If you try to push real load such as TM5 or Y-cruncher they can hit even 60 degrees with mild OC (7200) with 1.4v. Not many people have real world memory use but for those that do or who want a rock solid OC something like a "Memory cooling bracket" which hangs off the motherboard screw mounts from aliexpress can keep temps down by 10+ degrees. If you replace the 50mm fans they come with something pwm the noise might even be as good as a custom 120mm fan bracket.
RAM Coolers
- Thread starter Alexrose1uk
- Start date
Makes sense, I guess for overclocking it makes a lot of sense
Maybe I should update the BIOS first Gigabyte might have fixed the DDR voltage issue, their response to my support tickets doesn't suggest they'd be able to fix themselves a sandwich though
Well Blow me, I updated the BIOS and I can indeed now give it more than its stock 1.45v, of course all my config wiped and I now need to see what she'll do unleashed, hours of fun
. I at least wrote the primaries into the custom EXPO SPD, so its half way there.I wonder if they have fixed the WOL switches that didn't work and allowed magic packets from Apple TV to keep turning the bloody machine on, lets not get too excited
Last edited:
I did have a go but no luck, couldn't get 8000 to post @ a 40-48-48 and 1.65 nor could I get more than 6200 @ CL26, so i'll just be happy with my lot, it's pretty fast. 6400@c26 felt like it was on bit after an hour or so of Karhu it threw an error.
Last edited:
To be fair 6200@CL26 is a great clock, and certainly better latency compared to 8000 @CL34.
I am on a 2nd 9950X3d (the other was RMA'd for other reasons), and the first could not go above 7800; let alone be stable. Roll on BF6!
I am on a 2nd 9950X3d (the other was RMA'd for other reasons), and the first could not go above 7800; let alone be stable. Roll on BF6!
Last edited:
I was hoping with this CL26 binned RAM I'd be able to get 8000 with a CL 34/32 which would be the same or better latency than 6000CL26 but with additional bandwidth, but it wasn't to be.
Like you say 6200CL26 is not terrible though
and for me the whole system feels like such an upgrade from my old 5950x, it's the first time in a long time I have felt an upgrade was actually worthwhile.
Like you say 6200CL26 is not terrible though
and for me the whole system feels like such an upgrade from my old 5950x, it's the first time in a long time I have felt an upgrade was actually worthwhile.
Last edited:
- Joined
- 26 Apr 2004
- Posts
- 9,941
- Location
- Milton Keynes
Your latency is still beating mine; I've managed to get within around 1.5ns of you now though with continued tweaking. My memory controller doesnt seem to want to play that nice with 6200/6400, but I've got 6060 (1% Bclk bump) fairly stable
Not been able to match my prior numbers from Aida on latency, so something has changed with my updated tweaking, similar ball park 62-63 but better bandwidth, it's good enough, takes too long to test all these things though as I don't know what I am doing, just making number smaller, test, crash, reboot
Not tried a bclk actually but as this can run asynchronous now it might be worth looking into when I next get bored and want to tinker.
Not tried a bclk actually but as this can run asynchronous now it might be worth looking into when I next get bored and want to tinker.
- Joined
- 26 Apr 2004
- Posts
- 9,941
- Location
- Milton Keynes
My day to day is essentially tuned 6060C26, with the 9800X3D boosting to 5.5GHz (+200PBO, +1% BCLK), and IF at ~2150. I'm seeing 62-63.x NS on average using AIDAs test.
It's taken quite a bit of tuning and testing to get to what I'd consider stable, and the extra 1% BCLK threw me a red herring by making it look like my 9800X3D undervolt was unstable, but now its appearing more like the 1% bclk memory speed increase COMBINED with the undervolt was causing a slight deficiency in the RAM subsystem and causing an occasional memory error. Bumped RAM from 1.55v to 1.6V (which I wouldn't have done without a cooler), and I'm now error free and dialing my undervolt back in (as I'd dropped as low as -5 curve optimiser), with with a good 12hr+ AIDA stress stability test each time I up the undervolt to ensure I've not incurred another issue. It's now looking like I might be able to get back up to -15 to 25, which had failed the 12hr AIDA before seemingly due to the RAM voltage.
