*** AMD "Zen" thread (inc AM4/APU discussion) ***

[humbug]

Leaked Zen+ and Zen scores compared to measure IPC(taken off Reddit):

https://imgur.com/a/Do2dd

Dell also CBA making more than a few AMD systems it appears,as Intel is bigger:

http://www.channelpro.co.uk/news/10754/dont-expect-many-amd-chips-in-our-products-says-dell

I have a really serious issue with Dell saying not to expect many AMD systems from Dell because Intel are bigger.
For one it was Dell who took the bulk of Intel's bribes not to use AMD CPU's, some $800m a year, so they helped cause it and saying "Intel will continue to dominate our system because Intel have the biggest market share" is just a continuation of that.

But right now i want to concentrate on the new CPU's, 3.7% single-threaded and 8.5% multi-threaded uplift in AVERAGE IPC is significant, couple that with a 10% uplift in clock rates and AMD have a performance competitor to the 8700K at half its price.

 

[Alexrose1uk]

Yes, thats not bad at all for a refinement. Will be interesting to see whether the IPC uplift is the same with faster RAM, ie 3200, or whether it doesnt matter.

 

[humbug]

Yes, thats not bad at all for a refinement. Will be interesting to see whether the IPC uplift is the same with faster RAM, ie 3200, or whether it doesnt matter.

Memory speed for Ryzen has about the same impact on performance in the type of workloads tested here as it does for Intel, little or nothing.

Its only gaming where the memory performance matters.

 

[CAT-THE-FIFTH]

I have a really serious issue with Dell saying not to expect many AMD systems from Dell because Intel are bigger.
For one it was Dell who took the bulk of Intel's bribes not to use AMD CPU's, some $800m a year, so they helped cause it and saying "Intel will continue to dominate our system because Intel have the biggest market share" is just a continuation of that.

But right now i want to concentrate on the new CPU's, 3.7% single-threaded and 8.5% multi-threaded uplift in AVERAGE IPC is significant, couple that with a 10% uplift in clock rates and AMD have a performance competitor to the 8700K at half its price.

Its Dell EMC,ie,the more commercial part of it,but after Baidu,MS and QNAP started adopting Zen based products quite early on,it sounds more like excuses from them. I think they are scared Intel might throw a wobbly since Ryzen based systems would be cheaper,and hence it might force them to price the Intel based ones lower,and that would annoy Intel if they had to drop their prices,and if they didn't Dell would have to make their margins slimmer.

 

[CAT-THE-FIFTH]

Memory speed for Ryzen has about the same impact on performance in the type of workloads tested here as it does for Intel, little or nothing.

Its only gaming where the memory performance matters.

The more I have been following the Ryzen CPU and the Ryzen APU,it makes me wonder whether the gaming "issues" with slower RAM,are also partly down to the highish latency than anything else,looking by the general performance of the Ryzen APUs(single CCX),as faster RAM will also help in that regard. So,if Zen2 has improved inter-CCX bandwidth and an improved memory controller(lower latency),this might actually be the bigger jump in certain games.

 

[CAT-THE-FIFTH]

Some benchmarks of the Acer Nitro 5 which has an RX560:

https://www.notebookcheck.net/First...he-Intel-Core-i7-7700HQ-head-on.277336.0.html

It looks like the chassis has enough cooling for the Ryzen 7 2700U in the laptop to run at a 25W TDP.

 

[B12B6]

Leaked Zen+ and Zen scores compared to measure IPC(taken off Reddit):

https://imgur.com/a/Do2dd

This is 1700 with 2 cores disabled, OC to 3.8, 2600 is runnin 3.8 turbo.

 

[humbug]

Edit, my mistake

The more I have been following the Ryzen CPU and the Ryzen APU,it makes me wonder whether the gaming "issues" with slower RAM,are also partly down to the highish latency than anything else,looking by the general performance of the Ryzen APUs(single CCX),as faster RAM will also help in that regard. So,if Zen2 has improved inter-CCX bandwidth and an improved memory controller(lower latency),this might actually be the bigger jump in certain games.

The 2400G, which for those reading who don't know' like the 1500X has 4 cores and 8 threads and yet with high end GPU's the gaming performance is the same on the 2400G as it is for the 1500X, sometimes actually slightly slower due to having less L3

The 1500X is a 2+2 CCX, i would like to say latency between CCX's is the problem but it looks to me at least that its just yet another one of those things where reviewers like Ryan Shrout like to look for differences where AMD are less than Intel and then point at it in a that is a problem context, in fact Ryzen having slightly higher latency might just be that its a different architecture and is otherwise meaningless.
This is the thing with reviewers like that, they love to compare everything AMD to Intel and comparison fault find, its fashionable to do that and it can earn you money which at the end of the day is their only goal, 8 or more staff and an office or two is very very expensive and google adds don't scratch the surface of that expense, a couple of crates of CPU's 'for testing purposes' from Intel does.

The 2400G has no CCX interconect, its gaming performance is exactly the same as the 1500X, inter CCX latency being the fault of gaming performance is a whole lot of Male Bovine Manure.

