• 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.

Poll: Ryzen 7950X3D, 7900X3D, 7800X3D

Will you be purchasing the 7800X3D on the 6th?


  • Total voters
    191
  • Poll closed .
Nice, your calculations are off due to the RCD change.

RCD+RTP = RAS 49
RAS+RP = RC 86

If all is good measure Aida latency and maxxmem and note perfiormance.

Then you can try tightening things further, such as:

RDRDDD 4
RDRDSD 2
WRWRDD 8
WRWRSD 2

Measure Aida latency and maxxmem and see if latency drops and bandwidth goes up, check stability.

Then you can try RDRDSCL 5 and WRWRSCL 5 and repeat above.

If anything drops or latency gets worse, revert.
Been running these timings for a few days stable...


Reckon I should push for 6200?

VDD is 1.435 and VDDP at 1.335.
 
Been running these timings for a few days stable...


Reckon I should push for 6200?

VDD is 1.435 and VDDP at 1.335.
Yep. Might need to bump DRAM VDD/VDDQ, SOC and VDDP though. SOC may well be fine where it is, just have to test it.
 
Yep. Might need to bump DRAM VDD/VDDQ, SOC and VDDP though. SOC may well be fine where it is, just have to test it.
Not a good start, set to 6200, raised vdd/vvdq to 1.45/1.35, upped SOC to 1.27 and vddp a little but errors galore.

Then went 1.48/1.38 and put soc to 1.3, still errors. Changed RCD to 37, ras 49 and RC 86... took a bit longer but still got an error, any suggestions?
 
Not a good start, set to 6200, raised vdd/vvdq to 1.45/1.35, upped SOC to 1.27 and vddp a little but errors galore.

Then went 1.48/1.38 and put soc to 1.3, still errors. Changed RCD to 37, ras 49 and RC 86... took a bit longer but still got an error, any suggestions?
Try SOC 1.3v, VDDP 1.150v, DRAM VDD 1.48v-1.5v, VDDQ 1.38v-1.40v and RCD 38 (adjusting RP/RAS/RC accordingly).

If that's stable you can start backing down voltages (starting with SOC) one by one until instability reoccurs.

You could also try CL 30 instead of 28. 28 is going to require a voltage bump for M Die. My last m die kit needed more than 1.5v at 6200C28. A die kit can do it at lower voltage.

EDIT WRRD needs to be at 8 not 2.
 
Last edited:
Try SOC 1.3v, VDDP 1.150v, DRAM VDD 1.48v-1.5v, VDDQ 1.38v-1.40v and RCD 38 (adjusting RP/RAS/RC accordingly).

If that's stable you can start backing down voltages (starting with SOC) one by one until instability reoccurs.

You could also try CL 30 instead of 28. 28 is going to require a voltage bump for M Die. My last m die kit needed more than 1.5v at 6200C28. A die kit can do it at lower voltage.

EDIT WRRD needs to be at 8 not 2.
So tried those voltages, put rcd at 38, ras 50, rc 86 and even upped CL to 30... and changed wrrd to 8. Errors straight away.

Upped vdd/vddq to 1.5/1.4. Still errors. Just doesnt seem to like 6200. Don't really want to up voltage too much so might just leave it at 6000 cl28. Should I leave wrrd at 8 even at 6000?
 
So tried those voltages, put rcd at 38, ras 50, rc 86 and even upped CL to 30... and changed wrrd to 8. Errors straight away.

Upped vdd/vddq to 1.5/1.4. Still errors. Just doesnt seem to like 6200. Don't really want to up voltage too much so might just leave it at 6000 cl28. Should I leave wrrd at 8 even at 6000?
Yes leave at 8, 6000C28 now you can focus on trimming voltages.
 
Run Aida64 and maxxmem and share your results with the stock profile / EXPO profile vs the settings below and see how latency and bandwidth improves.
AxU9BbB.png
To disable geardown mode in the BIOS > DRAM menu look for an option called CMD Mode and set to Buf.
Stock
Aida63 Read 60585 MB/s, latency 84.7 ns
MaxxMem2 51.24 GByte/s, read 49063 MByte/s, write 43719 MByte/s, copy 60951 MByte/s

EXPO
Aida63 Read 78484 MB/s, latency 69.0 ns
MaxxMem2 54.48 GByte/s, read 55874 MByte/s, write 43588 MByte/s, copy 63964 MByte/s

Custom 1
Aida63 Read 84368 MB/s, latency 64.0 ns
MaxxMem2 55.74 GByte/s, read 58569 MByte/s, write 43942 MByte/s, copy 64708 MByte/s

Screenshot-2023-04-28-170850.png
 
Nice, your calculations are off due to the RCD change.

RCD+RTP = RAS 49
RAS+RP = RC 86

If all is good measure Aida latency and maxxmem and note perfiormance.

