New Ryzen 7 9700X, RTX 4080, 9700X build - EXPO not working

[_roman_]

I'm sorry - RMA the ram Kit. I just checked the webpage for the RAM number.

Per definition the ram kit should run only with the xmp profile.

• with a Ryzen 9000 cpu (the topic poster has a 9700X)
• with that ram partnumber
• with that mainboard + ryzen 9000 cpu

I'm just curious

Part number from post #1

F5-6000J3040G32GX2-TZ5NR - Specification - G.SKILL International Enterprise Co., Ltd.

Trident Z5 Neo RGB DDR5-6000 CL30-40-40-96 1.40V 64GB (2x32GB) AMD EXPO

www.gskill.com

Page says

• Reaching the rated XMP/EXPO overclock speed and system stability will depend on the compatibility and capability of the motherboard and CPU used.

page says

Check memory compatibility for your motherboard with the RAM Configurator tool, or check the QVL for a list of compatible motherboards with this memory kit.

-> https://www.gskill.com/qvl/165/390/1665020865/F5-6000J3040G32GX2-TZ5NR-QVL

the mainboard is on that page

• AMD X870 (Ryzen 9000 series CPUs)
• next line: ROG STRIX X870-F GAMING WIFI

when gskill make such claims. In my point of view. After a cmos reset and only activating the expo profile the box should run with that processor, ram and mainboard.

 

[LittleBro]

The main issue however is trying to find out if it's the CPU IMC, mobo or ram itself that is the culprit, RMA may not be needed anyway as in the UK and EU we have consumer laws that allows you to return any items purchased within 14 days, outside of this it could be tricky if they test the ram and it runs at its rated speed and the RMA is denied, but there's a lot of assuming going on now so I'll just stfu

So there in UK you can return unpacked and used goods within 14 days return period?
Here in EU goods must NOT be used and must have original and undamaged packaging.
So any sane owner of a shop should show you middle finger when they see sticker is done.

Either RMA that kit, or try your luck with RMA of CPU.

As others pointed out, official supported memory speed is in fact officially supported, anything higher is OC.
It's generally known that those chips can handle more. It's a silicon lottery. But not even reaching 6000 MHz on IMC is a rarity, though.

How did you set infinity fabric clock and memory controller clock back then before touching anything else? AFAIK, EXPO profile only sets few timings, voltage and DDR clock, nothing else.

Update your board's BIOS, there have already been 5 performance improvement BIOS updates released since the launch of the board.

 

[Mr_Engineer]

Yes I am running on the latest BIOS.

I have settled on the rated speeds (6000-30-40-40-96) with higher voltages, and tweaks to other timings. The PC has been up and running stress tests since yesterday without any issues. Below are the AIDA64 results with look much better too:

Notice that the L1, L2, L3 Cache numbers are much better, that's because CPU Boost is enabled this time "and" 105W mode. (Enabling 105W mode by itself didn't make much of a difference in AIDA64)

Memory latency went down too, and the only difference that I am aware of that may have improved this is increasing the tREFI from 5000 to 50000.

I'll do some game testing today and report here.

 

[LittleBro]

Yes I am running on the latest BIOS.

I have settled on the rated speeds (6000-30-40-40-96) with higher voltages, and tweaks to other timings.

Notice that the L1, L2, L3 Cache numbers are much better, that's because CPU Boost is enabled this time "and" 105W mode.

Why did you have CPU boost feature disabled? It is always enabled by default.

Memory latency went down too, and the only difference that I am aware of that may have improved this is increasing the tREFI from 5000 to 50000.

I'll do some game testing today and report here.

The latency is a bit too high for DDR5-6000 and CL30, you should be able to achieve something at least around 65 ns. Try lowering tRP to 36, tRAS to 60, tRC to 96.
Were it unstable, I'd increase VDIMM from 1.400V to 1.425V. This should also allow for tighter other primaries (30-38-38).

 

[Mr_Engineer]

Why did you have CPU boost feature disabled? It is always enabled by default.

I think it was on Auto, but I forced it to "Enabled".