 

[humbug]

This is 1700 with 2 cores disabled, OC to 3.8, 2600 is runnin 3.8 turbo.

It says at the bottom "IPC comparison locked at 3.8GHz and 2666MHz RAM" one would assume that's both CPU's. If not then the IPC for the 2600 would be lower given for several of those results the scores are the same.

 

[B12B6]

It says at the bottom "IPC comparison locked at 3.8GHz and 2666MHz RAM" one would assume that's both CPU's. If not then the IPC for the 2600 would be lower given for several of those results the scores are the same.

https://browser.geekbench.com/v4/cpu/compare/6690641?baseline=7255496
He used the old geekbench scores. Also came to the conclusion 2600 was tested with 2666 memory so he replicated the test with hes own 1700 2 cores disabled. 2600 is runnin 3,4-3,8 and "1600" is always 3,8. Actually we dont even know if it was 3,8 turbo but one would assume so as 1600 has 400mhz base-turbo.

 

[humbug]

https://browser.geekbench.com/v4/cpu/compare/6690641?baseline=7255496
He used the old geekbench scores. Also came to the conclusion 2600 was tested with 2666 memory so he replicated the test with hes own 1700 2 cores disabled. 2600 is runnin 3,4-3,8 and "1600" is always 3,8. Actually we dont even know if it was 3,8 turbo but one would assume so as 1600 has 400mhz base-turbo.

Ah, in that case the whole thing is BS, i see it now, its the same thing that's already been posted a couple of weeks ago, i used my own 1600 as a clock for clock comparison and my results were higher on the single threaded performance and the same on multithreaded, or the other way round, can't remember, i'll try and find it.

 

[humbug]

Ah, in that case the whole thing is BS, i see it now, its the same thing that's already been posted a couple of weeks ago, i used my own 1600 as a clock for clock comparison and my results were higher on the single threaded performance and the same on multithreaded, or the other way round, can't remember, i'll try and find it.

Here it is. Its an overclocked Ryzen 1, if you run the 1600/X at 4Ghz you would get the same MT results.

Yeah, i just ran 4.0.3, the same version.

The results are indeed different

Single threaded @ 3.8Ghz
Ryzen 1600: 4258
Ryzen 2600: 4269

Multi threaded @ 3.4Ghz
Ryzen 1600: 17342
Ryzen 2600: 20102 (+16%)

Reader make of that what you will.

https://browser.geekbench.com/v4/cpu/7121287
https://browser.geekbench.com/v4/cpu/7121351

 

[Martini1991]

If you saw a 3.5% gain in IPC with an application then that's your lot, 3.5%. It shouldn't change with multi threading.

 

[humbug]

If you saw a 3.5% gain in IPC with an application then that's your lot, 3.5%. It shouldn't change with multi threading.

I wouldn't say that was strictly true, 100% per core does not = 800% across 8 threads, there is some loss in scaling and if you can improve that scalling then you improve MT performance.

Intel vs AMD is an example of this, AMD's per thread IPC is about 8% down on Coffeelake, and yet with SMT the IPC is bar 1% exactly the same, its because AMD's SMT is better, or so it would seem.

@B12B6 sorry i forgot some context in my last post...

See this "AMD Ryzen: ZD2600BBM68AF_38/34_Y" its the product code from the apparently leaked benchmark, the _38/34 denotes 3.4Ghz Base with a 3.8Ghz turbo, its why i set my comparison clocks to 3.4 single-threaded and 3.8 multi-threaded.

The product code can easily be faked.

 

[CAT-THE-FIFTH]

Edit, my mistake



The 2400G, which for those reading who don't know' like the 1500X has 4 cores and 8 threads and yet with high end GPU's the gaming performance is the same on the 2400G as it is for the 1500X, sometimes actually slightly slower due to having less L3

The 1500X is a 2+2 CCX, i would like to say latency between CCX's is the problem but it looks to me at least that its just yet another one of those things where reviewers like Ryan Shrout like to look for differences where AMD are less than Intel and then point at it in a that is a problem context, in fact Ryzen having slightly higher latency might just be that its a different architecture and is otherwise meaningless.
This is the thing with reviewers like that, they love to compare everything AMD to Intel and comparison fault find, its fashionable to do that and it can earn you money which at the end of the day is their only goal, 8 or more staff and an office or two is very very expensive and google adds don't scratch the surface of that expense, a couple of crates of CPU's 'for testing purposes' from Intel does.

The 2400G has no CCX interconect, its gaming performance is exactly the same as the 1500X, inter CCX latency being the fault of gaming performance is a whole lot of Male Bovine Manure.

I am talking about the memory controller latency(read what I said before). If you look at the reviews of both the Ryzen CPU and APU,the memory controller seems relatively high,and its the same with the APUs.

 

[humbug]

I am talking about the memory controller latency(read what I said before). If you look at the reviews of both the Ryzen CPU and APU,the memory controller seems relatively high.

Does any of this by its self actually make any difference to performance though?

This is what i'm saying, the actual gaming performance of the 2400G with an overkill GPU is no better than the dual CCX 1500X.

