Ryzen AM5 ; 7000

[RasmusP]

You may have already been using it Rebel but here is the full guide for anyone that might find it useful.

AMD Precision Boost Overdrive 2 - SkatterBencher

Precision Boost Overdrive 2, is AMD's overclocking toolkit to adjust the parameters governing the Precision Boost 2 algorithm.

skatterbencher.com

Awesome information! Thanks mate!

No and again, thanks. If I’m this clueless and there’s no settings I can just copy, I probably shouldn’t be messing to much with this stuff until I learn how to test it out properly.

Hehe yes and no. I can tell you though:
Hitting the power limit (and the temperature limit) while doing a full core benchmark is completely normal with modern CPUs!
The important thing is that you aren't hitting these limits while simracing!

Simracing can best be simulated by running a 3 thread Cinebench r15 test:

However this is NOT clock stretching! That's simply the cores not being at full load long enough due to Windows shuffling the load from core to core for higher efficiency.

Here's how a 12 thread test (full multicore) looks like:

 

[demetri]

@RasmusP how much is 7600X getting with 3 threads?

 

[RasmusP]

@RasmusP how much is 7600X getting with 3 threads?

Around 800 without closing everything. One run gets me to 795, the next to 810. Iirc I got about 840 when closing everything I could.
CB r15 ofc!
Boost clock is 5590 Mhz for single thread runs but usually 5490 during normal use.

Here are my notes from a few years ago for various CPUs:

CPU:​	Threads: 3	FPS	%
Rasmus' i7 2600k	440​	60,00​	100,00​
i5 3570k	424​	57,82​	96,36​
i7 8700k @ 4.8 GHz	620​	84,55​	140,91​
i7 4790k	507​	69,14​	115,23​
Ryzen 2700x OC	544​	74,12​	123,53​
i3 8350k @ 4.9 GHz	607​	82,77​	137,95​
i9 9900k @ 4.7 GHz	615​	83,84​	139,73​
Ryzen 7600X PBO: +200	820​	111,82​	186,36​
7600X Hi-P. Powerplan:	925​	126,14​	210,23​

 

[RasmusP]

Here are my BIOS settings for PBO max, 6000CL30 @ 1.35V 3:2:1 mode.
No expo, 6000, 2000, 1:1 uclk=memclk.
They are slightly blurry to get rid of the monitor-stripes. If you fix the focus and then move the camera a little bit, the stripes disappear.

Translated from this:

 

[Rebel_488]

Perfect dude thanks, I’ll get a proper look tomoro night, long day today so early bed tonight.

 

[RasmusP]

Wow, I did another run with the high performance power plan and scored 925 points for 3 threads!
Repeated it 5x and constantly scored between 897-926 points!

Added that to my table above!

 

[Neilski]

Wow, I did another run with the high performance power plan and scored 925 points for 3 threads!

That's... just nuts

 

[demetri]

Around 800 without closing everything. One run gets me to 795, the next to 810. Iirc I got about 840 when closing everything I could.
CB r15 ofc!
Boost clock is 5590 Mhz for single thread runs but usually 5490 during normal use.

Here are my notes from a few years ago for various CPUs:

CPU:​	Threads: 3	FPS	%
Rasmus' i7 2600k	440​	60,00​	100,00​
i5 3570k	424​	57,82​	96,36​
i7 8700k @ 4.8 GHz	620​	84,55​	140,91​
i7 4790k	507​	69,14​	115,23​
Ryzen 2700x OC	544​	74,12​	123,53​
i3 8350k @ 4.9 GHz	607​	82,77​	137,95​
i9 9900k @ 4.7 GHz	615​	83,84​	139,73​
Ryzen 7600X PBO: +200	820​	111,82​	186,36​
7600X Hi-P. Powerplan:	925​	126,14​	210,23​

I was baffled when I saw 552 with my 12700K, but then I realized that Windows 10 scheduler in its infinite wisdom decided to run it on E-cores. I probably should upgrade to Win 11 soon

Rebooted with e-cores disabled, and it's 867 for 3 and 294 for a single thread. Should be 5.2GHz for 3 cores and 5.2-5.3 for a single one (depends on which core it runs on). DDR4 RAM

 

[RasmusP]

I was baffled when I saw 552 with my 12700K, but then I realized that Windows 10 scheduler in its infinite wisdom decided to run it on E-cores.

Windows at it again, lol!