Then you can try tightening things further, such as:

RDRDDD 4
RDRDSD 2
WRWRDD 8
WRWRSD 2

Measure Aida latency and maxxmem and see if latency drops and bandwidth goes up, check stability.

Then you can try RDRDSCL 5 and WRWRSCL 5 and repeat above.

If anything drops or latency gets worse, revert.
Not sure if I am going to get anymore from it without it going to error mode? Been testing most of the day on the settings and so far, no issues.
iUt1nzw.jpg
 
Stock
Aida63 Read 60585 MB/s, latency 84.7 ns
MaxxMem2 51.24 GByte/s, read 49063 MByte/s, write 43719 MByte/s, copy 60951 MByte/s

EXPO
Aida63 Read 78484 MB/s, latency 69.0 ns
MaxxMem2 54.48 GByte/s, read 55874 MByte/s, write 43588 MByte/s, copy 63964 MByte/s

Custom 1
Aida63 Read 84368 MB/s, latency 64.0 ns
MaxxMem2 55.74 GByte/s, read 58569 MByte/s, write 43942 MByte/s, copy 64708 MByte/s

Screenshot-2023-04-28-170850.png
Nice improvements there. :)
Not sure if I am going to get anymore from it without it going to error mode? Been testing most of the day on the settings and so far, no issues.
iUt1nzw.jpg
Call it a day and move onto curve optimiser then. :)
 
Nice improvements there. :)

Call it a day and move onto curve optimiser then. :)
Absolutely dude. :D I'm really grateful.

Just so I understand this next part.

Ai Tweaker - Precision Boost Overdrive - CPU Boost Clock Override - Enabled (Negative).

Ai Tweaker - Precision Boost Overdrive - Curve optimized - curve optimized - per core. (Set each core to negative sign. Set each core to 15.)

Reboot. Run Prime95. If a core fails go down to -14. If core passes go up to -16?

Every fail go down -1. Every pass go up -1.

Keep repeating until can't go any further than the set negative value for that core?
 
Absolutely dude. :D I'm really grateful.

Just so I understand this next part.

Ai Tweaker - Precision Boost Overdrive - CPU Boost Clock Override - Enabled (Negative).

Ai Tweaker - Precision Boost Overdrive - Curve optimized - curve optimized - per core. (Set each core to negative sign. Set each core to 15.)

Reboot. Run Prime95. If a core fails go down to -14. If core passes go up to -16?

Every fail go down -1. Every pass go up -1.

Keep repeating until can't go any further than the set negative value for that core?
Yep, but for reference I'm running -25 so you can probably start at -20 and go from there (or maybe even -25) and save yourself some work.

I just set it and it didn't crash, so I'll forget it, I can't imagine I can run -30 anyway :D (all cores)
 
Yep, but for reference I'm running -25 so you can probably start at -20 and go from there (or maybe even -25) and save yourself some work.

I just set it and it didn't crash, so I'll forget it, I can't imagine I can run -30 anyway :D (all cores)
Ok. Cool. Thanks dude. You also enabled CPU Boost Clock Override - Enabled (Negative)?

Also, how long do you normally let Prime95 run for?
 
Ok. Cool. Thanks dude. You also enabled CPU Boost Clock Override - Enabled (Negative)?

Also, how long do you normally let Prime95 run for?
About 10mins because I'm impatient :cry:

I figure with a new windows install and nothing really to go wrong if it crashes it's probably PBO, and if it does happen I'll lower the offset. No crashes so far and have hours in BF2042 and Star Wars :)

Yeah you have to make sure you enable it, and then set it to be negative too, then you can just fill in '25' (or your chosen value), and it will apply -25. :)
 
About 10mins because I'm impatient :cry:

I figure with a new windows install and nothing really to go wrong if it crashes it's probably PBO, and if it does happen I'll lower the offset. No crashes so far and have hours in BF2042 and Star Wars :)

Yeah you have to make sure you enable it, and then set it to be negative too, then you can just fill in '25' (or your chosen value), and it will apply -25. :)
I can't wait to try Star wars later. :D

Ok. 10min per test. See how we go. If I experience any crashses I might do a more extensive test when I go to bed.

Also. Say I am able to get a core stable working @ -30.. is there any benefit going higher than -30? Or is that the peak?
 
For a quick and dirty stress test of CO I can recommend the Aida64 Sha3 test. That thing blue screened instantly for me if a core was unstable from a negative offset despite prime95/ycruncher etc being stable. Just set it to 1 CPU and manually assign the affinity to each core (and its SMT partner). If you want absolute stability you kind of need to test in isolation ideally though. Ie only have the core you're testing with an offset and everything else 0. As one core being unstable can effect stability on another one. It's a quick test, and if it passes a few times you can move on. If they're all stable then you can try the full 16 CPU test on all cores, and it should pass.
 
Last edited:
Back
Top Bottom