The PC ran fine with tighter timings (30-38-38) and VDIMM @ 1.400V. But even with those timings latency was over 80ns, but I'll double check/test again to confirm. I believe the tREFI setting is what lowered the latency.

Also I might be getting slightly higher latency due to the 64GB size. (I could be wrong)

On Cinebench 2024 I am getting 137 (SC) and 1312 (MC) scores which is about average for a Ryzen 9700x (@105W)

 

[rv8000]

Why did you have CPU boost feature disabled? It is always enabled by default.

The latency is a bit too high for DDR5-6000 and CL30, you should be able to achieve something at least around 65 ns. Try lowering tRP to 36, tRAS to 60, tRC to 96.
Were it unstable, I'd increase VDIMM from 1.400V to 1.425V. This should also allow for tighter other primaries (30-38-38).

He has GDM enabled, thats a hefty latency penalty. I forget the exact comparison but itd be like running cl 32 with GDM off. Latency appears correct for CL 30 GDM on. As long as tphy timings are matching between channels (also a latency penalty if mismatched).

 

[Mr_Engineer]

He has GDM enabled, thats a hefty latency penalty. I forget the exact comparison but itd be like running cl 32 with GDM off. Latency appears correct for CL 30 GDM on. As long as tphy timings are matching between channels (also a latency penalty if mismatched).

I did try your suggestion to disable the GDM, but I think with the lower timings that I had at the time Windows wouldn't boot. Maybe needed slightly higher voltages?

I'll give it a try again later.

 

[rv8000]

I did try your suggestion to disable the GDM, but I think with the lower timings that I had at the time Windows wouldn't boot. Maybe needed slightly higher voltages?

I'll give it a try again later.

GDM off will need higher vdd/vddq probably 1.45-1.46 depending on kit quality.

 

[Mr_Engineer]

So I increased the vdd/vddq voltage to 1.435, tightened the timings to 30-38-38-38, disabled GDM, and increased tREFI to 65535. Below are AIDA results (mem latency still about the same though):

Cinebench 2024 gets slightly higher MC score of 1335.

 

[LittleBro]

He has GDM enabled, thats a hefty latency penalty. I forget the exact comparison but itd be like running cl 32 with GDM off. Latency appears correct for CL 30 GDM on. As long as tphy timings are matching between channels (also a latency penalty if mismatched).

I reached very good memory latency on DDR4 even with GDM on. I noticed that when GDM is turned off, system fails to boot randomly from time to time. GDM off requires tuning various voltages. I don't recommend gooing that way. As you can see above, it only helped very slightly. The biggest impact on latency have: tCL, tRFC, tREFI, then command rate, ProcODT, GDM. Someone suggests using CR2+GDM off rather than CR1+GDM on. I personally haven't experienced any significant difference, I'd say the differences were in margin for error.

Now, could you please raise tRFC to +30 ns higher and do a benchmark again? Yes, I meant to raise, not to lower the value.

 

[Mr_Engineer]

Now, could you please raise tRFC to +30 ns higher and do a benchmark again? Yes, I meant to raise, not to lower the value.

Ok, so I increased tRFC from 500 to 580, about 30ns higher. So far the AIDA results seem within margin of error:

I'll try a couple more tries with timings, then probably settle on something in between. Thanks.

 

[rv8000]

I reached very good memory latency on DDR4 even with GDM on. I noticed that when GDM is turned off, system fails to boot randomly from time to time. GDM off requires tuning various voltages. I don't recommend gooing that way. As you can see above, it only helped very slightly. The biggest impact on latency have: tCL, tRFC, tREFI, then command rate, ProcODT, GDM. Someone suggests using CR2+GDM off rather than CR1+GDM on. I personally haven't experienced any significant difference, I'd say the differences were in margin for error.

Now, could you please raise tRFC to +30 ns higher and do a benchmark again? Yes, I meant to raise, not to lower the value.

GDM on vs off isn’t going to produce margin of error differences; you also can’t conclude that from two posts where he changes several timings. GDM off provides significant latency benefits, as long as mclk is stable with the cpu sample, you will only have to adjust vdd/vddq.

Again, all of your rtt values should be left on auto, at ddr5 6000 theres no reason to be dumping BZoids settings in there.

