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

[CentroX]

Why is sm625 calling it a turd? At those clockspeeds thats good results by ryzen.

 

[.vodka]

He tried to deceive us in this post at first. He didn't tell us he was using overclocked baselines for the other processors. That G3258 is running at 4.4GHz, who knows about the other ones. Passmark reports CPUID string by default, not actual frequencies. See, having that information changes everything.

We pressure him, then he posts this, without the baselines. More graphs without any additional information apart from the CPUID string. After some more pressure, he then edits and posts this. [Archive.is before the mod edits, for posterity.] Of course, highly overclocked samples (I mean, well into the 4GHz range) vs a 3.4GHz Ryzen with turbo disabled, or very well having turbo enabled and Passmark not being able to read it because of unknown platform. Either way, it's quite impressive for Zen. In the end he did know the frequencies those samples were running at, and didn't include them in the graphs.

Why did he call it a turd at first? I don't know, I'd say he has an agenda. In the end he only managed to shoot himself in the foot with that first deceptive post. I might get a call out warning for this, but I think it's more than justified, all of this. All this drama could've been avoided by a well explained first post. You call out people when the BS is just too much to endure.


We have the baseline numbers so we can do our own comparisons vs stock processors, but Passmark's baseline search page is broken. Crap. I wonder if the benchmark's program inbuilt database search works... at least from the times I used performancetest it had that feature. I'll go download it.


edit: there you are, AMD ZD3406BAM88F4_38/34_Y

It's using a GTX 1080, 16GB DDR4, an MSI A320M PRO-VD (bummer, no X370) on W10 14393

edit2: Passmark knows how to read turbo VERY well athough it doesn't show such information when it builds the graphs. Have a look at my system vs this Ryzen baseline. It has correctly read my 4.5GHz turbo settings. We can be pretty sure those Intel baselines he chose are clocked at the speeds shown.

 

[inf64]

I'm barely reading this over my mobile phone and I'm enjoying so much. Entertainment level over 9000 .

It seems my initial estimate was a bit on the down side, Ryzen is faster than I expected.

 

[Rngwn]

edit: there you are, AMD ZD3406BAM88F4_38/34_Y

It's using a GTX 1080, 16GB DDR4, an MSI A320M PRO-VD (bummer, no X370) on W10 14393

Any way to know that the system was using dual-channel setup and not otherwise?

 

[.vodka]

It was a dual channel setup. The motherboard has two RAM slots and...

the baseline shows two 8GB sticks installed. I'll put up the individual memory results in a little while so you guys can see what the program is actually testing.


So, the individual tests are as follows (the results in the pictures are from my system):

Ryzen's scores from the baseline:
CPU mark: 15084
Integer: 39672 Mops/sec
Prime: 37 million primes /sec
Compression: 24723 KBytes/sec
Physics: 726 Frames/sec
CPU ST: 2046 Mops/sec
FP: 14807 Mops/sec
SSE: 717 Million Matrices /sec
Encryption: 3865 Mbytes/sec
Sorting: 15204 Thousand Strings/sec

Ryzen's results from the baseline:
Memory mark: 1855
Database: 78 Kops/sec
Memory read uncached: 14915 Mbytes/sec
Memory threaded: 34011 Mbytes/sec
Memory read cached: 28006 Mbytes/sec
Memory write: 7917 Mbytes/sec
Memory latency: 76 ns



Ryzen is running with DDR4 2400 17-17-17-39 probably 2T. Explains the poor results, those timings suck for that memory speed. Now we can either find baselines using stock Intel processors or you can download the benchmark, run it on a stock system and directly compare the results. You can't get any clearer than this on this leak.

 

[HurleyBird]

I'm going to take all of this with a grain of salt both ways. The absolutely terrible memory scores mean that there's a moderate amount of performance left on the table once that's fixed, but there's also no way to be sure of what clockspeed the Ryzen sample is running at. For all anyone knows the "automatic overclocking" feature might be activated, given that the processor is launching so soon.

