• Guest, The rules for the P & N subforum have been updated to prohibit "ad hominem" or personal attacks against other posters. See the full details in the post "Politics and News Rules & Guidelines."

AMD Ryzen 5000 Builders Thread

Page 39 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Dave3000

Golden Member
Jan 10, 2011
1,047
37
91
Should I really have to do that? It's
Do you have any fans blowing from top down on to the HSF / VRMs? I have one doing that.

Am running 5837 (same AGESA as your BIOS) on the PRIME X470-PRO, right now, prime95 Small FFT, I am getting ~80*C with a Gammaxx 400 and Ceramique 2 with the 5800X.

Try prime95 small fft to see if it stays around 80*C; IMHO I think prime95 is the 'best' stress test for a cpu.
No top case fan blowing down on the CPU cooler nor the VRMs. I thought that a top case fan should be used as exhaust since hot air rises. I don't have a top case fan. I did run Prime 95 v26.6 Small FFT for 5 minutes and my 5800x hit 89C after about 3 minutes and stayed there for the rest of the 5 minute duration. Frequencies were averaging around 4,470 MHz on all cores during the run. I also ran Prime 95 v30.3 Small FFT for 9 minutes and my 5800x hit 85C and stayed their for the rest of the 9-minute run. Frequencies were averaging around 4,183 MHz during that 9-minute run of Prime 95 v30.3.
 
Last edited:

thigobr

Member
Sep 4, 2016
174
112
116
Humm nice! I will try it out tonight on my X570 Unify!

A workaround to "fix" the L3 performance in AIDA64 is to set EDC > 300A. But I saw a small performance hit on other benchs/tests... I hope this fixes it without the performance hit
 

Det0x

Senior member
Sep 11, 2014
497
588
136
MB defaults on 5900x, 6800XT default clocks, Memory 3600 CL14. I went for the easy memory overclock, but haven't tried anything higher yet as MSI just dropped 1.2.0.0 on the 18th for my MB. Not sure what the CPU boost was during the run? 6800 XT has about 300Mhz of OC headroom currently left on the table.

View attachment 40028
5950x at bone stock with 3800CL14 tuned timings. Also 261 avg cpu game.
View attachment 40058
LOL i was benching the wrong version of the game.. Free version on steam was v505 (patch 15) which was missing 4 performance updates :D
Version 298 = patch 19

This is more like it :innocent:
CPU Game = Average 282 fps

282.png
1614277261011.png
 

pj-

Senior member
May 5, 2015
418
160
116
Is there a good guide for explaining curve optimizer?

I think I mostly understand the settings on the PBO page specifically, but I don't really know how those settings interact with other BIOS settings and what I should be doing to safely maximize performance.

On my 5950x I'm getting 5ghz on 2 cores in lightly threaded work and about 4.45ghz all core in r23, at 83C with a 2x 360 rad loop. Scores seem reasonable but not amazing, ~28500 r23 MT
 

Kenmitch

Diamond Member
Oct 10, 1999
8,405
2,146
136
Is there a good guide for explaining curve optimizer?

I think I mostly understand the settings on the PBO page specifically, but I don't really know how those settings interact with other BIOS settings and what I should be doing to safely maximize performance.

On my 5950x I'm getting 5ghz on 2 cores in lightly threaded work and about 4.45ghz all core in r23, at 83C with a 2x 360 rad loop. Scores seem reasonable but not amazing, ~28500 r23 MT
It's going to be trial and error from chip to chip, but I guess you could look over the following and see if it helps you get on your way....Good luck with your adventure!

.

There's also this thread which looks to discuss it which might also help you on your way.

 

Makaveli

Diamond Member
Feb 8, 2002
4,288
476
126
It's going to be trial and error from chip to chip, but I guess you could look over the following and see if it helps you get on your way....Good luck with your adventure!

.

There's also this thread which looks to discuss it which might also help you on your way.

+1

I used both of those links while tuning CO on my rig.
 

pj-

Senior member
May 5, 2015
418
160
116
It's going to be trial and error from chip to chip, but I guess you could look over the following and see if it helps you get on your way....Good luck with your adventure!

.

There's also this thread which looks to discuss it which might also help you on your way.