294 for a single thread

321 for me

 

[demetri]

Windows at it again, lol!


321 for me

Wonder if a good DDR5 or DDR4 B-die would've made any difference for me. I'm running a run of the mill Crucial DDR4 CL16 at 3733, CR=2T

 

[RasmusP]

Wonder if a good DDR5 or DDR4 B-die would've made any difference for me. I'm running a run of the mill Crucial DDR4 CL16 at 3733, CR=2T

Good question.. Not for Cinebench, as that's not sensitive to ram at all.
For simracing I sadly also barely have an idea tbh..

But DDR5 doesn't seem to give you any performance for ACC and for the other sims, your CPU performance is probably good enough?


PurePC.pl tested this:

Hardware Unboxed tested this:

 

[demetri]

Good question.. Not for Cinebench, as that's not sensitive to ram at all.
For simracing I sadly also barely have an idea tbh..

But DDR5 doesn't seem to give you any performance for ACC and for the other sims, your CPU performance is probably good enough?


PurePC.pl tested this:
View attachment 637819

Hardware Unboxed tested this:
View attachment 637820

Why is Intel so bad with DDR5? purepc.pl results could be explained by slower RAM with bad timings, but the second one probably had a much better RAM, so what gives?

Found the specs. Apparently, DDR4 B-dies are better than pretty decent DDR5

 

[RasmusP]

Why is Intel so bad with DDR5? purepc.pl results could be explained by slower RAM with bad timings, but the second one probably had a much better RAM, so what gives?

Found the specs. Apparently, DDR4 B-dies are better than pretty decent DDR5

View attachment 637824

Yeah good question.. Seems like acc is especially latency sensitive. So 3600 cl14 is still better than 6400 cl32.
In any case, a b-die kit might give you a few fps here and there but the money is invested better in a 14700k, 7800x3D or 8600X/8700X, going full ddr5 then.

 

[demetri]

Yeah good question.. Seems like acc is especially latency sensitive. So 3600 cl14 is still better than 6400 cl32.
In any case, a b-die kit might give you a few fps here and there but the money is invested better in a 14700k, 7800x3D or 8600X/8700X, going full ddr5 then.

There's no way I'm paying $250-300 for that. Not worth it at all, chasing those extra 5%. The only reason I decided to upgrade was a very good bundle deal with a decent Z690 mobo that I got for free when I bought the CPU. I paired that with my old DDR4 RAM, so the upgrade was very cheap, and spending almost as much on a fast binned RAM would make zero financial sense

 

[demetri]

Windows at it again, lol!


321 for me

I booted from a thumb drive with a fresh and up-to-date Win11 with lots of preinstalled junk already removed and CB15 scored 300/872 there. No e-cores disable necessary and 1% faster. Yay!

 

[RasmusP]

I booted from a thumb drive with a fresh and up-to-date Win11 with lots of preinstalled junk already removed and CB15 scored 300/872 there. No e-cores disable necessary and 1% faster. Yay!

Nice!

 

[Rebel_488]

Ok reduced pbo to -12 (temp ltd to 85) which got a low score, then to -14 (no temp limit). I did the 3 core test and I get 1867. I’m told I should be getting over 2000. That score is similar to what I had at -28 with temp limit of 85 however that was a single code test.

Cb23 btw should I switch to cb15 like u @RasmusP

 

[Rebel_488]

unfurl="true"]https://skatterbencher.com/amd-precision-boost-overdrive-2/[/URL]

This is great btw thanks for the link

 

[RasmusP]

Brilliant video from hwub again:

 

[RasmusP]

Ok reduced pbo to -12 (temp ltd to 85) which got a low score, then to -14 (no temp limit). I did the 3 core test and I get 1867. I’m told I should be getting over 2000. That score is similar to what I had at -28 with temp limit of 85 however that was a single code test.

Cb23 btw should I switch to cb15 like u @RasmusP

Sorry for the late replay!
The reason for CB15 is that it doesn't use avx instructions. I'm yet to fully understand what these are but basically they are really really heavy on the cpu. Intel has a setting for how much the clocks drop when using avx because the cpu shoots up in power consumption, heat and quickly becomes unstable.

But afaik rarely any games ever use avx instructions.
They apparently can be a lot more efficient overall but they seem to be difficult to code well.

So yeah, to simulate gaming loads and performance, always use CB15!

I'll edit in my cb23 3 core results in that next 30 minutes when I get to the pc