Discussion AMD Ryzen MegaThread! FAQ and Resources

010TheMaster010 · Feb 11, 2017

Placeholder for tracking!

[strike]Also, shouldn't this thread be created after the actual launch?[/strike]

Cheers!

Ditto!

juanrga · Feb 11, 2017

Found this online

Don't pay attention to the Pentium score. It is a single run made on a OC chip. However the other chips are average scores taken from different individual runs. For instance the Piledriver chip score is taken from averaging 566 samples. The margin of error is "low"

https://www.cpubenchmark.net/cpu.php?cpu=AMD+FX-8370+Eight-Core&id=2347

I will compare Zen to Piledriver Fx-8370. The RyZen ZD3406BAM88F4_38/34_Y is a quasi-final chip with 3.4GHz base and 3.8GHz turbo. It is commercially known as R7 1700X.

2046 * 4.3 / 3.8 = 2315

2315 / 1622 = 1.43

In this test Ryzen has 43% higher IPC than Piledriver.

jdwii · Feb 11, 2017

juanrga :

Man all this does is make me want to see legit benchmarks more. I mean if this is PURELY single threaded this would point towards a haswell IPC maybe a little lower then Haswell which is where we think it will be.

Gamer,Juan,jdwii,truegenius all think this. We could be wrong but hey either way i personally think it will be a success

juanrga · Feb 11, 2017

jdwii :

Well, 43% above Piledriver is more close to Sandy Bridge than Haswell.

Moreover a Sandy Bridge i5 @4.5GHz has a ST score of 2428, it would score about 2050 @ Zen clocks. However, it seems there are some problems with the memory subtest. Zen results are weak and could be a sign of either a non-final mobo or the IMC issue mentioned by sarinaide. In any case I don't think those memory subests would affect the score more than 5%.

illegaloperation · Feb 11, 2017

Supposedly some preliminary benchmarks for the Ryzen 7 1700X:

https://videocardz.com/65825/first-amd-ryzen-7-1700x-benchmarks-are-here

truegenius · Feb 11, 2017

jdwii said:
Man all this does is make me want to see legit benchmarks more. I mean if this is PURELY single threaded this would point towards a haswell IPC maybe a little lower then Haswell which is where we think it will be.

Gamer,Juan,jdwii, all think this. We could be wrong but hey either way i personally think it will be a success

^ hey count me too in that list because i also think that it will be around ivy ipc ±5%

subscribed •_•

sarinaide · Feb 11, 2017

The scores are lower in AVX related scores, and since AVX is a very limited adoption the Floating Point score is insanely high.

The Single thread Score and Floating Point math is high so I think it will score highly in Cinebench. I wouldn't read much into anything AVX driven, AMD will be weak in that department ie: IVY/Sandy level in AVX but this was already known by SpecINT

Turbos still not working so the 2045 would be a score at 3.4Ghz on that SKU. Which is rather strong.

sarinaide · Feb 11, 2017

My summary is that IPC is very strong but AVX is very weak, AMD's AVX was alwasys going to be bad, these parts are not designed for HPC.

sarinaide · Feb 11, 2017

jdwii :

juanrga :

Man all this does is make me want to see legit benchmarks more. I mean if this is PURELY single threaded this would point towards a haswell IPC maybe a little lower then Haswell which is where we think it will be.

Gamer,Juan,jdwii, all think this. We could be wrong but hey either way i personally think it will be a success

I did tell you it was faster than the 5960X

but yeah we were in that train faster than haswell but still a little off Broadwell, either way it is good. IMC though needs work big time.

sarinaide · Feb 11, 2017

juanrga :

jdwii :

Well, 43% above Piledriver is more close to Sandy Bridge than Haswell.

Moreover a Sandy Bridge i5 @4.5GHz has a ST score of 2428, it would score about 2050 @ Zen clocks. However, it seems there are some problems with the memory subtest. Zen results are weak and could be a sign of either a non-final mobo or the IMC issue mentioned by sarinaide. In any case I don't think those memory subests would affect the score more than 5%.

