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

[The Stilt]

You forgot that GTX 1050 Ti will be locked by no 6 pin connector. It can be bottlenecked by this factor, not the process.

Except the GTX 1050 Ti has a 6-pin connector, since it is a 100W TDP rated card 😉

 

[Dresdenboy]

Last time I checked GB4 doesn't correctly report boost clocks.

GB3 for XV correctly reports 0.5x boost clock. GB4 might still use the same code.

 

[inf64]

Yes it's not there, I searched for all Oct 2 results, not available. Probably a fake from a random user.

Looks like all Zeppelin results are gone from DB.

Edit:

For those who speculated that leaked early GB results were done at ~1Ghz:

Old ST result 984pts (@1Ghz?)
New result was 3078pts at presumably full Turbo clock of 2.9Ghz.

3078pts x 1Ghz / 2.9Ghz = 1061pts or pretty much what the first leak showed 🙂.

 

[lolfail9001]

Looks like all Zeppelin results are gone from DB.

Edit:

For those who speculated that leaked early GB results were done at ~1Ghz:

Old ST result 984pts (@1Ghz?)
New result was 3078pts at presumably full Turbo clock of 2.9Ghz.

3078pts x 1Ghz / 2.9Ghz = 1061pts or pretty much what the first leak showed 🙂.

Old ST result was 948, so "1061" is not "pretty much what the first leak showed".

If the "1Ghz" hypothesis is correct, it means it should be about 15% slower per clock than 6700k. Well, that is actually good. Not good enough to actually brag about it, but good enough to be useful for 3 months before Skylake-X comes out.

 

[Glo.]

Old ST result was 948, so "1061" is not "pretty much what the first leak showed".

If the "1Ghz" hypothesis is correct, it means it should be about 15% slower per clock than 6700k. Well, that is actually good. Not good enough to actually brag about it, but good enough to be useful for 3 months before Skylake-X comes out.

http://wccftech.com/amd-zen-naples-soc-benchmarks/
This is the first score for this SoC, that appeared in the internet.

 

[inf64]

Old ST result was 948, so "1061" is not "pretty much what the first leak showed".

If the "1Ghz" hypothesis is correct, it means it should be about 15% slower per clock than 6700k. Well, that is actually good. Not good enough to actually brag about it, but good enough to be useful for 3 months before Skylake-X comes out.

Difference is just 7% and can be attributed to pretty much anything on a platform like this.
My Haswell at 4.2Ghz scores 4060pts, just for reference. At 2.9Ghz it would score around 2800pts.

edit:
As shown above the score fluctuated to 1141pts.
edit2 : the score of 1141 was achieved on 64bit version of GB4 so not directly comparable.

Here is Skylake 6700K vs my 4690K @ 4.2Ghz in 32bit GB3:
http://browser.primatelabs.com/geekbench3/compare/8082269?baseline=8083536

Minor difference in ST, pretty much below 5%.

 

[mikk]

Here is Skylake 6700K vs my 4690K @ 4.2Ghz in 32bit GB3:
http://browser.primatelabs.com/geekbench3/compare/8082269?baseline=8083536

Minor difference in ST, pretty much below 5%.

Are you sure this is at 4.2 Ghz? My i7-6700k at 4.0 Ghz already scores higher than this: http://browser.primatelabs.com/geekbench3/8083540

 

[inf64]

Are you sure this is at 4.2 Ghz? My i7-6700k at 4.0 Ghz already scores higher than this: http://browser.primatelabs.com/geekbench3/8083540

I'm sure my i5 is at 4.2Ghz. I have no idea what speed that 6700K was running at since I just picked one that had a score that was matching the average scores for this chip. Your chip is 5% and 3% faster in ST and MT respectively (vs the one I used). You are scoring 7% higher than me in GB3 ST. I am running 42x100 setup with 2x8GB DDR3-2400Mhz.

edit:

Here is the GB3 result with the same version as yours (I used the older one):
http://browser.primatelabs.com/geekbench3/8083556

 

[cdimauro]

Division by what? You don't know MT scaling, you don't know SMT scaling and without those, you can't divide.

