Should I keep my DDR5 ram on 6400 or 6000 ?

[g335]

Hello,

 

I just replaced the DDR5 6000mhz 2x32gig  for my Ryzen 7950X with DDR5 6400mhz 2x48gigs.

 

The memory speed in bios was set on Auto on the Asus ROG B650E-E board I have(I have one of the newer bios updates).   It set the new memory at 6000mhz  and the system came on fine.   I went back and changed it to DOCP1/EXPO 1 6400mhz and its running fine.

 

The timings for the new ram is CL32-39-39-102.   I ran performance test by Passmark and the score is 3816 with memory latency 51.

 

I ran Aida 64 stability test for 11 mins, system did not crash. The temperature  of the dims was dim2 55 degrees dim 4 50 degrees  On the Cache & Memory Benchmark, it showed the latency at 87.2

 

Why would one software show latency 51 and the other 87?

 

Will keeping the ram on 6400mhz affect performance in a bad way?  The new bios updates for Ryzen boards allow 48 gig ram in each channel.

 

Should I keep it on 6400mhz?

 

 

[TatamiMatt]

1 minute ago, g335 said:

Hello,

 

I just replaced the DDR5 6000mhz 2x32gig  for my Ryzen 7950X with DDR5 6400mhz 2x48gigs.

 

The memory speed in bios was set on Auto on the Asus ROG B650E-E board I have(I have one of the newer bios updates).   It set the new memory at 6000mhz  and the system came on fine.   I went back and changed it to DOHC1/EXPO 1 6400mhz and its running fine.

 

The timings for the new ram is CL32-39-39-102.   I ran performance test by Passmark and the score is 3816 with memory latency 51.

 

I ran Aida 64 stability test for 11 mins, system did not crash. The temperature  of the dims was dim2 55 degrees dim 4 50 degrees  On the Cache & Memory Benchmark, it showed the latency at 87.2

 

Why would one software show latency 51 and the other 87?

 

Will keeping the ram on 6400mhz affect performance in a bad way?  The new bios updates for Ryzen boards allow 48 gig ram in each channel.

 

Should I keep it on 6400mhz?

 

 

Run a few more tests, take the average then change it to 6000 CL30 and adjust timings accordingly and then run another few tests, average and compare, whichever provides better performance, roll with that

[g335]

5 minutes ago, TatamiMatt said:

Run a few more tests, take the average then change it to 6000 CL30 and adjust timings accordingly and then run another few tests, average and compare, whichever provides better performance, roll with that

How do I know what good performance numbers are?   I will change to 6000 and try but will not mess with the timings, dont know how do it and dont want to kill any parts on the computer.

 

[TatamiMatt]

2 minutes ago, g335 said:

How do I know what good performance numbers are?   I will change to 6000 and try but will not mess with the timings, dont know how do it and dont want to kill any parts on the computer.

 

only issue with that is, with the same timings but slower running, it quite simply will be worse with no reason to compare, you need to adjust the timings for it to be equivalent, are you running XMP/DOCP profiles? Or is the 6400CL32 EXPO based?

[g335]

12 minutes ago, TatamiMatt said:

only issue with that is, with the same timings but slower running, it quite simply will be worse with no reason to compare, you need to adjust the timings for it to be equivalent, are you running XMP/DOCP profiles? Or is the 6400CL32 EXPO based?

DOCP/EXPO 1 is what it says in bios

[TatamiMatt]

5 minutes ago, g335 said:

DOCP/EXPO 1 is what it says in bios

so DOCP should have profiles and you might be able to set a 6000CL30 profile thats automatically loaded as that in bios, if it is EXPO theyll have to be set manually, if you dont want to tinker with timings to set 6000CL30 manually and 6400 is running smoothly then id say stick with that

[Somerandomtechyboi]

ill just summon @RONOTHAN##

For ram benchmarks im not that sure which ones are actually decent, all i know is spi, wprime, y cruncher vst, and linpack

 

As for ram freq if you can run 6400 in 1:1 vsoc set to 1.3v then keep it at that, otherwise downclock to 6200 or 6000 and set trfc to 500

[g335]

3 minutes ago, TatamiMatt said:

so DOCP should have profiles and you might be able to set a 6000CL30 profile thats automatically loaded as that in bios, if it is EXPO theyll have to be set manually, if you dont want to tinker with timings to set 6000CL30 manually and 6400 is running smoothly then id say stick with that

After I installed the ram and the system booted up fine, I went into bios.  Bios had it set automatically at 6000mhz.  I have to set it at DOCP1/EXPO1.

