• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD Ryzen 7 9800X3D

Well, even easy YouTube Buildzoid (Actually Hardware Overclocking) timings would be vastly superior to those god awful timings. Look up his easy Hynix timings. Given you have an 8000 kit, it’s guaranteed A-Die.

However, I can provide you with some tighter timings myself that ‘should’ run, but I can’t guarantee it. All this assumes that you’re comfortable doing timings in the BIOS.

My suggestions just off the top of my head for A-Die would be -

Set trfc to around 500 and trefi to 65535 asap. Those are killing your latency.

Then TRAS to 52. Set trc to 100. Tfaw to 32. Twtrs to 4. Twtrl to 16. Twr to 48.

RDRDSCL and WRWRSCL both to 8.

Trp to 16. RDWR to 20. WRRD to 6

Put 1.45v into VDD and VDDQ
Great thanks I'll take a look and have a test. I can always reset the bios if needs be.
 
Great thanks I'll take a look and have a test. I can always reset the bios if needs be.

Yeah exactly. Those timings I’ve given you should be good, but every chip and stick is different. As said, if you want easy timings, just watch Buildzoid.

Try 1.45v as said. If no joy, try 1.5v. I wouldn’t go higher than that though if your case cooling isn’t good.

You should see a significant improvement in latency and 1% lows.
 
Those are some bogus aida numbers - looks too out of date I think - single CCD chips cannot hit those read speeds for example :)
Yeah I think you're right, updated aida now.
Screenshot-912.png
 
Yeah exactly. Those timings I’ve given you should be good, but every chip and stick is different. As said, if you want easy timings, just watch Buildzoid.

Try 1.45v as said. If no joy, try 1.5v. I wouldn’t go higher than that though if your case cooling isn’t good.

You should see a significant improvement in latency and 1% lows.

how do I post a picture in the chat posting the above link with https or http doesn't seem to work lol

Changes made:

tREFI 15554 > 65535
tRFC1 1177 > 500
tRAS 126 > 52
tRC 176 > 100
tRRDL 16 > 4
tRRDS 8 > 4
tFAW 40 > 32
tRP 48 > 46
tRCDWR 48 > 46
tRCDRD 48 > 46
tWTRL 40 > 32
tWTRS 10 > 6
tRDRDSCL 14 > 8
tWRWRSCL 34 >16
tRDWR 27 > 20
tWRRD 8 > 6
tWR 96 > 48
tRFC2 639 > 256
tRFCSB 519 > 220

cinebench 24

from 1420 > 1470 multi core
137 > 141 single core.
 
Last edited:

how do I post a picture in the chat posting the above link with https or http doesn't seem to work lol

Changes made:

tREFI 15554 > 65535
tRFC1 1177 > 500
tRAS 126 > 52
tRC 176 > 100
tRRDL 16 > 4
tRRDS 8 > 4
tFAW 40 > 32
tRP 48 > 46
tRCDWR 48 > 46
tRCDRD 48 > 46
tWTRL 40 > 32
tWTRS 10 > 6
tRDRDSCL 14 > 8
tWRWRSCL 34 >16
tRDWR 27 > 20
tWRRD 8 > 6
tWR 96 > 48
tRFC2 639 > 256
tRFCSB 519 > 220

cinebench 24

from 1420 > 1470 multi core
137 > 141 single core.

Ahh, so my timings were mostly good? Nice. That latency is so much better. Game performance should have increased a lot too.
 

how do I post a picture in the chat posting the above link with https or http doesn't seem to work lol
If you right click the image after clicking your link should be able to copy image link or similar and use that instead. Alternatively when you are uploading it there is usually a get share links or something button where you can grab the bb code. So for your actual link to pic url is https://i.imgur.com/4glIz3x.png which you can add to have it embedded:

4glIz3x.png
 
2200 is not stable, try 2167. Aida latency should go down, if it goes up or stays the same FCLK needs to be dropped.

This is not 100% true. FCLK and UCLK need to be in sync to provide the lowest bandwidth. The issue with this is that bandwidth metrics will be lower.
By using other ratios you may see bandwidth improvement but you will see the latency go up.

Just divide UCLK by 3 on 1:1 ratio

6000/2000
6200 / 2066
6400/2133
6600/2200
8000/2000 ( in 2:1 8000, UCLK = MCLK/2, 8000:2, so... 4000 UCLK ... and for FCLK you need to divide UCLK it by 2 instead of 3 ... or just do 8000/4=2000 )
8200/2050 (no way to sync this .... )
8400/2100
...