I did have a bit of a play with the Core Tuning Config setting, Gigabyte seems to default to Level 2, Aida64 latency test prefers legacy, and Level 1 or Legacy seem to produce the best gaming results, although the difference isn't huge; so I left it to Legacy for now to help me dial the memory in using AIDA more easily.
I'd rather take the time to aggressively stress test (24hrs Karhu passed at these new RAM settings, along with multiple hours YCruncher VT3, OCCT AVX2 and TM5 Absolut all passed) and now just using AIDA64 Stress Stability 12hr test runs to validate and dial in the CPU undervolt configuration.
It's kinda funny, the more aggressive curve optimiser settings were generating quick fails (increasingly so when the optimiser was more agggressive), leading me to think it was the CPU UV when my RAM was still set to 1.55v, but multiple failed values have all passed 12 hours so far now the RAM has been bumped to 1.6, so it really does seem my bclk 1% JUST pushed the RAM/CPU/IMC subsystem to the edge on the voltage it had, and now that the RAM has been given an extra voltage threshold, the bottleneck has been resolved I should be able to pull a decent UV back. -15 has already passed again, and I'm validating -20 now. I might even be able to test -25/30 again if this continues to go well.
Either way happy to be dialing it in, as if stable that'll help bring heat and power consumption down a bit, and allow it to boost that bit harder and longer (oy vey!)
Minimimum framerates in CPU bottlenecked situations are easily pushing another 5% over stock in many cases (for example if I run it through Shadow of the Tomb Raider 1080p low), and Total Warhammer 3 average FPS in the heaviest included benchmark went from 131 to mid 140s (145-147), so there is definately some real world benefits being pulled out of this. Kinda shows up the *******s the mainstream tech youtube world pushes that boost does it all these days, and that CPU and RAM tuning/speed is pointless; yes don't get me wrong it doesn't come with the 25+% gains it used to, and it's time consuming, but a load of rot to suggest there's nothing left, and I have severe doubts about some of the videos I've seen showing that RAM speed does nothing, as I think there'll be a point where the additional bandwidth will offset the 2:1 divider, but as MY particular memory controller doesnt really seem to like much above 6000, I'll leave it until later platforms to push that further.
Anyone with a brain or who's just been in this space long enough knows there are still extra levers and knobs to tune, as there's always something left on the table in the name of ensuring there's somewhat of an overhead at stock. Expo/my motherboard actually seems to set somewhat reasonable timings for my settings too, but even they were able to be tuned further.
My 9070XT unfortunately isnt as blessed, the VRAM seems very temperamental in regards to triggering black screens once moved from stock, but there's still a bit in the tank in regards to undervolting and allowing more power to the core as well.
Honestly, once I've finished dialing in my undervolt, I'll be able to finally take some time to actually properly tweak the fan profiles to reduce the noise, as that's all been left on auto, and at the very least I want the AIO pump and RAM cooler to run at a static RPM; and tune down the fans a little as they have a definitive drone at load and even idle; I just can't remember what's plugged into exactly which header so I'll likely need to take the back panel off to check.
This all being said though, much as I AM taking some serious time to do it, this is all for gaming and day-to-day performance. I want stability, and I'm not interested in extreme OC/benchmark chasing, just want my day to day to be as snappy as possible, and get my minimum FPS as high as I can, and frametimes as stable as possible.
I still see a lot of people chasing synthetic benchmarks, or pushing unstable overclocks, which'll error out randomly; and that just doesn't interest me.