[g335]

Just now, Somerandomtechyboi said:

ill just summon @RONOTHAN##

For ram benchmarks im not that sure which ones are actually decent, all i know is spi, wprime, y cruncher vst, and linpack

 

As for ram freq if you can run 6400 in 1:1 vsoc set to 1.3v then keep it at that, otherwise downclock to 6200 or 6000 and set trfc to 500

So I look for something called vsoc and set it to 1:1 and 1.3v.   Or if downclock set to trfc 500.  What is vsoc and what is trfc?

[Somerandomtechyboi]

7 minutes ago, g335 said:

So I look for something called vsoc and set it to 1:1 and 1.3v.   Or if downclock set to trfc 500.  What is vsoc and what is trfc?

Vsoc = soc voltage for the imc and the max is 1.3v

 

Trfc is a subtiming found in the ram timings submenu, 500 should work for your sticks since theyre hynix a die, could go lower like 480 or 470 but 500 is conservative enough that itll work for most hynix a dies and alot better than the 600-700+ the mobo would probably auto set

 

For 1:1 i was reffering to uclk so that should be in 1:1 mode

[RONOTHAN##]

5 hours ago, g335 said:

Why would one software show latency 51 and the other 87?

They're different benchmarks that calculate it differently. Practically speaking however, most benchmarks that give memory latency in a ns value aren't actually that good. Aida's specifically is known for bouncing all over the place and not lining up with reality for memory performance, I haven't seen too much with Passmark relating to real life. 

 

5 hours ago, g335 said:

Will keeping the ram on 6400mhz affect performance in a bad way?

Yes, it's very likely because it will default to UCLK == MEMCLK/2 mode (also referred to as 2:1 mode or Gear 2). This causes the memory controller to run at half the speed of the memory, resulting in a massive memory latency spike but the ability to run much higher frequency. Howeer, unless you get DDR5 7600+ to work, the latency disadvantage of 2:1 mode is not worth the extra clock speed, and given you're running dual rank memory you're not likely to get much above 6800MT/s, so you want to avoid 2:1 mode if possible. 

 

DDR5 6000, on the other hand, will default to 1:1 mode and have significantly better memory latency as a result, so it's much more likely to perform well. You can manually try setting it to 1:1 mode when going for higher frequency as most chips should do 6200 and a handful can do 6400MT/s in 1:1 mode. This setting on ASUS boards is somewhere buried in the AMD overclocking menu on in the advanced tab, I forget where, but if you manually set that to 1:1 mode you can try going for higher frequencies without worrying about memory performance decreasing, just know that there's a good chance your CPU won't work at that frequency. Increasing the SOC voltage can help, 1.3V is the maximum, but if that doesn't work, drop down to 6200MT/s, then ditch it all together and just go for 6000MT/s.

 

As for the timings, you can do custom sub-timings if you really feel like it, but practically speaking since this is a production rig I wouldn't bother. Just leave it at the XMP timings no matter what and you should be fine. 

 

4 hours ago, Somerandomtechyboi said:

Trfc is a subtiming found in the ram timings submenu, 500 should work for your sticks since theyre hynix a die, could go lower like 480 or 470 but 500 is conservative enough that itll work for most hynix a dies and alot better than the 600-700+ the mobo would probably auto set

They don't have A die, they have 24Gb M die, which does significantly worse tRFC than A die or M die. The best you can even dream of with those would be something like 550, though realistically it will need to be in the 600-700 range. 

[Somerandomtechyboi]

10 hours ago, RONOTHAN## said:

They don't have A die, they have 24Gb M die, which does significantly worse tRFC than A die or M die. The best you can even dream of with those would be something like 550, though realistically it will need to be in the 600-700 range. 