And so on ... Usually games prefer lower latency to higher memory bandwidth... but benchmarks should be done because 1 title is not the same as the other and some may prefer bandwidth more than others.
 
Been playing a few games today, and virtually all of them show some improvement at 2200. Others are around the same. But yes, I thought it was true that latency will rise slightly when going higher and out of ratio.

In my case though, it’s 0.2ns, and the performance increases far outweigh that.
 
Last edited:
Ahh, so my timings were mostly good? Nice. That latency is so much better. Game performance should have increased a lot too.
Yeah thanks for that. Interesting when I reset the bios trefi was around the 70k mark so when set to auto at some point it decided to up it dramatically. I have the profile saved now I might try and reduce it to 6400 with the built in settings see what the difference is
 
Yeah thanks for that. Interesting when I reset the bios trefi was around the 70k mark so when set to auto at some point it decided to up it dramatically. I have the profile saved now I might try and reduce it to 6400 with the built in settings see what the difference is

8000 is a nice number, but in gaming, you’ll almost certainly see performance increases at 6400 1:1 over 8000. Depending on how good your chip is, even 6600 1:1 isn’t impossible.

Now you have the profile saved, you have a solid base to work from, and something that works.

I would try and shoot for 6400CL28 if you can and 3200 UCLK&MCLK, 2200FCLK. Maybe even CL26. Depends how good your individual silicon is, but you have a good RAM kit and mobo.
 
6400 > 8000 ... unless your IOD can't get you stable 6400 without a lot of VSOC. if you need a lot of VSOC, then going 2:1 is a good approach since it needs a lot less voltage.

6400CL28 on MDie = 1.29vsoc
8000CL34 on MDie = 1.1vsoc

6400CL8 on ADie = 1.28vsoc.

It's all about math and if you are ok running high VSOC, but high VSOC contributes to the overall power envelope, meaning, if you want to OC your CPU you need to have that in consideration :)
 
6400 > 8000 ... unless your IOD can't get you stable 6400 without a lot of VSOC. if you need a lot of VSOC, then going 2:1 is a good approach since it needs a lot less voltage.

6400CL28 on MDie = 1.29vsoc
8000CL34 on MDie = 1.1vsoc

6400CL8 on ADie = 1.28vsoc.

It's all about math and if you are ok running high VSOC, but high VSOC contributes to the overall power envelope, meaning, if you want to OC your CPU you need to have that in consideration :)

My chip/IMC is quite strange. At 1.27v, I was getting FCLK errors in P95 large FFT’s. That’s at 6400CL28 and 1:1 with 2200FLCK.

Dropping the VSOC to 1.17v stabilised the FCLK and the RAM stayed stable.

High FCLK does often need lower soc voltage to be stable, but I was quite lucky that the RAM stayed stable at a lower VSOC. I think I have a very strong IMC.

That being said, I would rather run 64001:1 2200FCLK with a 1.3 VSOC than 8000 2:1 with a low VSOC.

8000 is a flexy number, but outside of synthetic benchmarks and productivity, it’s not worth it. 64001:1 is consistently shown to outperform it in games.

To outperform 6400 1:1, you’re likely looking at 8400+
 
Last edited:
The most important consideration is probably RAM voltage and temps. At 6400 1:1 and tight timings, you are going to need quite a bit of VDD/VDDQ voltage. My Gskill needs 1.65v on both. That’s well within safe limits (1.8v) but even with 2x50mm fans over my sticks, I was seeing 50c ram temps in Prime95. Admittedly Prime95 is brutal, but I was still averaging 42c ram temps in Karhu.

Without good case airflow and active ram cooling, a lot of people will be pushing it on voltage to achieve tight timings at 1:1 without causing heat related errors. If you see errors later in to your ram test runs, it’s probably heat related.
 
I will also say that the ‘concerns’ over 1.3 VSOC are largely unwarranted. I see a lot of people intentionally holding back on achieving peak performance because they’re concerned about VSOC voltage.

VSOC voltage was absolutely a problem when some of the boards were bugged and pushing through way too much VSOC. But it’s been fixed for ages.

Will 1.3 VSOC cause degradation? Doubtful. And you’re talking decades if it does. It’s perfectly safe.

If you require 1.3vsoc to get stable optimum ram timings/performance? Just run it.
 
Last edited:
Back
Top Bottom