Thanks, I'll check those out. Also found this one which seems pretty thorough:

There does seem to be some conflicting info. One says that your best cores should have the highest negative offsets (makes sense to me), but others say they should have the smallest offset?? Maybe the theory is that your bad cores suck so you might as well save some voltage and push your good ones?
 

Dave3000

Golden Member
Jan 10, 2011
1,047
37
91
I realize that 3Dmark03 is an old benchmark but I think I'm not getting the performance out of my CPU that I should be getting based on the Troll's Lair test. I have a Ryzen 5800x, GTX EVGA GTX 1080 Ti Black SC, and 32GB DDR4-3200 (16-18-18-36) in my system. I'm getting around 1350 points in the Troll's Lair test. At the point where the camera zooms in on the girl's face at the entrance of the room it drops down to 163 fps. I checked Youtube videos and found a video showing 3Dmark03 benchmark running and at that same point it drops down to 206 fps on a 7700k overclocked to 4.8GHz (my CPU at stock should still be a little faster than this in single-threaded performance) with a GTX 1080 FW2 (my video card should still be a little faster than this). Is that point where I see 163 fps in Troll's Lair down to some mechanism in the 5800x that is slower than an i7-7700k that I'm not aware of? I think I should be getting around 210-220 fps instead of 163 fps at that point in Troll's Lair based on that youtube video I checked out. I could not find a youtube video showing a Ryzen 3000 or 5000 series running 3Dmark03 but I do know that still a 5800x should be better than a 7700k overclocked at 4.8GHz in gaming.
 

Shmee

Memory and Storage, Graphics Cards
Super Moderator
Sep 13, 2008
4,956
791
126
... BIOS fix for L3
( hopefully my issue with 1933 and above )

2021/02/24 AGESA ComboAM4PIV2 1.2.0.1 Beta BIOS which fixes L3 cache performance issue

MSI Beta BIOS
Anyone try this new AGESA here? Would you recommend it? Any caveats to know before flashing?
 

thigobr

Member
Sep 4, 2016
174
112
116
I tried the new 1.2.0.1 AGESA and I didn't see any significant difference when compared to 1.2.0.0 except for the AIDA64 L3 bandwidth numbers. FCLK and Curve Optimizer behavior looks identical.
 

lightmanek

Senior member
Feb 19, 2017
280
512
136
I realize that 3Dmark03 is an old benchmark but I think I'm not getting the performance out of my CPU that I should be getting based on the Troll's Lair test. I have a Ryzen 5800x, GTX EVGA GTX 1080 Ti Black SC, and 32GB DDR4-3200 (16-18-18-36) in my system. I'm getting around 1350 points in the Troll's Lair test. At the point where the camera zooms in on the girl's face at the entrance of the room it drops down to 163 fps. I checked Youtube videos and found a video showing 3Dmark03 benchmark running and at that same point it drops down to 206 fps on a 7700k overclocked to 4.8GHz (my CPU at stock should still be a little faster than this in single-threaded performance) with a GTX 1080 FW2 (my video card should still be a little faster than this). Is that point where I see 163 fps in Troll's Lair down to some mechanism in the 5800x that is slower than an i7-7700k that I'm not aware of? I think I should be getting around 210-220 fps instead of 163 fps at that point in Troll's Lair based on that youtube video I checked out. I could not find a youtube video showing a Ryzen 3000 or 5000 series running 3Dmark03 but I do know that still a 5800x should be better than a 7700k overclocked at 4.8GHz in gaming.
For that benchmark to perform you would need to lower your memory latency a bit. Optimized 3200 CL14 does miracles for old engined video performance.
 

Justinus

Platinum Member
Oct 10, 2005
2,458
592
136
I realize that 3Dmark03 is an old benchmark but I think I'm not getting the performance out of my CPU that I should be getting based on the Troll's Lair test. I have a Ryzen 5800x, GTX EVGA GTX 1080 Ti Black SC, and 32GB DDR4-3200 (16-18-18-36) in my system. I'm getting around 1350 points in the Troll's Lair test. At the point where the camera zooms in on the girl's face at the entrance of the room it drops down to 163 fps. I checked Youtube videos and found a video showing 3Dmark03 benchmark running and at that same point it drops down to 206 fps on a 7700k overclocked to 4.8GHz (my CPU at stock should still be a little faster than this in single-threaded performance) with a GTX 1080 FW2 (my video card should still be a little faster than this). Is that point where I see 163 fps in Troll's Lair down to some mechanism in the 5800x that is slower than an i7-7700k that I'm not aware of? I think I should be getting around 210-220 fps instead of 163 fps at that point in Troll's Lair based on that youtube video I checked out. I could not find a youtube video showing a Ryzen 3000 or 5000 series running 3Dmark03 but I do know that still a 5800x should be better than a 7700k overclocked at 4.8GHz in gaming.
Can you share your total score and the FPS breakdown?
 