You were asked a different question: since you consider the test was indicative of IPC as the appropriate values are one division away, please develop on this further and explain how division is to be applied. Otherwise having people "read up" on what IPC means is hardly helpful for the purpose of this conversation.

We know that:

IPC = the average number of instructions executed for each clock cycle. Per processor. Usually it's referred to a single processor core, and that's what I used. But even if you decide to take all instructions executed by all cores, at the end it doesn't change the overall result; see below.

ExecutedInstructions = Time * Frequency * Cores * IPC ; Time in seconds

But we have some fixed values for Frequency (3Ghz) and Cores (8), both for Zen and Broadwell-E.

So we leave them as they are, and only substitute the different values for the two processors:

ExecutedInstructions.Zen = Time.Zen * Frequency * Cores * IPC.Zen
ExecutedInstructions.BDW = Time.BDW * Frequency * Cores * IPC.BDW

Since the application used is exactly the same, as well as the executed job, we can roughly assume that the number of executed instructions is the same in both cases. Here I'm assuming that there's no specific code path for the two processors: only one, the same, is used for both.

So:

ExecutedInstructions = Time.Zen * Frequency * Cores * IPC.Zen
ExecutedInstructions = Time.BDW * Frequency * Cores * IPC.BDW

hence:

Time.Zen * Frequency * Cores * IPC.Zen = Time.BDW * Frequency * Cores * IPC.BDW

Simplifying (read: the infamous division which I talked about is here, applied to the Cores):

Time.Zen * IPC.Zen = Time.BDW * IPC.BDW

But we know that:

Time.BDW = 1.02 * Time.Zen

So:

Time.Zen * IPC.Zen = 1.02 * Time.Zen * IPC.BDW

Simplifying:

IPC.Zen = 1.02 * IPC.BDW

Which is the expected result.

 

[lolfail9001]

http://wccftech.com/amd-zen-naples-soc-benchmarks/
This is the first score for this SoC, that appeared in the internet.

It's GB4 result, not GB3.

Difference is just 7% and can be attributed to pretty much anything on a platform like this.
My Haswell at 4.2Ghz scores 4060pts, just for reference. At 2.9Ghz it would score around 2800pts.

edit:
As shown above the score fluctuated to 1141pts.
edit2 : the score of 1141 was achieved on 64bit version of GB4 so not directly comparable.

Here is Skylake 6700K vs my 4690K @ 4.2Ghz in 32bit GB3:
http://browser.primatelabs.com/geekbench3/compare/8082269?baseline=8083536

Minor difference in ST, pretty much below 5%.

I used 6200Mhz 6700k scores as reference, rechecking 4800Mhz 6700k with DDR4@3733 leads to same results in points per Ghz.

Finally, there's the scaling. Suddenly it went from 17x to 20x.

 

[lolfail9001]

Spoiler

We know that:

IPC = the average number of instructions executed for each clock cycle. Per processor. Usually it's referred to a single processor core, and that's what I used. But even if you decide to take all instructions executed by all cores, at the end it doesn't change the overall result; see below.

ExecutedInstructions = Time * Frequency * Cores * IPC ; Time in seconds

But we have some fixed values for Frequency (3Ghz) and Cores (8), both for Zen and Broadwell-E.

So we leave them as they are, and only substitute the different values for the two processors:

ExecutedInstructions.Zen = Time.Zen * Frequency * Cores * IPC.Zen
ExecutedInstructions.BDW = Time.BDW * Frequency * Cores * IPC.BDW

Since the application used is exactly the same, as well as the executed job, we can roughly assume that the number of executed instructions is the same in both cases. Here I'm assuming that there's no specific code path for the two processors: only one, the same, is used for both.