Aaand im being a retard and forgot that its a 2x48gb kit and not a 2x32gb kit

 

Still though manually setting to that 600-700 range will probs still be better than whatever the hell the board auto sets

[g335]

15 hours ago, Somerandomtechyboi said:

Vsoc = soc voltage for the imc and the max is 1.3v

 

Trfc is a subtiming found in the ram timings submenu, 500 should work for your sticks since theyre hynix a die, could go lower like 480 or 470 but 500 is conservative enough that itll work for most hynix a dies and alot better than the 600-700+ the mobo would probably auto set

 

For 1:1 i was reffering to uclk so that should be in 1:1 mode

 

11 hours ago, RONOTHAN## said:

They're different benchmarks that calculate it differently. Practically speaking however, most benchmarks that give memory latency in a ns value aren't actually that good. Aida's specifically is known for bouncing all over the place and not lining up with reality for memory performance, I haven't seen too much with Passmark relating to real life. 

 

Yes, it's very likely because it will default to UCLK == MEMCLK/2 mode (also referred to as 2:1 mode or Gear 2). This causes the memory controller to run at half the speed of the memory, resulting in a massive memory latency spike but the ability to run much higher frequency. Howeer, unless you get DDR5 7600+ to work, the latency disadvantage of 2:1 mode is not worth the extra clock speed, and given you're running dual rank memory you're not likely to get much above 6800MT/s, so you want to avoid 2:1 mode if possible. 

 

DDR5 6000, on the other hand, will default to 1:1 mode and have significantly better memory latency as a result, so it's much more likely to perform well. You can manually try setting it to 1:1 mode when going for higher frequency as most chips should do 6200 and a handful can do 6400MT/s in 1:1 mode. This setting on ASUS boards is somewhere buried in the AMD overclocking menu on in the advanced tab, I forget where, but if you manually set that to 1:1 mode you can try going for higher frequencies without worrying about memory performance decreasing, just know that there's a good chance your CPU won't work at that frequency. Increasing the SOC voltage can help, 1.3V is the maximum, but if that doesn't work, drop down to 6200MT/s, then ditch it all together and just go for 6000MT/s.

 

As for the timings, you can do custom sub-timings if you really feel like it, but practically speaking since this is a production rig I wouldn't bother. Just leave it at the XMP timings no matter what and you should be fine. 

 

They don't have A die, they have 24Gb M die, which does significantly worse tRFC than A die or M die. The best you can even dream of with those would be something like 550, though realistically it will need to be in the 600-700 range. 

I went into bios and without changing anything here is what bios shows

 

DOCP1/EXPO1

Memory Frequency DDR5 6400

FCLK Frequency Auto

CPU Core Voltage 1.440V-1.432V Auto

CPU SOC Voltage 1.320V  Auto

DRAM VDD Voltage 1.35000

DRAM VDDQ Voltage 1.3500

Memory Timings

CL32

TRCD39

TRP39

TRAS102

UCLK is set to Auto

TRFC is set to Auto

 

Notes in bios says DOCP1/EXPO1 Load the DIMMS default DOCP/EXPO memory timings and other memory parameters optimized by ASUS

 

DOCP/EXPO Tweaked Load DOCP/EXPO profile with tweaks for improved performance if config matches

 

DCOP2/EXPO2 Load the DIMMS complete default DOCP/EXPO profile

 

So seems to be fine. Do I need to change anything?

 

I can change ram speed but do I need to?

 

Whats the big differences between the DOCP/EXPO settings?

 

Is it worth trying the tweaked option?

 

[RONOTHAN##]

12 minutes ago, g335 said:

UCLK is set to Auto

This you want to manually set to 1:1 if it isn't already. If you're only going to touch one thing, have it be this. This is also the most likely to cause instability, and if it does, drop the memory speed down to 6200 or 6000.

 

13 minutes ago, g335 said:

Whats the big differences between the DOCP/EXPO settings?

DOCP is the open source name for XMP or the Intel optimized RAM overclock. EXPO is the AMD optimized RAM overclock. In practice one works just fine on the other. 

TL;DR: Marketing. 

 

15 minutes ago, g335 said:

Is it worth trying the tweaked option?