Dave3000

Golden Member
Jan 10, 2011
1,047
37
91
For that benchmark to perform you would need to lower your memory latency a bit. Optimized 3200 CL14 does miracles for old engined video performance.
Even if the CL of RAM did effect that scene at that specific point in the benchmark, it surely would not go from 163 fps to 210-220 fps just by lowering the CL of my RAM from 16 to 14. It would probably be around 166-168 I'm guessing if it was really reliant on memory latency at that point of the scene. I'm wondering if the memory write performance of a single CCD is what is holding it back for this specific case. Would a 5900x and 5950x get around 210-220 because those CPUs have double the memory write bandwidth of the 5800x and 5600x due to a dual CCD design?

Update: I increased the latencies of my RAM from 16-18-18-36 to 18-20-20-38 and the frame rate at that point in Troll's Lair went from 163 fps to 162 fps but then again sometimes I got 162 fps even at 16-18-18-36. I don't think 3200 CL14 RAM would make any difference, basically within margin of error, compared to 3200 CL16 at that point in Troll's Lair if going from 16-18-18-36 to 18-20-20-38 at 3200 does not seem to make a difference at that specific point. Also I don't want to run my RAM faster than it's advertised specs (16-18-18-36 at 3200) and that is why I tested it with slower timings for comparing instead of faster timings than what my RAM is advertised for at 3200. I'm still wondering if I'm memory write bandwidth limited in that specific point in Troll's Lair in 3Dmark 03.
 
Last edited:

Justinus

Platinum Member
Oct 10, 2005
2,458
592
136
Even if the CL of RAM did effect that scene at that specific point in the benchmark, it surely would not go from 163 fps to 210-220 fps just by lowering the CL of my RAM from 16 to 14. It would probably be around 166-168 I'm guessing if it was really reliant on memory latency at that point of the scene. I'm wondering if the memory write performance of a single CCD is what is holding it back for this specific case. Would a 5900x and 5950x get around 210-220 because those CPUs have double the memory write bandwidth of the 5800x and 5600x due to a dual CCD design?

Update: I increased the latencies of my RAM from 16-18-18-36 to 18-20-20-38 and the frame rate at that point in Troll's Lair went from 163 fps to 162 fps but then again sometimes I got 162 fps even at 16-18-18-36. I don't think 3200 CL14 RAM would make any difference, basically within margin of error, compared to 3200 CL16 at that point in Troll's Lair if going from 16-18-18-36 to 18-20-20-38 at 3200 does not seem to make a difference at that specific point. Also I don't want to run my RAM faster than it's advertised specs (16-18-18-36 at 3200) and that is why I tested it with slower timings for comparing instead of faster timings than what my RAM is advertised for at 3200. I'm still wondering if I'm memory write bandwidth limited in that specific point in Troll's Lair in 3Dmark 03.
You're scoring higher than I do with a 5950x with 3800CL14 on a 6900xt.

I think you're reading into this old benchmark too much. Was the youtube video even running W10 or were they running it on windows 7 or XP, where it performs significantly better?
 
  • Like
Reactions: Tlh97 and Makaveli

Dave3000

Golden Member
Jan 10, 2011
1,047
37
91
You're scoring higher than I do with a 5950x with 3800CL14 on a 6900xt.