 

[CAT-THE-FIFTH]

Does any of this by its self actually make any difference to performance though?

This is what i'm saying, the actual gaming performance of the 2400G with an overkill GPU is no better than the dual CCX 1500X.

Well you said yourself gaming performance seems relatively more dependent on memory bandwidth than non-gaming performance,and AMD gains more from faster RAM than Intel does.

If anything,if you look at reviews which collate performance overall,apparently IPC looks stronger in non-gaming stuff at similar thread counts,than with non-gaming situations. Core IPC is not the issue here,or clockspeeds. If that is the case the Core i5 8400 shouldn't be edging a Ryzen 5 1600X at similar clockspeeds overall. Remember it even has less threads.

I would argue if there is more latency with regards to memory access it would actually impact certain engines worse,especially if the core balancing is not great and the games is doing most of the computational stuff over one or two cores. If there is a bigger delay in memory access when compared to Intel,that would be a bottleneck and also explain why Ryzen seems to scale better and better with faster and low latency RAM in games when compared to Intel. Engines which tend to thread much more evenly don't seem to show as much difference between the two CPUs.

Edit!!

Not to say games devs who CBA is not helping since they should be trying to optimise for Ryzen as well,but I even remember back to the Phenom II X6,that its gaming performance rose dramatically with CPU-NB tweaking and lower latency RAM.

Its why I think if AMD makes some bigger improvements with the memory controller it would help,and inter-CCX bandwidth needs to increase if they do increase core count per CCX with Zen2.

I can't see average DDR4 memory speeds dramatically increasing by 2019,especially with the way things are going now. The RAM cartel is just not helping here at all.

 

[Martini1991]

I wouldn't say that was strictly true, 100% per core does not = 800% across 8 threads, there is some loss in scaling and if you can improve that scalling then you improve MT performance.

Intel vs AMD is an example of this, AMD's per thread IPC is about 8% down on Coffeelake, and yet with SMT the IPC is bar 1% exactly the same, its because AMD's SMT is better, or so it would seem.

@B12B6 sorry i forgot some context in my last post...

See this "AMD Ryzen: ZD2600BBM68AF_38/34_Y" its the product code from the apparently leaked benchmark, the _38/34 denotes 3.4Ghz Base with a 3.8Ghz turbo, its why i set my comparison clocks to 3.4 single-threaded and 3.8 multi-threaded.

The product code can easily be faked.

I wasn't comparing Intel and AMD, it's irrelevant to the point.
And 100% scaling means you'd be at the gain of the single threaded IPC improvement, you shouldn't get over 100% scaling. SMT/HT don't change that, other than the gains that come from using SMT/HT. But until Ryzen, no one ever used HT in the context of IPC improvements (And they still don't, only you do). Haswell 4690K and 4790K are considered to be of equal IPC for example. If we use your logic, then that's no longer true.

 

[humbug]

Well you said yourself gaming performance seems relatively more dependent on memory bandwidth than non-gaming performance,and AMD gains more from faster RAM than Intel does.

If anything,if you look at reviews which collate performance overall,apparently IPC looks stronger in non-gaming stuff at similar thread counts,than with non-gaming situations. Core IPC is not the issue here,or clockspeeds. If that is the case the Core i5 8400 shouldn't be edging a Ryzen 5 1600X at similar clockspeeds overall. Remember it even has less threads .

I would argue if there is more latency with regards to memory access it would actually impact certain engines worse,especially if the core balancing is not great and the games is doing most of the computational stuff over one or two cores. If there is a bigger delay in memory access when compared to Intel,that would be a bottleneck and also explain why Ryzen seems to scale better and better with fast RAM. Engines which tend to thread much more evenly don't seem to show as much difference between the two CPUs.

All this is pseudo logic.
The gaming performance is lower than it perhaps should be, the IMC latency is a little higher than Intel and memory speed performance scaling is higher than Intel, put all that together it must be the reason, right? As people like PcPer concluded it must be because the CCX interconnect latency is causing a slowdown in IPC tasks heavily dependent on gaming.

That ^^^ is a theory.

The 2400G has no CCX interconnect latency because it has no CCX interconnect and yet its no better than the one with the CCX interconnect and its inherent latency.

The 2400G is the experiment to test the aforementioned theory, there is no difference in gaming performance between the two, the experiment proves the theory wrong.

There could be any number of reasons for the perceived lower gaming performance, i say perceived because for a CPU like the 8700K clocked 25% higher than the 1600 is 9 in 10 times significantly less than that faster in games, same goes for Ryzen's performance scaling with memory speed, its better than Intel maybe because it just is that way for other reasons, the 2400G does not have CCX's and yet it scales in the same way. explain that.

 

[humbug]

@Martini1991 Its very simple, MT scaling is not 100%, i score 160 ST, i have 12 threads, i score 1325 with all 12, that's very roughly 70% scaling, if i can make changes to the architecture to bring that scaling up to 80% then i have gained 10% MT IPC, my MT score will go up from 1325 to 1460 and yet my per core performance remains the same.