Since this is a production system and the stress testing would take 1-2 days for me to trust it 24/7 because of how much RAM you have, no. Practically, I'd drop the speed down to 6000 and be done with it because I wouldn't want a production system to be out of commission for that long. 

[Somerandomtechyboi]

Just now, g335 said:

CPU SOC Voltage 1.320V  Auto

Uhhh i think youll wanna update your bios cause it shouldnt be going over 1.3v and thats because of the burning ryzen 7000 shenanigans a few months ago due to >1.3v vsoc

 

3 minutes ago, g335 said:

UCLK is set to Auto

This is set by uclk div1 mode setting located in the ai tweaker menu which you wanna set to uclk = memclk which is running the uclk at the same freq as memclk which in this case is 3200 since the rams are at a real freq of 3200 (advertised at 6400 since they effectively run at 6400 due to ddr double data rate)

 

13 minutes ago, g335 said:

TRFC is set to Auto

Located in the dram timing control submenu in the ai tweaker menu

 

Cant find any screenshots highlightning the subs for now so id just stick with 650 or 700trfc

 

3 minutes ago, RONOTHAN## said:

Since this is a production system and the stress testing would take 1-2 days for me to trust it 24/7 because of how much RAM you have, no. Practically, I'd drop the speed down to 6000 and be done with it because I wouldn't want a production system to be out of commission for that long. 

^^^

Unless the system is for whatever reason left unused for a few days

 

Probably just update bios first and foremost, div1 set to uclk = memclk, trfc set to 700, and be done with it

[g335]

2 minutes ago, RONOTHAN## said:

This you want to manually set to 1:1 if it isn't already. If you're only going to touch one thing, have it be this. This is also the most likely to cause instability, and if it does, drop the memory speed down to 6200 or 6000.

 

DOCP is the open source name for XMP or the Intel optimized RAM overclock. EXPO is the AMD optimized RAM overclock. In practice one works just fine on the other. 

TL;DR: Marketing. 

 

Since this is a production system and the stress testing would take 1-2 days for me to trust it 24/7 because of how much RAM you have, no. Practically, I'd drop the speed down to 6000 and be done with it because I wouldn't want a production system to be out of commission for that long. 

If it is stable now and the timings are good and the voltages, I dont know if I want to mess with the UCLK.  What are some signs to look for that would say to change the UCLK to 1:1?

 

You would run a stress test for 24 hours?

[g335]

2 minutes ago, Somerandomtechyboi said:

Uhhh i think youll wanna update your bios cause it shouldnt be going over 1.3v and thats because of the burning ryzen 7000 shenanigans a few months ago due to >1.3v vsoc

 

This is set by uclk div1 mode setting located in the ai tweaker menu which you wanna set to uclk = memclk which is running the uclk at the same freq as memclk which in this case is 3200 since the rams are at a real freq of 3200 (advertised at 6400 since they effectively run at 6400 due to ddr double data rate)

 

Located in the dram timing control submenu in the ai tweaker menu

 

Cant find any screenshots highlightning the subs for now so id just stick with 650 or 700trfc

 

^^^

Unless the system is for whatever reason left unused for a few days

 

Probably just update bios first and foremost, div1 set to uclk = memclk, trfc set to 700, and be done with it

So Auto for the UCLK means the ram is at 3200?  I dont understand, the bios and even 3DMark shows my ram at 6400 speed.   So how is at 3200?  What is happening if I dont change the UCLK from Auto?  If I am not lowering the speed, why do I need to change the TRFC?

 

I have the bios update from December 2023.

[RONOTHAN##]

4 minutes ago, Somerandomtechyboi said:

Uhhh i think youll wanna update your bios cause it shouldnt be going over 1.3v and thats because of the burning ryzen 7000 shenanigans a few months ago due to >1.3v vsoc

ASUS reads the SOC voltage closer to the VRM than the silicon, 1.32V corresponds to 1.3V at the silicon. 

 

4 minutes ago, g335 said:

What are some signs to look for that would say to change the UCLK to 1:1?