So:

ExecutedInstructions = Time.Zen * Frequency * Cores * IPC.Zen
ExecutedInstructions = Time.BDW * Frequency * Cores * IPC.BDW

hence:

Time.Zen * Frequency * Cores * IPC.Zen = Time.BDW * Frequency * Cores * IPC.BDW

Simplifying (read: the infamous division which I talked about is here, applied to the Cores):

Time.Zen * IPC.Zen = Time.BDW * IPC.BDW

But we know that:

Time.BDW = 1.02 * Time.Zen

So:

Time.Zen * IPC.Zen = 1.02 * Time.Zen * IPC.BDW

Simplifying:

IPC.Zen = 1.02 * IPC.BDW

Which is the expected result.

Your assumption here is beyond rough because you are measuring single core IPC while ignoring the fact that SMT exists. So your IPC value has nothing to do with single thread performance.
Next: your assumption here is even more beyond rough because you consider code executed absolutely the same. When no evidence to this was ever provided.

 

[cdimauro]

The 10 uops cycle probabily is avoid bottlenecks and keep scheduler queues almost empty.

Nevertheless, the queues can receive a maximum of 6 (total) uops per cycle, wheres the schedulers can dispatch 10 of them.

Even if also INTEL has a 6uop cycle from the cache, the shared uop scheduler can be a bottleneck

Why?

and so potentially Zen can have greater IPC than intel's in corner cases (and we hope also in more common cases)...

It might be possible: different microarchitectures have different results depending on the specific task.

Anyway I am still convinced that Zen will be clocked higher than SKL...

It might be, but actually we have no information.

I remember that Zen was compared to similarly clocked INTEL CPUs in Aots and scored similar... If not, my fault...

Take a look at the first post of the thread: the Aots ones are there, with some nice chart with processors stick all at 3.2Ghz, making easier the comparison.

 

[cdimauro]

Your assumption here is beyond rough because you are measuring single core IPC while ignoring the fact that SMT exists. So your IPC value has nothing to do with single thread performance.

I'm not ignoring SMT, because it's part of the calculations.

It's you that are talking of different things, splitting the IPC definition in ST & MT, which doesn't make sense.

Next: your assumption here is even more beyond rough because you consider code executed absolutely the same. When no evidence to this was ever provided.

You have to better read: not absolutely, but roughly the same.

 

[Sweepr]

So this is apparently a GB3 result and it is not in the database. Probably fake.

An elaborate hoax then? Too bad. 🙁

 

[Abwx]

Difference is just 7% and can be attributed to pretty much anything on a platform like this.

About plateforms that s a server one and should be compared to eventual server plateforms as well..

http://browser.primatelabs.com/geekbench3/search?dir=desc&q=Intel+Xeon+E5-2699+LINUX&sort=score

 

[inf64]

An elaborate hoax then? Too bad. 🙁

How do you know it is a hoax? All Zeppelin entries are gone. If the score is legit which I think is the case, it matches perfectly with the claims of 40% higher ST IPC over XV core.

A couple of XV scores (3.8Ghz ST Turbo, locked):
https://browser.primatelabs.com/geekbench3/7933740
https://browser.primatelabs.com/geekbench3/6953761

ST score of ~2500pts or so at 3.8Ghz.
At 1Ghz it should score 657pts. 40%more is around 920pts. Leaked Zeppelin scored 984pts at unknown clock. Unknown clock's likelihood of being 1Ghz is 99% now 😉.

 

[R0H1T]

Are you sure this is at 4.2 Ghz? My i7-6700k at 4.0 Ghz already scores higher than this: http://browser.primatelabs.com/geekbench3/8083540

Didn't notice the mem scores did you ~ https://browser.primatelabs.com/geekbench3/compare/8082269?baseline=8083540

 

[Sweepr]

How do you know it is a hoax? All Zeppelin entries are gone. If the score is legit which I think is the case, it matches perfectly with the claims of 40% higher ST IPC over XV core.

