Overclock.net - An Overclocking Community

Overclock.net - An Overclocking Community (https://www.overclock.net/forum/)
-   AMD CPUs (https://www.overclock.net/forum/10-amd-cpus/)
-   -   Theories on why the SMT hurts the performance of gaming in Ryzen and some recommendations for the future (https://www.overclock.net/forum/10-amd-cpus/1624566-theories-why-smt-hurts-performance-gaming-ryzen-some-recommendations-future.html)

gtbtk 03-04-2017 03:08 AM

Quote:
Originally Posted by ChronoBodi View Post

what would be the option to set 3.9 ghz on all cores, not just one or two at full load?

with a higher BCLK? that would be 3900Mhz/125 = 31.2

 

using only the multiplier with default BCLK is 3900/100=39.

 

Any overclock on Ryzen is currently equal on all operational cores. I dont believe they have enabled independent core overclocking as yet. You can disable cores 2 at a time to get increased overclocks on the remaining cores. 


Undervolter 03-04-2017 03:25 AM

Quote:
Originally Posted by navjack27 View Post

it goes down to 400mhz on the lowest one. you have to enable 'custom' on each one starting with p0 to see the other ones accurately.

Well, that means, that if you choose it, then most likely the CPU will idle at 400Mhz instead of whatever the default set by AMD is. That's a very good feature for undervolters. I 've read people that would resort to 3rd party programs to do that. Like, try to run Phenom or FX at lower idle P-state than the default 1400Mhz.

gtbtk 03-04-2017 06:03 AM

Quote:
Originally Posted by Undervolter View Post
 
Quote:
Originally Posted by navjack27 View Post

it goes down to 400mhz on the lowest one. you have to enable 'custom' on each one starting with p0 to see the other ones accurately.

Well, that means, that if you choose it, then most likely the CPU will idle at 400Mhz instead of whatever the default set by AMD is. That's a very good feature for undervolters. I 've read people that would resort to 3rd party programs to do that. Like, try to run Phenom or FX at lower idle P-state than the default 1400Mhz.

I think that it will use that one when it is in power saving idle mode


Undervolter 03-04-2017 06:21 AM

Quote:
Originally Posted by gtbtk View Post

I think that it will use that one when it is in power saving idle mode

Yes, for the P-States to work, you need Cool N Quiet enabled, or whatever name AMD has given it nowdays (PowerNow? Powersomething?). When "Cool N Quiet" is enabled, the CPU has various P-States, from the lowest (idle) to the highest (the upper turbo). If the motherboard allows to edit them, then you can change them all or if possible, only some. The only trick, is that, if you are allowed to change vcore for each one, you should be certain that the vcore you put, is enough to keep the CPU stable.

gtbtk 03-04-2017 06:30 AM

Quote:
Originally Posted by Undervolter View Post
 
Quote:
Originally Posted by gtbtk View Post

I think that it will use that one when it is in power saving idle mode

Yes, for the P-States to work, you need Cool N Quiet enabled, or whatever name AMD has given it nowdays (PowerNow? Powersomething?). When "Cool N Quiet" is enabled, the CPU has various P-States, from the lowest (idle) to the highest (the upper turbo). If the motherboard allows to edit them, then you can change them all or if possible, only some. The only trick, is that, if you are allowed to change vcore for each one, you should be certain that the vcore you put, is enough to keep the CPU stable.

 

It is doing exactly the same thing that GPUs do.

 

I cant speak for AMD GPus but Nvidia Pascal units use Pstate0 in gaming 3d loads but they use Pstate2 if you run something like lexmark to render a 3d image.

 

Is there a utility that monitors what Pstate the CPU us actually operating in under different loads? What settings can actually be changed with regards pstate performance? it would be interesting to compare the Pstate in cinebench and the pstate in GTA V for example. Tuning those may well be the place to find improvements


Undervolter 03-04-2017 06:37 AM

Quote:
Originally Posted by gtbtk View Post

It is doing exactly the same thing that GPUs do.

I cant speak for AMD GPus but Nvidia Pascal units use Pstate0 in gaming 3d loads but they use Pstate2 if you run something like lexmark to render a 3d image.