If your kit is Hynix A-die: 120ns trfc is your min, 140ns is safe buffer (OP’s kit is likely a-die)

If your kit is Hynix M-die: 160ns trfc is your min, 170-180ns is a safe buffer.

CL, GDM off, and resfresh window (min trfc + max trefi) are the biggest changes you can make to improve latency; fclk overclocking can additionally improve latency but there are thresholds and while he’s adjusting timings there is no reason to touch fclk until stabilized.

 

[Mr_Engineer]

GDM on vs off isn’t going to produce margin of error differences; you also can’t conclude that from two posts where he changes several timings. GDM off provides significant latency benefits, as long as mclk is stable with the cpu sample, you will only have to adjust vdd/vddq.

Again, all of your rtt values should be left on auto, at ddr5 6000 theres no reason to be dumping BZoids settings in there.

If your kit is Hynix A-die: 120ns trfc is your min, 140ns is safe buffer (OP’s kit is likely a-die)

If your kit is Hynix M-die: 160ns trfc is your min, 170-180ns is a safe buffer.

CL, GDM off, and resfresh window (min trfc + max trefi) are the biggest changes you can make to improve latency; fclk overclocking can additionally improve latency but there are thresholds and while he’s adjusting timings there is no reason to touch fclk until stabilized.

I noticed the last few days that while I can play with most timings, if I reduce tRFC much lower than 500, the PC wouldn't even boot. I think even 400 didn't work (maybe I needed higher voltages).

The other issue is that, at my current settings, when running games/benchmarks, the DRAM temperature reaches up to 62C (144F). A bit high maybe?

I'll change the rtt settings to auto, increase the voltages and adjust other the timings a bit (including tRFC) and retest today

 

[rv8000]

I noticed the last few days that while I can play with most timings, if I reduce tRFC much lower than 500, the PC wouldn't even boot. I think even 400 didn't work (maybe I needed higher voltages).

The other issue is that, at my current settings, when running games/benchmarks, the DRAM temperature reaches up to 62C (144F). A bit high maybe?

I'll change the rtt settings to auto, increase the voltages and adjust other the timings a bit (including tRFC) and retest today

Reducing trefi to 50000 will help with temps. Leave your termination settings at auto as well as rtt values (boards are pretty good at handling these while @1:1). Ideally, under heavy load you want dimm temps to be at 50c +/- max, 62 is incredibly hot for low to medium loads such as games.

Not a direct comparison but at 6400 1.53vdd 1.44vddq, my dimms run around ~38c after 2-3hrs of games. Peak load temps during karhu/ycruncher/memtest are 48-50c. I’m using an nh-d15 with a single fan at max rpm.

If you have an aio that can be a reason for increased memory temps, as theres a lack of direct airflow around your memory dimms.

Tightening trfc will require higher voltages for sure, and will always vary by kit. 512 trfc should be relatively easy to achieve without blasting voltage. At 512 trfc and 50000 trefi you have already massively improve latency benefits for your refresh window, tightening timings any further won’t provide large enough benefits to offset temp and voltages increases - little tangible day to day benefits outside of epeen benchmarking.

 

[Mr_Engineer]

Yes I am using a 360 AIO. My case is Corsair 5000D which has openings from all sides for air flow and I have an extra fan added in the back, but I noticed the fans were running in quiet mode at only 40% speed. A Cinebench MC run would get the memory to about 59-60C, and the CPU would get caped at 95C.

After increasing the fan speed to 80% (a bit noisy), ran Cinebench again and the memory was only at 44-45C, and the CPU at around 83C. Much better temperatures

 

[Palladium]

My personal theory was it was voltage related but I'm not a huge fan of having people mess around with memory voltages or recommending that regardless of what is listed as safe online.

It's always preferable to have a kit that works out of the box.

My experience with mem OCing since 8700K is not to trust any of the voltages set on auto, especially on XMP. They are always boosted way beyond than what is needed for stability.

Besides if XMP/EXPO speeds are so bulletproof and easy to achieve they have been long baked into the JEDEC spec and technically covered under warranty, instead of these smoke and mirrors over what's really a OC guideline.