2600K is 3.4 to 3.8 so it wouldn't actually, it would score the same as it would on baseline score meaning it is faster than Sandy, but everyone already knew this.

sarinaide · Feb 11, 2017

A $400US chip tested on a A320 board? with memory timings the hint at gimping results.

This makes it look even better as hamstrung to the extreme, Ryzen still delivers good performance

A 4.5Ghz 2500K - 2428

Ryzen at 3.4- 2048

2048*4.5/3.4 = 2710.58

11% faster than Sandy bridge.

Per Cinebench ST

135 2600K at 3.8 turbo
140 5960X at 3.5 turbo

converted to 3Ghz

106
120

that is 11% difference

Ryzen=Haswell

There is margin to play here as the Ryzen was very held back.

sarinaide · Feb 11, 2017

Notice Ryzen is not running any turbo.

5960X is taking a clear beatdown

@3.3ghz = 1782.70
@3.5ghz = 1890.74

Ryzen @3.4 = 2048

Nope 1151 · Feb 11, 2017

Has anyone seen this yet? I am not sure if this faked or not:
https://videocardz.com/65825/first-amd-ryzen-7-1700x-benchmarks-are-here

8350rocks · Feb 11, 2017

juanrga :

jdwii :

Well, 43% above Piledriver is more close to Sandy Bridge than Haswell.

Moreover a Sandy Bridge i5 @4.5GHz has a ST score of 2428, it would score about 2050 @ Zen clocks. However, it seems there are some problems with the memory subtest. Zen results are weak and could be a sign of either a non-final mobo or the IMC issue mentioned by sarinaide. In any case I don't think those memory subests would affect the score more than 5%.

Chips that start with a 4XXX and fall within 100 points of the score of the Ryzen CPU are haswell chips, not SB.

sarinaide · Feb 11, 2017

8350rocks :

juanrga :

Chips that start with a 4XXX and fall within 100 points of the score of the Ryzen CPU are haswell chips, not SB.

FX 8370

@3.4Ghz = 1283 (1622*3.4/4.3)

Roughly 46% IPC gain

Sandy ~33-35% faster than Vishera

Zen 11-13% faster than sandybridge

Got it wrong by a lot

2500K at 3.4Ghz only scores 1834, just saying

Nope 1151 · Feb 11, 2017

Has anyone done a comparison to the CPC leaks? I am not (mathematically) competent enough for that.

8350rocks · Feb 11, 2017

Nope 1151 :

Those specific numbers put it squarely around haswell performance...in single thread...and apparently right around SL/KL when threaded, and it blows the doors off that 6800K.

Also, reading the original thread where these CPU Mark benchmarks came from on Anandtech, lots of the processors that are not Zen are overclocked by the estimates of many. The clockspeeds on these links are evidently only showing stock speeds, but not what the actual CPU speed is at currently. For example, a 5960X was shown beating a 6900K in several examples, which should not be accurate, additionally, they had one where a heavily overclocked FX9590 was ahead in physics score and prime numbers, which should also not be possible, as an 8c Ryzen CPU would have twice the FPUs.

Something is not completely right with this...

sarinaide · Feb 11, 2017

8350rocks :

Nope 1151 :

Those specific numbers put it squarely around haswell performance...in single thread...and apparently right around SL/KL when threaded, and it blows the doors off that 6800K.

Also, reading the original thread where these CPU Mark benchmarks came from on Anandtech, lots of the processors that are not Zen are overclocked by the estimates of many. The clockspeeds on these links are evidently only showing stock speeds, but not what the actual CPU speed is at currently. For example, a 5960X was shown beating a 6900K in several examples, which should not be accurate, additionally, they had one where a heavily overclocked FX9590 was ahead in physics score and prime numbers, which should also not be possible, as an 8c Ryzen CPU would have twice the FPUs.

Something is not completely right with this...

Yes Ryzen per version 9 trial on Passmark has no turbo score only 3.4Ghz meaning that it is faster marginally than Haswell

here is a better representation.

8350rocks · Feb 11, 2017

sarinaide :

Thanks for the baseline...

Using this knowledge...here we go.