Please come back with an entry or something substantial and then we talk. It's easy to believe what you want to see. 😉

Edit: LOL, that pic comes from WCCFTech comments section.

 

[lolfail9001]

I'm not ignoring SMT, because it's part of the calculations.

It's you that are talking of different things, splitting the IPC definition in ST & MT, which doesn't make sense.

You have to better read: not absolutely, but roughly the same.

Because if you are not making this split then it leads to entire confusion since suddenly a core performance means either single threaded performance (that won't be equal to IPC*clock in your terms), or single core throughput, that will be given by your definition IPC.

Next, even "roughly" is unproven, even if evidently simple messing with flags does not affect it that much.

Anyways, this 1Ghz hypothesis gives AMD fans a hope, so we'll let it live until first proper reviews. Zen certainly won't be anything groundbreaking even in this best case though.

Please come back with an entry or something substantial and then we talk. It's easy to believe what you want to see. 😉

Edit: LOL, that pic comes from WCCFTech comments section.

Wait, really. Welp. Even if there are sometimes some interesting things, that ruins it.

 

[inf64]

Please come back with an entry or something substantial and then we talk. It's easy to believe what you want to see. 😉

Edit: LOL, that pic comes from WCCFTech comments section.

Check my edit above. It is just common sense and logic.

 

[Arachnotronic]

ST score of ~2500pts or so at 3.8Ghz.
At 1Ghz it should score 657pts. 40%more is around 920pts. Leaked Zeppelin scored 984pts at unknown clock. Unknown clock's likelihood of being 1Ghz is 99% now 😉.

99%? Seems too high a likelihood given how much we still don't know...

 

[inf64]

99%? Seems too high a likelihood given how much we still don't know...

Well we know the score of XV and we know what AMD publicly stated (multiple)times about ST IPC increase. Add two and two and the leaked scores only make sense if they were done at 1Ghz. If you think they were done at 1.4Ghz then there is zero IPC gain vs XV on a core that has roughly double the resources of its predecessor. Which makes absolutely no sense.

Take your pick 🙂.

 

[Glo.]

Well we know the score of XV and we know what AMD publicly stated (multiple)times about ST IPC increase. Add two and two and the leaked scores only make sense if they were done at 1Ghz. If you think they were done at 1.4Ghz then there is zero IPC gain vs XV on a core that has roughly double the resources of its predecessor. Which makes absolutely no sense.

Take your pick 🙂.

I would actually agree here with Arach, that we do not know if the CPU in question was running at 1 GHz. It is only our assumption.
If it does, and it scores 1000 Pos in Geekbench for 1 GHz in ST, then its very good for AMD.

 

[Arachnotronic]

Well we know the score of XV and we know what AMD publicly stated (multiple)times about ST IPC increase. Add two and two and the leaked scores only make sense if they were done at 1Ghz. If you think they were done at 1.4Ghz then there is zero IPC gain vs XV on a core that has roughly double the resources of its predecessor. Which makes absolutely no sense.

Take your pick 🙂.

You are discounting the possibility that Zen isn't so great. I don't think it will be so bad that it's barely edging out an XV clock-for-clock, but you are sort of working backwards...in other words, "Zen must be +40% better than XV per clock, therefore let me construct a scenario that proves my point."

GB detected a 1.44GHz CPU, so it's not 100% (or even 99%) that it was running at 1.0GHz. The possibility of background tasks nerfing the performance or perhaps the user adjusting the clock inside of windows with some utility (so it booted at 1.44GHz but could have been running at something different).

Anyway, AMD has given optimistic projections before ("33% more cores, 50% more performance") and they have often failed to materialize.

So I say, let's just wait until the CPU makes it out and people have gotten a chance to test it before drawing conclusions one way or another.

 

[inf64]

Well you can be a pessimist all you want, I'm just thinking logically here. To me it makes no sense that the score was achieved at 1.4Ghz. If Zen is to launch at CES in January that means we will likely have some leaks in December (if we are lucky). That is 2 months away, not a long time.