 

[oxrufiioxo]

My experience with mem OCing since 8700K is not to trust any of the voltages set on auto, especially on XMP. They are always boosted way beyond than what is needed for stability.

Besides if XMP/EXPO speeds are so bulletproof and easy to achieve they have been long baked into the JEDEC spec and technically covered under warranty, instead of these smoke and mirrors over what's really a OC guideline.

I haven't seen much of that on AM4/5 other than the whole soc voltage debacle. On intel it's been a problem for a long time Z390 was the worse but X99 was also problematic for me.

I agree though most motherboards do the opposite and set voltage higher than they need to be although it's been a while since I've had one set it to something I thought was unsafe my Z390 Code was the last board that did that.

 

[Mr_Engineer]

Reducing trefi to 50000 will help with temps. Leave your termination settings at auto as well as rtt values (boards are pretty good at handling these while @1:1). Ideally, under heavy load you want dimm temps to be at 50c +/- max, 62 is incredibly hot for low to medium loads such as games.

I tried changing the rtt values and termination settings to Auto, but no boot again.

The settings I found online by this guy (stable 128GB (4x32GB) DDR5-6000 CL30) are what work perfectly with my memory. I guess mine are also Hynix-A die.

 

[LittleBro]

You can't copy paste settings for 4 dimms to your 2 dimms. It does not work like that. Logically, because less dimms put less stress on IMC, you should be able to squeeze out more of your dimms than that guy but don't take it for granted. It's silicon lottery. CPU's IMC is silicon lottery. DIMMs are silicon lottery.

You dies are most probably Hynix M. If you could post photo of label on your dimm, we will know better.

Anyway, I found out when I was tightening my B-Dies, that after lowering tRFC to certain value, the latency did not keep lowering, but the opposite.

tRFC and tREFI are sensitive to high temps. Your temps are way too high for 1.4V. Let the air come to them or install additional cooling fan for them.

 

[Palladium]

I haven't seen much of that on AM4/5 other than the whole soc voltage debacle. On intel it's been a problem for a long time Z390 was the worse but X99 was also problematic for me.

I agree though most motherboards do the opposite and set voltage higher than they need to be although it's been a while since I've had one set it to something I thought was unsafe my Z390 Code was the last board that did that.

My MSI Z370 XMP was crazy, pushing 1.3V into SA and 1.25V into I/O for 2x16GB DDR4-3200C16 when the stock 1.05V/0.95V was enough to be stable.

I feel like PC stuff gets way overvolted out of the box since 2016. My GTX 1070 actually performed better in Witcher 3 at 1800MHz @ 0.8V max versus the stock curve drawing 150W.

 

[rv8000]

You can't copy paste settings for 4 dimms to your 2 dimms. It does not work like that. Logically, because less dimms put less stress on IMC, you should be able to squeeze out more of your dimms than that guy but don't take it for granted. It's silicon lottery. CPU's IMC is silicon lottery. DIMMs are silicon lottery.

You dies are most probably Hynix M. If you could post photo of label on your dimm, we will know better.

Anyway, I found out when I was tightening my B-Dies, that after lowering tRFC to certain value, the latency did not keep lowering, but the opposite.

tRFC and tREFI are sensitive to high temps. Your temps are way too high for 1.4V. Let the air come to them or install additional cooling fan for them.

They’re probably not m-die, none of the m-die kits I have can run rrds below 8, let alone post. Hwinfo should tell him what the ic’s are or typhoon. Most m-die are 24/48gb configs. You also probably cannot post at trfc 166ns at 1.435v or lower like he did.

@OP I’m surprised auto rtt’s and termination settings don’t work, if the previous settings were stable what works works I guess. I’ve literally only worked with single rank kits (7 various corsair, gskill and tg kits I currently have).

 

[Mr_Engineer]

You can't copy paste settings for 4 dimms to your 2 dimms. It does not work like that. Logically, because less dimms put less stress on IMC, you should be able to squeeze out more of your dimms than that guy but don't take it for granted. It's silicon lottery. CPU's IMC is silicon lottery. DIMMs are silicon lottery.