Ryzen @ 3.4 no turbo = 2046

FX9590 @ 5.15 no turbo = 1925

2046 * 5.15/3.4 = 3099 (If only Ryzen could OC to 5.15! :ouch:

)

3099/1925=1.60 :heink:

So...now that we have clockspeed baselines...this benchmark shows that Ryzen on a clock per clock basis is 60% faster than Piledriver's most advanced version.

Since, we have some speculating that SB is ~40% faster than PD, with Ryzen being 60% faster than that...we are looking at Ryzen being ~15-20% faster than SB using very simple math.

Now, SB to IB was about 6%, and IB to haswell was about 10% so, SB to haswell ~16%.

My guesstimations are that we see right around haswell performance in singlethreaded benchmarks after all...

Grab a pinch of NaCl as that is mostly cocktail napkin math, but it should be in the ballpark on the low side...on the high side, haswell is only 5% behind skylake in IPC. If they really pulled off a coup, it would match skylake...I am not holding my breath...but I am quietly optimistic that benchmarks may be better than I initially thought.

EDIT: For grins and giggles...why not:

Note the 7700K @ 5.0 GHz. The extrapolated Ryzen score @ 5.0 GHz is 3008, where the 7700K comes out slightly less...

juanrga · Feb 11, 2017

sarinaide :

Turbo is working. It has been working since CPCHardware review and before...

juanrga · Feb 11, 2017

sarinaide :

Ryzen does 2046 at 3.8GHz. Therefore it is

2428 * 3.8 / 4.5 = 2050

Aka RyZen has about Sandy Bridge IPC.

juanrga · Feb 11, 2017

sarinaide :

If turbo was disabled I guess it would report the same frequency than in the base field. "N/A" surely means that the mobo (without XFR support) is having problems reading the parameters of the chip. Recall that RyZen has two single-core turbo frequencies, f_TMT, and F_TMax, and F_Tmax is not available on that mobo.

We can check that turbo was enabled. The ST score is obtained from single-thread versions of the floating point, string sorting, and data compression tests.

6900k vs RyZen
----------------------
FP: 12.5%
String: 10.4%
Compr: 17.7%
---------------------
Average: 13.5%

ST: 12.8%

If turbo was disabled, then the ST gap would be higher, about 15% higher, because the 6900k would be pushing clocks up to 3.7GHz in the ST gap. However, we can see that the gap is about the same in ST and MT, which implies that RyZen is working with turbo enabled. In fact

13.5 * (3.8 / 3.4) / (3.7 / 3.2) = 13%

which is virtually equivalent to the 12.8% measured in ST.

juanrga · Feb 11, 2017

8350rocks :

juanrga :

jdwii :

Well, 43% above Piledriver is more close to Sandy Bridge than Haswell.

Moreover a Sandy Bridge i5 @4.5GHz has a ST score of 2428, it would score about 2050 @ Zen clocks. However, it seems there are some problems with the memory subtest. Zen results are weak and could be a sign of either a non-final mobo or the IMC issue mentioned by sarinaide. In any case I don't think those memory subests would affect the score more than 5%.

Chips that start with a 4XXX and fall within 100 points of the score of the Ryzen CPU are haswell chips, not SB.

I was mentioning an i5-2500k, which is SB.

8350rocks · Feb 11, 2017

juanrga :

8350rocks :

juanrga :

Chips that start with a 4XXX and fall within 100 points of the score of the Ryzen CPU are haswell chips, not SB.

I was mentioning an i5-2500k, which is SB.

You mean the chip that has a CPU Mark of 1825 @ 3.4 GHz + turbo? In other words, 100 points lower clock per clock?

juanrga · Feb 11, 2017

Nope 1151 :

This Passmark bench puts 8C Ryzen 3.4/3.8GHz average multithreading performance somewhat between 6C and 8C Broadwell, and the former CPC leak did the same

Discussion AMD Ryzen MegaThread! FAQ and Resources

Honorable

Distinguished

Splendid

Distinguished

Honorable

Distinguished

Splendid

Splendid

Splendid

Splendid

Splendid

Splendid

Commendable

Distinguished

Splendid

Commendable

Distinguished

Splendid

Distinguished

Distinguished

Distinguished

Distinguished

Distinguished

Distinguished

Distinguished

Share this page