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

[Abwx]

Yes 8 cores, up to 16 threads, up to 3,4+
They don't say 18threads at full load at 3,4 +

If you have better infos than this then you can post them eventually, otherwise you are just making hollow theories....

They gave enough infos and their slides are explicit, now it seems that Zen will be much better than what some people expected, FI we are a long shot from the 2.8GHz base that were thrown by some "specialists" in various Zen dedicated threads, so make the conclusions you want but so far all nay sayers were wrong in all possible discussed metrics, and i think that it s not over.

 

[TheELF]

If you have better infos than this then you can post them eventually, otherwise you are just making hollow theories....

They gave enough infos and their slides are explicit,

If you have better infos then you post them please.

They gave enough infos and their slides are explicit,

Oh yeah,what's senseMI?

 

[cytg111]

Yes 8 cores, up to 16 threads, up to 3,4+
They don't say 18threads at full load at 3,4 +

Applied logic tells me it is a 3.4 base clock. Adjusting for PR departments relationship with reality I could dial that back to whatever.. However with recent leak of max OC 4.3 for A0, max 4.5, a 3.4 baseclock is not that far fetched imo?

 

[Chicken76]

So the event starts at 21:00 UTC? Any estimate of its duration?

 

[TheELF]

Applied logic tells me it is a 3.4 base clock. Adjusting for PR departments relationship with reality I could dial that back to whatever.. However with recent leak of max OC 4.3 for A0, max 4.5, a 3.4 baseclock is not that far fetched imo?

Even for those leaks it's not clear if they where testing for stability with 8 or 16 threads.
It might very well be base clock, I'm just saying that it's not clear,no way near clear actually.

Applied logic would tell you to examine all slides as one entity so 3,4+ could very well refer to auto overclock together with the power saving stuff and all the rest.

 

[Abwx]

If you have better infos then you post them please.

Oh yeah,what's senseMI?

Changing the goal posts..?..

What is the relevance with frequency and thread counts, because those latter datas are explicit.

Applied logic tells me it is a 3.4 base clock. Adjusting for PR departments relationship with reality I could dial that back to whatever.. However with recent leak of max OC 4.3 for A0, max 4.5, a 3.4 baseclock is not that far fetched imo?

Thye need about 10% frequency margin, wich is not mandatory but is an accepted enginering design rule, so if their samples hit 4.3 they could release an SKU that turbo up to 3.8-3.9 when less than 8 cores are loaded.

 

[TheELF]

Changing the goal posts..?..

How is it changing the goalposts?
If you think that every thing on that slide relates to each other (instead of just being a list of features) than please explain what senseMi is and what it has to do with clocks.

 

[Atari2600]

They gave enough infos and their slides are explicit

AMD's marketing slides are always "selectively" explicit.


For example:
"We expect across a wide variety of workloads for Barcelona to outperform Clovertown by 40 percent"

This was **true**, even giving the gimp'd Barcelona clocks. But it was only true whenever the user required full memory bandwidth and the IMC in Barcelona could stride clear of the FSB/Clovertown combination.

Now, would you consider that to be a wide variety of workloads?

 

[blublub]

Lol - no. But I hope AMD isn't that dumb anymore...

 

[leoneazzurro]

There is no "workload" word anywhere in the slides, nor a dependency on the number of threads. It is written 3,4+ GHz, so it is safe to assume that is the base clock of the highest performing SKU.

 

[KTE]

If you have better infos than this then you can post them eventually, otherwise you are just making hollow theories....

They gave enough infos and their slides are explicit, now it seems that Zen will be much better than what some people expected, FI we are a long shot from the 2.8GHz base that were thrown by some "specialists" in various Zen dedicated threads, so make the conclusions you want but so far all nay sayers were wrong in all possible discussed metrics, and i think that it s not over.

Still miles better than the 4.0-5GHz base hyped by some specialists

I think 3.2-3.4GHz base would be good and right inline with my projections based on the released info.

Altho hugely backwards in frequencies but that was a given. 5.5yrs ago they started at 3.6GHz-4.2GHz.

Sent from HTC 10
(Opinions are own)

 

[rainy]

Altho hugely backwards in frequencies but that was a given. 5.5yrs ago they started at 3.6GHz-4.2GHz.

Sent from HTC 10
(Opinions are own)

With a "small" exception that FX-8150 was an 8-thread CPU and top consumer Zen is 8c/16t.

Btw, that was a bit over 5 years ago (October 2011).

 

[mikk]

Can i have some link where its quantified how much performance brings that 128MB of edram to the table?

And if games are so cache dependent, why use them as a proof of anything when talking about IPC (gains)?

You have to search for some tests. You can see it even in this test. Important is, you cannot compare the IPC when one CPU does include edram and the other doesn't. Games are not necessarily cache dependent, they are memory and highly IPC dependent as well. Obviously you are really clueless if you ask such questions.

 

[deasd]

Still miles better than the 4.0-5GHz base hyped by some specialists

I think 3.2-3.4GHz base would be good and right inline with my projections based on the released info.

(Opinions are own)

Eh, IIRC I didn't see anyone on the web claimed 4-5Ghz based for an 8C Zen, but there was indeed some 2.8Ghz based 'guesstimation' around when AotS bench leaked.