I think you're reading into this old benchmark too much. Was the youtube video even running W10 or were they running it on windows 7 or XP, where it performs significantly better?
Here's the link to the youtube video that I mentioned and at 2:23-2:25 is where I mentioned that it drops to 163 fps on my system and in this person's system in the video it drops to 207 fps at that same spot:

 
Last edited:

B-Riz

Golden Member
Feb 15, 2011
1,295
364
136
Should I really have to do that? It's


No top case fan blowing down on the CPU cooler nor the VRMs. I thought that a top case fan should be used as exhaust since hot air rises. I don't have a top case fan. I did run Prime 95 v26.6 Small FFT for 5 minutes and my 5800x hit 89C after about 3 minutes and stayed there for the rest of the 5 minute duration. Frequencies were averaging around 4,470 MHz on all cores during the run. I also ran Prime 95 v30.3 Small FFT for 9 minutes and my 5800x hit 85C and stayed their for the rest of the 9-minute run. Frequencies were averaging around 4,183 MHz during that 9-minute run of Prime 95 v30.3.
Finally got back to messing with the 5800X setup; what is your ambient temperature? I am in the basement at ~66*F. I can try moving upstairs, but that is kept at 68*F in the winter and ~ 75*F in the summer, Midwest USA.

Also, something I saw on Reddit is to put a negative offset on for CPU voltage, -.15. I am testing that now in prime95. That is what I did with the 2700X, neg offset, and it ran beautifully and passed a 24 prime95 small fft test back in the day.

Or you can do like @IEC said:


@Dave3000 Enable PBO2 and set your PPT to 105W (+/-). Depending on silicon lottery, you should find performance closer to stock but with at least 10C cooler load temperatures.

I am also NOT using XMP, just setting memory up manually in the BIOS.
 

lightmanek

Senior member
Feb 19, 2017
280
512
136
Even if the CL of RAM did effect that scene at that specific point in the benchmark, it surely would not go from 163 fps to 210-220 fps just by lowering the CL of my RAM from 16 to 14. It would probably be around 166-168 I'm guessing if it was really reliant on memory latency at that point of the scene. I'm wondering if the memory write performance of a single CCD is what is holding it back for this specific case. Would a 5900x and 5950x get around 210-220 because those CPUs have double the memory write bandwidth of the 5800x and 5600x due to a dual CCD design?

Update: I increased the latencies of my RAM from 16-18-18-36 to 18-20-20-38 and the frame rate at that point in Troll's Lair went from 163 fps to 162 fps but then again sometimes I got 162 fps even at 16-18-18-36. I don't think 3200 CL14 RAM would make any difference, basically within margin of error, compared to 3200 CL16 at that point in Troll's Lair if going from 16-18-18-36 to 18-20-20-38 at 3200 does not seem to make a difference at that specific point. Also I don't want to run my RAM faster than it's advertised specs (16-18-18-36 at 3200) and that is why I tested it with slower timings for comparing instead of faster timings than what my RAM is advertised for at 3200. I'm still wondering if I'm memory write bandwidth limited in that specific point in Troll's Lair in 3Dmark 03.
Main timings are not going to do that much difference, hence I said optimized timings. Where the difference will show is doing complete main and sub-timing tweak, it has a large influence on benchmark like these. I would run it on my PC, but I'm busy till weekend, with little to no spare time in the week.
I've seen uplifts north of 20% by just setting all the memory timings manually on a good B-Die modules. Also, some motherboards have a BIOS setting for 3DMark03 specificaly.

What also matters is which vendor GPU you have and generation, as old 3DMark engine relies on specific DX functions no longer supported in hardware on newer cards, so there is a lot to the final performance.
 

B-Riz

Golden Member
Feb 15, 2011
1,295
364
136
Finally got back to messing with the 5800X setup; what is your ambient temperature? I am in the basement at ~66*F. I can try moving upstairs, but that is kept at 68*F in the winter and ~ 75*F in the summer, Midwest USA.

Also, something I saw on Reddit is to put a negative offset on for CPU voltage, -.15. I am testing that now in prime95. That is what I did with the 2700X, neg offset, and it ran beautifully and passed a 24 prime95 small fft test back in the day.

Or you can do like @IEC said:
Enable PBO2 and set your PPT to 105W (+/-). Depending on silicon lottery, you should find performance closer to stock but with at least 10C cooler load temperatures.
I am also NOT using XMP, just setting memory up manually in the BIOS.

@Dave3000

Offset did not do anything, still hit 80*C. Tried PPT Limit at 120, that worked to bring it down to ~74*C.

So the PPT limit is your best option if you have a warm to hot ambient temp and did not load the case up with fans.

I am going to fiddle with PPT some more, but may just end up running stock config. If AMD designed for 80*C limit, does not bother me.
 

ASK THE COMMUNITY