Ddr6000 is 2000 infinity fabric. It’s the IF that’s I’m getting it. It needs to run in a 1:1 ratio or you lose performance
Last edited:
Advice please on DDR5 RAM 48GB (2x24GB) 8000MT/s CL40-48-48-128 1.30V
- Thread starter rapid64k
- Start date
Soldato
- Joined
- 12 Feb 2014
- Posts
- 3,128
- Location
- Somewhere Only We Know
DDR 6000 is Really 3000mhz doubled date rated upto 6000MT/s, the 3000mhz is your MCLK (memory clock) that you see in Zen timings, so 8000MT/s is really 4000mhz double data rate, megahertz vs mega transfers.
But because we cant run FCLK which is our fabric at 3000mhz, which would then make it 1:1, we generally back it off to 3:2 so 3000mhz MCLK=DDR6000 and 2000mhz Fabric.
On AM4 we could, we used to run 3200MT/s RAM which was 1600mhz MCLK, so we would also run our fabric at 1600mhz which was indeed 1:1, but AM5 changed all of that.
Last edited:
Soldato
- Joined
- 12 Feb 2014
- Posts
- 3,128
- Location
- Somewhere Only We Know
My UCLK is at 2000 because im running DDR8000 ive set UCLK/MCLK=2 in the bios so UCLK is running at half speed, theres no way to make it work at 4000mhz the same as my MCLK.
But if I was running DDR6400 I would still run IF at 2200, but I would set UCLK/MCLK=1 in the bios which would put my UCLK at 3200mhz which would be the same as my MCLK 3200mhz / DDR6400
Because IF is completely desynchronised, run it as fast as you can, test with Linpack extreme in safe mode to make sure nothing else interferers with it.
Last edited:
My UCLK is at 2000 because im running DDR8000 ive set UCLK/MCLK=2 in the bios so UCLK is running at half speed, theres no way to make it work at 4000mhz the same as my MCLK.
But if I was running DDR6400 I would still run IF at 2200, but I would set UCLK/MCLK=1 in the bios which would put my UCLK at 3200mhz which would be the same as my MCLK 3200mhz / DDR6400
Because IF is completely desynchronised, run it as fast as you can, test with Linpack extreme in safe mode to make sure nothing else interferers with it.
so with 7200mhz ram my UCLK is now at 1800, so to run higher speed ram you would have to be looking at 8000mhz so as to not lose a little performance?
what would be the real world differance of running 1800 as apposed to 2000 UCLK
im now at 7200mhz 34/42/42/84 UCLK 1800, i was running 6000mhz 30/36/36/76 UCLK 2000
i could probably tighten the timings i haven't tried, i just set EXPO for 7200mhz
Last edited:
Soldato
- Joined
- 12 Feb 2014
- Posts
- 3,128
- Location
- Somewhere Only We Know
Do you mean FCLK ? you cant change UCLK really, its either 1:1 or 2:1
Run FCLK as fast as you can, in all honesty, theres not really any point running 7200, the speeds is probably equivalent to 5600.
What I found was 7800 CL34 was the same as running 6400, 8000 beats both, but 6000 is probably better than 7200.
What Ram do you have ?
with any CPU I get, I set the bios up and before changing anything I to do with CPU & RAM, I find out what FCLK it can do and set SOC voltage to 1.2v a medium voltage, once I know what FCLK it can do, it stays there, not all my 9950X3Ds would do 2200, one of them would do 2167 and one 2133, then I start pushing RAM to get 8000 or above, I test with TM5 as it picks up errors really quickly and I dont have to run 12 hours of it to find an error.
Just to give you an idea.
Last edited:
no the UCLK will change based on the ram speed, i have 7200mhz so the UCLK is now set to 1800mhz.
the MCLK is 2x the UCLK, so at 7200mhz MCLK is 3600 making the UCLK 1800
i have a 48gb 7200mhz T-Force Delta R
Last edited:
Soldato
- Joined
- 12 Feb 2014
- Posts
- 3,128
- Location
- Somewhere Only We Know
Ahh yes sorry my bad, Debauer did a test on UCLK and found that running it 2:1 made little to no difference on performance, i'll see if I can find the video.