But yes the '3.4Ghz+' description is quite confused and misleading, especially this digit doesn't represent any retail SKUs' clock, just a range that 8C might land in.

I agree with 3.2-3.4Ghz, which is a little better than I expected, but still I won't care about few MHz less or more then it's released.

Altho hugely backwards in frequencies but that was a given. 5.5yrs ago they started at 3.6GHz-4.2GHz.

No. Frequency is an overall characteristic of both architecture and process node, you cannot compare between different design directly.

 

[bjt2]

Still miles better than the 4.0-5GHz base hyped by some specialists

I think 3.2-3.4GHz base would be good and right inline with my projections based on the released info.

Altho hugely backwards in frequencies but that was a given. 5.5yrs ago they started at 3.6GHz-4.2GHz.

Sent from HTC 10
(Opinions are own)

The CPU is not out yet. Maybe 3.4GHz is conservative and the mean clock will be 4. Anyway 4.5Ghz max turbo seems to be true, according to the graph and if each line is 1Ghz...

 

[Arachnotronic]

The CPU is not out yet. Maybe 3.4GHz is conservative and the mean clock will be 4. Anyway 4.5Ghz max turbo seems to be true, according to the graph and if each line is 1Ghz...

Why would each line be 1GHz?

 

[Dresdenboy]

Yes 8 cores, up to 16 threads, up to 3,4+
They don't say 18threads at full load at 3,4 +

If they meant "up to", you'll usually find the string "up to" on the slide, even if just 1 pixel high. Without having checked everything, I wouldn't say, this is true for 100% of their slides presented during the last 2 years, but the number should be rather high.

Otherwise, shouldn't they write "-3.4 GHz" then?

 

[Abwx]

For example:
"We expect across a wide variety of workloads for Barcelona to outperform Clovertown by 40 percent"

It was a statement made relatively to the server market and particularly HPC, and they were right, in intensive FP computation like LS Dyna it was 40% faster clock/clock than the Intel equivalent, as aknowledged publicly by Intel themselves, now dont come with the fallacy that consist to pretend that they stated so for DT consumer markets SKUs.

 

[Toettoetdaan]

The CPU is not out yet. Maybe 3.4GHz is conservative and the mean clock will be 4. Anyway 4.5Ghz max turbo seems to be true, according to the graph and if each line is 1Ghz...

Why would each line be 1GHz?

Yes? As the chart says: "chart for illustrative purposes only"
(On the other hand, the minimal steps in that chart should represent 25Mhz)

 

[Arachnotronic]

Yes? As the chart says: "chart for illustrative purposes only"
(On the other hand, the minimal steps in that chart should represent 25Mhz)

The Y axis isn't even labeled.

 

[TheELF]

If they meant "up to", you'll usually find the string "up to" on the slide, even if just 1 pixel high. Without having checked everything, I wouldn't say, this is true for 100% of their slides presented during the last 2 years, but the number should be rather high.

Otherwise, shouldn't they write "-3.4 GHz" then?

So SMT will not be "turn-off-able" ?
You will always be forced by AMD law to run 16 threads?

 

[PhonakV30]

Yes 8 cores, up to 16 threads, up to 3,4+
They don't say 18threads at full load at 3,4 +

3.4+ means Full load ( 16 threads ) running at 3.4 , now you can overclock it up to 4.0GHz but i think maximum clock for Zen (due to GF's 14nm Process ) is around 3.8Ghz ( Full load with 16 threads).even Core i7-6900K can't be overclocked to 4.5Ghz.

 

[Atari2600]

It was a statement made relatively to the server market and particularly HPC, and they were right, in intensive FP computation like LS Dyna it was 40% faster clock/clock than the Intel equivalent, as aknowledged publicly by Intel themselves, now dont come with the fallacy that consist to pretend that they stated so for DT consumer markets SKUs.

I've already detailed exactly why it outperformed Clovertown. I know, I was using varieties of both CPUs for HPC and general use at the time.

AMD said they outperformed Clovertown over a wide variety of workloads.

Getting trounced in workloads that didn't require the capacity of the IMC doesn't seem to be a 40% advantage* over a wide variety to me.

*Note, the AMD PR did not qualify their advantage as clock-for-clock, but rather the much more ambiguous term "outperform"... which is essentially misleading when the clock ceiling was looking at being so low... before even the TLB bug.


[As an aside, in most instances a touch less than half of the performance deficit of Clovertown could be negated by setting process affinity for each thread. Thread thrashing across cores was frequently resulting in L1 and L2 cache clearances across the FSB further crippling an already saturated memory interface.]

 

[sirmo]

3.4+ means Full load ( 16 threads ) running at 3.4 , now you can overclock it up to 4.0GHz but i think maximum clock for Zen (due to GF's 14nm Process ) is around 3.8Ghz ( Full load with 16 threads).even Core i7-6900K can't be overclocked to 4.5Ghz.

We seem to have it on "good authority" that it will OC ~4.3Ghz. Is your 3.8Ghz your personal guestimate or you know something from the actual real world testing?

 

[EightySix Four]

Yes 8 cores, up to 16 threads, up to 3,4+
They don't say 18threads at full load at 3,4 +

Don't be obtuse. 3.4GHz+ clearly means 3.4GHz or higher, which means the base clock is 3.4GHz.