A massive latency penalty. Every single Ryzen 7000 CPU should be able to do 6000MT/s with two DIMMs, so that's a safe bet, and that's in 1:1 mode. You can check if it's in 2:1 mode inside Windows with HWInfo, there will be a UCLK reading somewhere that you can then compare to the MCLK setting (memory frequency) to see if they're in a 1:1 or 2:1 ratio. I'd just set it to 1:1 permanently to be safe though. 

 

6400 2:1 mode is slower than 6000 1:1 mode, so if you don't want to stress test, drop to 6000 in 1:1 mode. 

 

8 minutes ago, g335 said:

You would run a stress test for 24 hours?

For a production system, 24 hours is the minimum I'd want. Realistically, for me to 100% trust that amount of RAM with any amount of tune on it, I'd want more like 48-72 hours of stress testing, so set it to run over the weekend. Not practical a lot of the time. 

 

If this was just a gaming system with 32-48GB of RAM, 4-6 hours is usually good enough, 24 hours if you want to be positive it's fine, but with over double the memory and for a system where blue screens cost you more money than the minutes it would save on render time, you want to test more than not. 

 

3 minutes ago, g335 said:

So how is at 3200?

3200MHz == 6400MT/s. RAM manufacturers advertise the data rate and therefore it's what most people communicate in, but the actual frequency is half of the data rate (6400Mt/s is the data rate). 

 

4 minutes ago, g335 said:

So Auto for the UCLK means the ram is at 3200? 

Not really. Auto UCLK usually translates to defaulting to 2:1 mode for speeds over 6000MT/s, so the UCLK would actually be running at just 1600MHz. 

 

6 minutes ago, g335 said:

If I am not lowering the speed, why do I need to change the TRFC?

Just don't worry about the sub-timings, if I were in your situation I would not touch them. Ignore anything said about tRFC, tREFI, tRAS, or anything else that has a t- in front. 

[g335]

12 minutes ago, RONOTHAN## said:

ASUS reads the SOC voltage closer to the VRM than the silicon, 1.32V corresponds to 1.3V at the silicon. 

 

A massive latency penalty. Every single Ryzen 7000 CPU should be able to do 6000MT/s with two DIMMs, so that's a safe bet, and that's in 1:1 mode. You can check if it's in 2:1 mode inside Windows with HWInfo, there will be a UCLK reading somewhere that you can then compare to the MCLK setting (memory frequency) to see if they're in a 1:1 or 2:1 ratio. I'd just set it to 1:1 permanently to be safe though. 

 

6400 2:1 mode is slower than 6000 1:1 mode, so if you don't want to stress test, drop to 6000 in 1:1 mode. 

 

For a production system, 24 hours is the minimum I'd want. Realistically, for me to 100% trust that amount of RAM with any amount of tune on it, I'd want more like 48-72 hours of stress testing, so set it to run over the weekend. Not practical a lot of the time. 

 

If this was just a gaming system with 32-48GB of RAM, 4-6 hours is usually good enough, 24 hours if you want to be positive it's fine, but with over double the memory and for a system where blue screens cost you more money than the minutes it would save on render time, you want to test more than not. 

 

3200MHz == 6400MT/s. RAM manufacturers advertise the data rate and therefore it's what most people communicate in, but the actual frequency is half of the data rate (6400Mt/s is the data rate). 

 

Not really. Auto UCLK usually translates to defaulting to 2:1 mode for speeds over 6000MT/s, so the UCLK would actually be running at just 1600MHz. 

 

Just don't worry about the sub-timings, if I were in your situation I would not touch them. Ignore anything said about tRFC, tREFI, tRAS, or anything else that has a t- in front. 

Thanks everyone.

 

I downloaded and used HWinfo, it shows the MCLK at 32.00mhz, and  UCLK at 16.00. So that means change the UCLK in bios?  If I dont change the UCLK,  what situations would the latency affect performance? Gaming?  3D animation and rendering?

 

What stress software would you suggest? The ones mentioned earlier are all from 2010 and 2013.  No updates.

[RONOTHAN##]

Just now, g335 said:

it shows the MCLK at 32.00mhz, and  UCLK at 16.00. So that means change the UCLK in bios?

Yes, change the UCLK mode to UCLK == MEMCLK. 