But for sure, the initial signs are very promising.

 

[lolfail9001]

Ryzen's results here that would show in the boxes:
Memory mark: 1855
Database: 78 Kops/sec
Memory read uncached: 14915 Mbytes/sec
Memory threaded: 34011 Mbytes/sec
Memory read cached: 28006 Mbytes/sec
Memory write: 7917 Mbytes/sec
Memory latency: 76 ns

Holy cow, that is bad bad bad in writes and latency department.

For all anyone knows the "automatic overclocking" feature might be activated, given that the processor is launching so soon.

I am convinced single threaded passmark was certainly running at 3.8Ghz. Do not know about multi threaded tests though, but it should be enough to establish that rest of tests were not running on higher clock than that.

 

[.vodka]

I'm going to take all of this with a grain of salt both ways. The absolutely terrible memory scores mean that there's a moderate amount of performance left on the table once that's fixed, but there's also no way to be sure of what clockspeed the Ryzen sample is running at. For all anyone knows the "automatic overclocking" feature might be activated, given that the processor is launching so soon.

But for sure, the initial signs are very promising.

No automatic overclocking / XFR, the motherboard is using the A320 chipset which doesn't support overclocking along with A300. B350, X300 and X370 are the overclocking enabled chipsets.

So we know the definitive range of this sample is between 3.4 and 3.8GHz if turbo is enabled or working. It's not exact, but it's a small range.

The memory scores look bad, although I'd like to see what happens when you get some better memory in there, both speed and timings. Motherboards were the cause of the delay IIRC, BIOSes are being polished, maybe there's something more to extract out of the memory subsystem.

 

[Rngwn]

Here with i7 6700HQ bench with 2x 2133MT/s DDR4 15-15-15-36, for comparision's sake

Seems like the DDR4 is "worse" than vodka's DDR3 under passmark benches.

BTW, since the memory appears to be the culprit, the question now is, which of the CPU benchmarks are memory dependent?

 

[HurleyBird]

No automatic overclocking / XFR, the motherboard is using the A320 chipset which doesn't support overclocking along with A300. B350, X300 and X370 are the overclocking enabled chipsets.

I stand corrected. In that case, these results seem pretty promising indeed.

 

[inf64]

sm625 has gone missing. He unintentionally fuelled the hype train, trolled himself and set us all for a journey to infinity... and beyond

 

[lolfail9001]

sm625 has gone missing. He unintentionally fuelled the hype train, trolled himself and set us all for a journey to infinity... and beyond

Well, he managed to throw more actual coal into the hype loco than AMD over half a year, if we are honest. If that memory issue does not get resolved on launch, i can see it biting AMD in the arse in gaming benchmarks though.

 

[bjt2]

Yeah, if that's at 3.4GHz, then it's faster per-clock than Haswell (3.7GHz boost i5-4590), but only very slightly (1.8%).

Since BWE is 3-4% faster and SKL/KBL is another 3-4% faster, we can say that it's 5% slower than SKL/KBL...
Not bad...
Nothing that 200MHz more of clock can't compensate...

 

[bjt2]

edit2: Passmark knows how to read turbo VERY well athough it doesn't show such information when it builds the graphs. Have a look at my system vs this Ryzen baseline. It has correctly read my 4.5GHz turbo settings. We can be pretty sure those Intel baselines he chose are clocked at the speeds shown.

There is a problem with Ryzen caches: 16 L1? 16 L2? Something is broken...

 

[bjt2]

It was a dual channel setup. The motherboard has two RAM slots and...

the baseline shows two 8GB sticks installed. I'll put up the individual memory results in a little while so you guys can see what the program is actually testing.