Is there a utility that monitors what Pstate the CPU us actually operating in under different loads? What settings can actually be changed with regards pstate performance? it would be interesting to compare the Pstate in cinebench and the pstate in GTA V for example. Tuning those may well be the place to find improvements

The thing is, i don't have Ryzen, so it's all guesswork. If in your BIOS, you can see somewhere the P-States, then you can understand what's running inside Windows, if you have a software that shows clock in real time. For example, the lowest P-state, corresponds to the lowest clock. The next one, to the next attainable clock and so on. AMD in FX, had released a small software called PSCheck, that was showing exactly P-States all the time. But it's of little practical use. When your CPU is at max clock, you are also at the top P-State. As about to performance, i don't know what possibilities the BIOS gives. In AM3, software like K10Stat, was allowing you to set multi (and thus clock), vcore and NB voltage. In AMDMsrTweaker (see my signature), you can adjust multi (thus clock) and vcore.

CrazyElf 03-04-2017 09:07 AM

There is some evidence that the bandwidth may be double, but we will need testing.



Quote:
Originally Posted by gtbtk View Post

Warning: Spoiler! (Click to show)
This is the first post that is actually looking for answers in the right place.

I am really surprised that no-one else has noticed the correlation yet between CPU and GPU performance. All the CPU benchmarks that only hit the CPU seem to provide really good results that beat can Broadwell-E while All the benchmarks that rely on the CPU plus a strong GPU are under performing. GPU load impacts CPU performance.  

Together with the memory clock speed limitations, Isnt it obvious to everyone that the gaming performance issue is being caused because of a weakness in performance somewhare in the interface between the CPU and GPU? The PCIe 3.0 bus is running at a fixed rate x16 so it is not the size of the pipe that is limiting things so there is only one thing left that can possibly be causing the issue and that is the part of the Chip that manages the chip IO (PCIe and Memory controllers - the fabric that you are discussing in the original post).

CPU's, and it doesn't matter if it is Intel or AMD, all have to juggle interdependent resources to get peak performance, too much strength on one side will overwhelm the other side and will reduce performance. Given that none of the Reviewer "experts" nor, apparently any of the Motherboard "engineering" Marketing people have mentioned it would seem to indicate that they don't really understand what is going on are are just trying to follow an overclocking process that they have memorized in the past. 

This is just from a thought experiment but I believe that I can tell you The Solution.

I am pretty certain that the workaround to this apparent conundrum will be to use a higher BCLK frequency with a lower multiplier and fine tune the IO/SOC voltages. It will improve memory clocking limitations and will increase the the number of cycles per second that the PCIe controller can deal with data flow between the CPU and GPU allowing a better balanced system. It will also allow you to  improve fast frequency memory kits ability to clock past 2933Mhz.

I would start by Setting BCLK to 125 and the CPU multiplier to around 32 or 33. Set memory to a higher bin frequency and adjust Ram timings. I don't know what the exact best multiplier/BCLK frequency combination will be, It could be 150/20 for a 4Ghz CPU frequency. That will require experimentation by the people who have the chip in hand. I suspect that the 1800X and 1700X chips, because of higher complexity at the SOC level that currently seems to be having problems, will benefit the most from this approach.


I'm not sure if that solution will work.

Keep in mind that unlike Skylake, the Base clock is not isolated from the rest of the other components. PCie signals begin to degrade after 105 MHz, like on Sandy Bridge. The overclocks to Base Clock may be limited like P67/Z68 was. To the best of my knowledge, there's nothing like the "strap" function on Intel CPUs.


We need mature BIOSes as well to see what the RAM OC is. If the RAM is the last level cache then we need all the OC to RAM that we can get.

sterob 03-04-2017 10:15 AM

Can the SMT problem be fixed with microcode, windows update and software optimization or users should wait out till the next generation?

cloppy007 03-04-2017 10:29 AM

If I had a Ryzen CPU, I would benchmark with different affinity settings, I'm fairly confident that will have a big impact in those games that perform so-so. If that's the case, an updated OS scheduler (or game) will be able to fix, or minimise, that.

navjack27 03-04-2017 10:39 AM

The p States have something to do with the XFR. Yes like a GPU. Each one has its own power offset settings and final clock.

I was planning on messing with process lasso and affinity settings with stuff. It's not like I haven't done that before to tighten frame times in CSGO with Intel CPUs and hyperthreading.


All times are GMT -7. The time now is 01:33 AM.

Powered by vBulletin® Copyright ©2000 - 2019, Jelsoft Enterprises Ltd.

User Alert System provided by Advanced User Tagging (Pro) - vBulletin Mods & Addons Copyright © 2019 DragonByte Technologies Ltd.
vBulletin Security provided by vBSecurity (Pro) - vBulletin Mods & Addons Copyright © 2019 DragonByte Technologies Ltd.

vBulletin Optimisation provided by vB Optimise (Pro) - vBulletin Mods & Addons Copyright © 2019 DragonByte Technologies Ltd.