You dies are most probably Hynix M. If you could post photo of label on your dimm, we will know better.

Anyway, I found out when I was tightening my B-Dies, that after lowering tRFC to certain value, the latency did not keep lowering, but the opposite.

tRFC and tREFI are sensitive to high temps. Your temps are way too high for 1.4V. Let the air come to them or install additional cooling fan for them.

I get what you are saying, although somehow these settings are working well on mine so far.

The temps are better ofc after increasing the case fan speed, the highest memory temps I've seen are around 46C. But more and longer stress testing is in order.

They’re probably not m-die, none of the m-die kits I have can run rrds below 8, let alone post. Hwinfo should tell him what the ic’s are or typhoon. Most m-die are 24/48gb configs. You also probably cannot post at trfc 166ns at 1.435v or lower like he did.

@OP I’m surprised auto rtt’s and termination settings don’t work, if the previous settings were stable what works works I guess. I’ve literally only worked with single rank kits (7 various corsair, gskill and tg kits I currently have).

So right now I have settled on 28-38-38-40 timings at 6000, GDM off, and higher 1.50vdd 1.44vddq . AIDA numbers don't look that impressive but I noticed a bigger improvement in games (example: Black Myth Wukong). Below is the HWiNFO screenshot:

Cinebench 2024 (MC could be better I guess):

Black Myth: Wukong (Free Benchmark - Cinematic - 100% - 4K)

This is where the numbers went much higher. Initially, with the same graphics settings below, I got 31 fps average and 25 fps low. That was with 30-40-40-96 timings, GDM on, tREFI 5000 and fan speed at 40% (CPU was throttling at 95 C). With the improved timings and faster fans, I am getting 37 fps average and 28 fps lows which is a pretty good increase for such a demanding game.

I guess the performance is okay overall but I'll check the Benchmarks section to be able to compare.

Thanks guys for the help and suggestions, really appreciate it.

 

[LittleBro]

This is where the numbers went much higher. Initially, with the same graphics settings below, I got 31 fps average and 25 fps low. That was with 30-40-40-96 timings, GDM on, tREFI 5000 and fan speed at 40% (CPU was throttling at 95 C). With the improved timings and faster fans, I am getting 37 fps average and 28 fps lows which is a pretty good increase for such a demanding game.
View attachment 376202

I guess the performance is okay overall but I'll check the Benchmarks section to be able to compare.

Thanks guys for the help and suggestions, really appreciate it.

Are you sure about that fps increase? More than 20% by just adjusting RAM timings on Zen is quite ... well, miracle? There must be something other behind this, game update that fixes performance or GPU drivers update. With really aggresive timings you can squeeze 7-12% out of the kits, but 20+% is far too much.

 

[Mr_Engineer]

Are you sure about that fps increase? More than 20% by just adjusting RAM timings on Zen is quite ... well, miracle? There must be something other behind this, game update that fixes performance or GPU drivers update. With really aggresive timings you can squeeze 7-12% out of the kits, but 20+% is far too much.

I can double check and retest again. But I think it was the memory timings, voltages, and the CPU reaching 95C as my fans were in quiet mode. This game seems to be affected by both the CPU and GPU at 4K.

Probably if I drop a 9800x3d in my system with the same setting, it will get noticeably higher frame rates in this game even at 4K.

This video shows Expo vs tweaked timings make a big difference in games on Ryzen 9000 series: DDR5 : EXPO vs Subtiming Tweak // Test in 9 Games

 

[BitingChaos]

I know this is an old thread, but there are two things you could try.

1) You could try the previous 0606 BIOS.

2) Stick with 0706 but change VDCI to Predictive.

BIOS 0706 w/ AGESA 1.2.0.2b introduced new settings that cause timing issues with memory training.

If you go to AMD Overclocking > DDR and Infinity Fabric Frequency/Timings > Infinity Fabric Frequency and Dividers, you can change "Fclk VDCI Mode Pref" from "Auto" to "Predictive".
This change to VDCI (Voltage Dependent Clock Increment) will make BIOS 0706 work more like BIOS 0606 and offer better stability at high voltages & frequencies. .