So, the individual tests are as follows (the results in the pictures are from my system):

Ryzen's scores here that would show in the boxes:
CPU mark: 15084
Integer: 39672 Mops/sec
Prime: 37 million primes /sec
Compression: 24723 KBytes/sec
Physics: 726 Frames/sec
CPU ST: 2046 Mops/sec
FP: 14807 Mops/sec
SSE: 717 Million Matrices /sec
Encryption: 3865 Mbytes/sec
Sorting: 15204 Thousand Strings/sec

Ryzen's results here that would show in the boxes:
Memory mark: 1855
Database: 78 Kops/sec
Memory read uncached: 14915 Mbytes/sec
Memory threaded: 34011 Mbytes/sec
Memory read cached: 28006 Mbytes/sec
Memory write: 7917 Mbytes/sec
Memory latency: 76 ns



Ryzen is running with DDR4 2400 17-17-17-39 probably 2T. Explains the poor results, those timings suck for that memory speed. Now we can either find baselines using stock Intel processors or you can download the benchmark, run it on a stock system and directly compare the results. You can't get any clearer than this on this leak.

Bulldozer/Excavator have the NB at 2400MHz. L3 cache clocked at NB clock.
It seems Ryzen has the same configuration. But 5x the bandwidth was promised.
Since the bus is 4x, we can infer that NB clock in retail chips must be at least 3GHz (+25%, the difference to make 4x a 5x)
But this is an ES. Who knows what the NB clock is.
Another unknown for the memory benchmark.

 

[lolfail9001]

But this is an ES.

It is a QS, in fact, it's spec looks to match that of 1700x at glance.

EDIT: Hm, 16 caches is weird, indeed, lemme check real quick.

 

[bjt2]

It is a QS, in fact, it's spec looks to match that of 1700x at glance.

EDIT: Hm, 16 caches is weird, indeed, lemme check real quick.

Ok, but the MB? Looking at memory score, the BIOS seems to be not much ok...

 

[blublub]

Hmm, a 3.8 GHz Ryzen losing out against the 3.6 GHz i3 on ST is a very telling about its IPC.

Can't be real. Wen know Ryzen can beat 6900k in MT.
So to loose this badly in St but beat in MT, AMDs SMT implementation must be so much superior than Intel's that it just impossible.

 

[itsmydamnation]

Holy cow, that is bad bad bad in writes and latency department.

meh, write are @1:2 to to read, so not really a big deal. 76ns isn't bad its just not good, skylake with 2400mhz is around 52ns.

 

[cytg111]

Yeah, if that's at 3.4GHz, then it's faster per-clock than Haswell (3.7GHz boost i5-4590), but only very slightly (1.8%).

- Damn, how much money did we bet? I am pretty sure I put my car down on 5% slower IPC than broadwell..

 

[vissarix]

Can't be real. Wen know Ryzen can beat 6900k in MT.
So to loose this badly in St but beat in MT, AMDs SMT implementation must be so much superior than Intel's that it just impossible.

Who told you that Ryzen can beat the i7 6900k?
Amd showed a blend test without numbers or anything just said they finished the test about the same time...such a reliable benchmark..

everything outside Amd own benchmarks points Ryzen underperforming badly...Canard benchmarks, sandra sisoftware benchmarks etc

 

[itsmydamnation]

everything outside Amd own benchmarks points Ryzen underperforming badly...Canard benchmarks, sandra sisoftware benchmarks etc

Your a really comedian, you should do standup,
1. CPC was @ 3.15ghz and doesn't look bad at all, we have 3.6ghz coming
2. You ignore the the Ryzen h264 demo
3. You ignore the BF1 demo
4. even the passmark data doesn't look bad

Then you completely ignore price!!!! ROLFCOPTER to you my friend.

 

[The Stilt]

How can a product, which delivers more than it originally promised be underperforming?
Zen, like every other x86 design has it's weaknesses and compromises but still?

 

[flash-gordon]

Badly? Au contraire my friend! These Passmark results are on the edge of awesome for the rumored prices!

sm625, I bet you're drunk when you posted this, what a confusion you've made!

 

[Glo.]

Intel Damage Control efforts, starts in 3, 2, 1...

Ryzen appears to be MUCH better than EVERYBODY has anticipated. Keller, and AMD team did engineering marvel. We are looking at something that is not very often seen in this industry. Hats off, AMD.