I was looking to make this same mistake, have a B850 & 9950X3D, how easy was it to get it to boot and to get it to the 7000 speeds / timings
TBF i just set EXPO 2(7200mhz) and it booted
I had just built it and was messing around with the timings i'm still loading all my "Stuff" back after so have not been back to the timings. I will have a look when I get home. But it probably as simple as what maddness said.
It was durring the build so didn't want to keep messing with it while I had more things to add to the build. May mess with it later but the last thing I had adjusted was the Memory speed which I sent to 7000mhz.
Soldato
- Joined
- 12 Feb 2014
- Posts
- 3,128
- Location
- Somewhere Only We Know
TBH theres not much point buying 7000, or 7200 etc mhz kits, either grab a highly binned 6000 kit like G.Skill's CL26, or grab an 8000 kit which you can either go for 8000, or downclock it and run it at 6000 or 6400, they are pretty cheap now a days, even the G.Skill 32mb CL28 kit is just under £130 and would clock pretty well.
The kit you see in my benchmark above is G.Skill 6000 CL26 kit running at 8000 CL34, it will do 8200, but the voltage requirements are not worth the extra 200mhz, everything needss to be maxed out, SOC, VDDIO, VDDP, VDDQ, VDD etc, whereas I can keep them well within reason for 8000 CL34 and still get decent results.
The kit you see in my benchmark above is G.Skill 6000 CL26 kit running at 8000 CL34, it will do 8200, but the voltage requirements are not worth the extra 200mhz, everything needss to be maxed out, SOC, VDDIO, VDDP, VDDQ, VDD etc, whereas I can keep them well within reason for 8000 CL34 and still get decent results.
Last edited:
I did buy an 8000 DDR5. it's ok any way so I'm not worried I may try to eek a little more out of it but it's ok for now.
Last edited:
Soldato
- Joined
- 12 Feb 2014
- Posts
- 3,128
- Location
- Somewhere Only We Know
Is it the XMP or EXPO kit ? can you noit just load EXPO profile or does it not boot ?
XMP! can't remember if it didn't boot or if it went straight to 5600mhz.
Soldato
- Joined
- 12 Feb 2014
- Posts
- 3,128
- Location
- Somewhere Only We Know
You need quite a bit of voltage to run the XMP kit at 8000 CL34, I needed 1.20v VSOC, 1.4v VDDIO, 1.12v VDDP, 1.58v VDD, and 1.55v VDDQ, its gets quite hot (especially in this weather whilst testing) but you can probably get away with sticking a fan over them just whist testing, you can remove the fan after youve done all your testing, monitor ram temps with HWInfo, you dont really want them going over 53-55c, you will start getting heat related errors after that, personally I strip the heatspreaders off and put iceman heatspeaders on them, cut off the LEDs and stick them under a water block, I know thats not to everyones taste though and you do have to be careful.
However, you could probably get away with less VDD and VDDQ if you ran them at 8000 36-16-48-48-48 and still get a pretty respectable benchmark score, you 9950X3D should be more than capable of achieving this.
Play with tRFC until you get 125ns in zen timings reading,
tRFC2 and tRFCsb can be set to 1 as AMD desktop CPUs dont use those,
tWR always to 48,
tFAW always to 32,
tREFI to 65535 if it will do it, but 50000 will allow it to run cooler,
GDM enabled,
power down disabled,
tRC to 96 (THIS IS ALWAYS TRAS+TRP)
tRDWR = 16,
tRTP = 16,
tWRRD =2,
Set both your SCLs to 8
tRDRDSC = 1
tRDRDSD = 1
tRDRDDD = 1
tWRWRSC = 1
tWRWRSD = 1
tWRWRDD = 1
Leave the rest on auto and see what auto give, you can play with those later on, which should only really leave you with tWTRS, tWTRL, tRRDS and tRRDL to play with.
Set RAM VDD and VDDQ to 1.50v and see if you get any errors, I test with TM5 (Test mem 5) using the 1usmus profile, it picks up errors very quickly, if you dont get any, reduce those 2 voltages to 1.48v and try again.
Last edited:
what about real world performance out side of benchmarks?
at 1080p its about 5%
at 1440 / 4k its 0%
but a bonus is faster ram dose make the CPU draw more power so you know its a win win...
Last edited:
The percentage improvement is dependant on the software (not just the resolution): anything from 1% to 20%. Agreed 5% is probably the average.
The main advantage is improved "1% lows" fps