would allegedly require tight timings to get optimal performance. Tight timings are not a mandatory requirement ofc.

I get that, I'm just wondering why Intel wouldn't require said timings for optimal performance.

 

[Dygaza]

OS language changes from english to chinese between those 2 cpu-id shots. Also cpuid font and frames are different. Most likely different machine.

 

[coercitiv]

I get that, I'm just wondering why Intel wouldn't require said timings for optimal performance.

Blast from the past. Take that info as an example of different solutions to a problem, add current AMD marketing talk about Neural Net and Smart Prefetch working together to increase performance (responsible for ~25% of Zen performance uplift), then consider how this might help hide or at least partially compensate for an alleged weakness in Zen IMC.

Since Intel's and AMD implementations are different, they may have different weakpoints and different scaling with memory timings.

[EDIT] on a second note, when SKL launched it was also presented as a CPU that benefits from "faster" RAM, with the only exception that this time no reviewer bothered to isolate latency from bandwidth - they simply used kits that optimized both (ex: 3000 CL15 vs. 2133 CL15)

 

[zinfamous]

My thoughts exactly. Even if they average around ~4-4.1GHz on non-golden samples, I'll still probably get one. But if they reach ~4.4-4.6? My, oh my. They should sell like hotcakes. Let's just hope we don't get a mining craze-style demand-outstripping-supply situation with chips selling at 2x MSRP. That would destroy this, and hurt AMD badly.

This is probably why AMD again re-worked their wafer agreement to pay the GloFo penalty to use other fabs. Granted, If they are producing wafers at different fabs, does that increase the silicon lottery variable?

Either way, it looks like they were planning to utilize all available resources to pump out a metric shitload of Zen dies for this initial release and going forward this year (I would assume that maybe the Naples cores will be localized to one fab, while Ryzen is localized to another?) and are confident enough in what they have to pay whatever penalty to make it happen.

 

[dahorns]

Same user posted this also: "The 179.99 Dollar R3 1300 is going to be 5-8% slower than a 7700K and be 150 dollars cheaper". <- I assume he is talking about OCing on air this sucker to 4.3Ghz? Has to be since base and Turbo of this poor chip are very low.

He must mean the R5? I don't see an R3 1300.

 

[Glo.]

I would like to point out something that nobody seems to notice, about the 6C/12T CPU in that benchmarks.

ZD3301BBM6IF4_37/33_Y

Now lets look at what Eng Sample numbers mean, and focus on TDP package.

The 6C/12T CPU here has TDP indicator as: BB = 65W(!).

Every other CPU Eng sample, especially with ZD is final version that we will see in retail.
The CPU benchmarked does not have XFR. It is not 1600X. It does not cost 259$.

 

[Gideon]

Every other CPU Eng sample, especially with ZD is final version that we will see in retail.
The CPU benchmarked does not have XFR. It is not 1600X. It does not cost 259$.

Where do you read, that it doesn't have XFR? (don't see any reference to it on the codename decoding picture).

EDIT: Ahh, now i get id, no XFR CPU has 65W TDP

Hopefully they do indeed have a Vanilla Ryzen R5 1600

 

[Glo.]

Where do you read, that it doesn't have XFR? (don't see any reference to it on the codename decoding picture).

EDIT: Ahh, now i get id, no XFR CPU has 65W TDP

Hopefully they do indeed have a Vanilla Ryzen R5 1600

Here you have it:

 

[bjt2]

wow , Is that even Possible? How about SKL/KB ? Do they have ?

AMD since Bristol Ridge has DLDO that allows to avoid power gating altogether (CC6) entering the core a low power state called Pmin, with extremely low Vcore (more like 0.6-0.7V on 28nm, so i think about 0.5 on 14nm, but 0.375V can be possible) and retaining of L1 and L2, so no flushing and refill. This has very low overhead and entry/exit latency.

EDIT: on 28nm the minimum Vcore required to retain caches is 0.55V according to the papers, but they used 0.6-0.7 to have a margin. I suppose that on 14nm can be lower...

 

[sm625]

This was uploaded on February 13... the first baseline #771904 was uploaded on February 10... hmm.. interesting..

Something else that's interesting is that baseline #771904 no longer shows up in PerformanceTest 9.0 when you search for it. They must be manually filtering it out because the adjacent baselines do appear in searches. Kind of reminds me of shadowbanning on reddit/twitter. It's there, but its not there. I still have it bookmarked. I can still make charts with it. But cant search for it. lol. It's got me wondering how many other baselines are hidden away....

 

[sm625]

AMD since Bristol Ridge has DLDO that allows to avoid power gating altogether (CC6) entering the core a low power state called Pmin, with extremely low Vcore (more like 0.6-0.7V on 28nm, so i think about 0.5 on 14nm, but 0.375V can be possible) and retaining of L1 and L2, so no flushing and refill. This has very low overhead and entry/exit latency.

A DLDO with bristol ridges? With low entry/exit latency? I'm surprised they dont sell better.

 

[.vodka]

Something else that's interesting is that baseline #771904 no longer shows up in PerformanceTest 9.0 when you search for it. They must be manually filtering it out because the adjacent baselines do appear in searches. Kind of reminds me of shadowbanning on reddit/twitter. It's there, but its not there. I still have it bookmarked. I can still make charts with it. But cant search for it. lol. It's got me wondering how many other baselines are hidden away....

Noticed that too, I couldn't get any results for #771904 anymore (clean installed Windows a day ago, so that baseline was lost). As you say, it's still there, someone at reddit managed to get the link on their website.

You can be pretty sure this new baseline will get unlisted in a few days or so. It's an interesting data point nonetheless, showing progression from an engineering sample to final silicon in #771904

 

[Face2Face]

That CPU-Z bench is really impressive, especially if that's a 65w part.

Ryzen 6/12 3.3 - 3.7Ghz

ST Score: 1888
MT Score: 12544

My 5820K @ 4.25GHz

ST: 1929
MT Score: 11872

Amazing if true. I was planning on buying the cheapest 6/12 Ryzen CPU for my Plex server, but man; this is making me want to replace my Main rig with the 8/16 variant...

I'll run the test at stock speeds when I get the chance. Probably tonight