 

Just now, g335 said:

If I dont change the UCLK,  what situations would the latency affect performance? Gaming?  3D animation and rendering?

Gaming quite a bit, and probably the animation and rendering though I have less of an intuition about those. 

 

1 minute ago, g335 said:

What stress software would you suggest? The ones mentioned earlier are all from 2010 and 2013.  No updates.

The ones I've had the best luck with at finding memory issues on DDR5 are TestMem5 with the 1usmus_v3 preset (you want to increase the number of loops to something like 100 to run for ~24-48 hours) and Y-Cruncher with the VT3 stress test. Both of those together is what I run on my personal system, with 2-3 hours of VT3 usually being plenty and the rest being TestMem5. 

[g335]

Just now, RONOTHAN## said:

Yes, change the UCLK mode to UCLK == MEMCLK. 

 

Gaming quite a bit, and probably the animation and rendering though I have less of an intuition about those. 

 

The ones I've had the best luck with at finding memory issues on DDR5 are TestMem5 with the 1usmus_v3 preset (you want to increase the number of loops to something like 100 to run for ~24-48 hours) and Y-Cruncher with the VT3 stress test. Both of those together is what I run on my personal system, with 2-3 hours of VT3 usually being plenty and the rest being TestMem5. 

Thanks!!!

[g335]

1 minute ago, RONOTHAN## said:

Yes, change the UCLK mode to UCLK == MEMCLK. 

 

Gaming quite a bit, and probably the animation and rendering though I have less of an intuition about those. 

 

The ones I've had the best luck with at finding memory issues on DDR5 are TestMem5 with the 1usmus_v3 preset (you want to increase the number of loops to something like 100 to run for ~24-48 hours) and Y-Cruncher with the VT3 stress test. Both of those together is what I run on my personal system, with 2-3 hours of VT3 usually being plenty and the rest being TestMem5. 

So after setting the UCLK, watch out for crashes and blue screens? Run the stability tests?

[g335]

4 minutes ago, RONOTHAN## said:

Yes, change the UCLK mode to UCLK == MEMCLK. 

 

Gaming quite a bit, and probably the animation and rendering though I have less of an intuition about those. 

 

The ones I've had the best luck with at finding memory issues on DDR5 are TestMem5 with the 1usmus_v3 preset (you want to increase the number of loops to something like 100 to run for ~24-48 hours) and Y-Cruncher with the VT3 stress test. Both of those together is what I run on my personal system, with 2-3 hours of VT3 usually being plenty and the rest being TestMem5. 

Why would they release ram, say it works with mb, say set to DOCP1/EXPO1 and then not say to change the UCLK?   The mb I have stock bios says it supports up to 6400mhz ram.  I updated the bios to the December 23 bios.   So that means people are buying ram and many not know to change the UCLK are not receiving the advertised speed?

[RONOTHAN##]

6 minutes ago, g335 said:

So that means people are buying ram and many not know to change the UCLK are not receiving the advertised speed?

Well, they are receiving the advertised speed, but it's getting worse performance than if they dropped down to 6000MT/s. Higher number does not equal faster with RAM, contrary to popular belief. 

 

As for everyhting else in that post, the entire RAM EXPO/XMP scene is complicated with all sorts of things that make no sense (8400MT/s rated kits, for instance basically never work). 

 

14 minutes ago, g335 said:

So after setting the UCLK, watch out for crashes and blue screens? Run the stability tests?

More or less, yeah. Run the stability tests more than anything if you want to keep 6400MT/s, and the first sign of instability just drop it down to 6200 or even 6000. 

[g335]

1 minute ago, RONOTHAN## said:

Well, they are receiving the advertised speed, but it's getting worse performance than if they dropped down to 6000MT/s. Higher number does not equal faster with RAM, contrary to popular belief. 

 

As for everyhting else in that post, the entire RAM EXPO/XMP scene is complicated with all sorts of things that make no sense (8400MT/s rated kits, for instance basically never work). 

 

More or less, yeah. Run the stability tests more than anything if you want to keep 6400MT/s, and the first sign of instability just drop it down to 6200 or even 6000. 

Thanks!!!