With great power comes great stability
It's taken quite a bit of tuning and testing to get to what I'd consider stable, and the extra 1% BCLK threw me a red herring by making it look like my 9800X3D undervolt was unstable, but now its appearing more like the 1% bclk memory speed increase COMBINED with the undervolt was causing a slight deficiency in the RAM subsystem and causing an occasional memory error. Bumped RAM from 1.55v to 1.6V (which I wouldn't have done without a cooler), and I'm now error free and dialing my undervolt back in (as I'd dropped as low as -5 curve optimiser), with with a good 12hr+ AIDA stress stability test each time I up the undervolt to ensure I've not incurred another issue. It's now looking like I might be able to get back up to -15 to 25, which had failed the 12hr AIDA before seemingly due to the RAM voltage.
I did have a bit of a play with the Core Tuning Config setting, Gigabyte seems to default to Level 2, Aida64 latency test prefers legacy, and Level 1 or Legacy seem to produce the best gaming results, although the difference isn't huge; so I left it to Legacy for now to help me dial the memory in using AIDA more easily.
I'd rather take the time to aggressively stress test (24hrs Karhu passed at these new RAM settings, along with multiple hours YCruncher VT3, OCCT AVX2 and TM5 Absolut all passed) and now just using AIDA64 Stress Stability 12hr test runs to validate and dial in the CPU undervolt configuration.
It's kinda funny, the more aggressive curve optimiser settings were generating quick fails (increasingly so when the optimiser was more agggressive), leading me to think it was the CPU UV when my RAM was still set to 1.55v, but multiple failed values have all passed 12 hours so far now the RAM has been bumped to 1.6, so it really does seem my bclk 1% JUST pushed the RAM/CPU/IMC subsystem to the edge on the voltage it had, and now that the RAM has been given an extra voltage threshold, the bottleneck has been resolved I should be able to pull a decent UV back. -15 has already passed again, and I'm validating -20 now. I might even be able to test -25/30 again if this continues to go well.
Either way happy to be dialing it in, as if stable that'll help bring heat and power consumption down a bit, and allow it to boost that bit harder and longer (oy vey!)
Minimimum framerates in CPU bottlenecked situations are easily pushing another 5% over stock in many cases (for example if I run it through Shadow of the Tomb Raider 1080p low), and Total Warhammer 3 average FPS in the heaviest included benchmark went from 131 to mid 140s (145-147), so there is definately some real world benefits being pulled out of this. Kinda shows up the *******s the mainstream tech youtube world pushes that boost does it all these days, and that CPU and RAM tuning/speed is pointless; yes don't get me wrong it doesn't come with the 25+% gains it used to, and it's time consuming, but a load of rot to suggest there's nothing left, and I have severe doubts about some of the videos I've seen showing that RAM speed does nothing, as I think there'll be a point where the additional bandwidth will offset the 2:1 divider, but as MY particular memory controller doesnt really seem to like much above 6000, I'll leave it until later platforms to push that further.
Anyone with a brain or who's just been in this space long enough knows there are still extra levers and knobs to tune, as there's always something left on the table in the name of ensuring there's somewhat of an overhead at stock. Expo/my motherboard actually seems to set somewhat reasonable timings for my settings too, but even they were able to be tuned further.
My 9070XT unfortunately isnt as blessed, the VRAM seems very temperamental in regards to triggering black screens once moved from stock, but there's still a bit in the tank in regards to undervolting and allowing more power to the core as well.
Honestly, once I've finished dialing in my undervolt, I'll be able to finally take some time to actually properly tweak the fan profiles to reduce the noise, as that's all been left on auto, and at the very least I want the AIO pump and RAM cooler to run at a static RPM; and tune down the fans a little as they have a definitive drone at load and even idle; I just can't remember what's plugged into exactly which header so I'll likely need to take the back panel off to check.
This all being said though, much as I AM taking some serious time to do it, this is all for gaming and day-to-day performance. I want stability, and I'm not interested in extreme OC/benchmark chasing, just want my day to day to be as snappy as possible, and get my minimum FPS as high as I can, and frametimes as stable as possible.
I still see a lot of people chasing synthetic benchmarks, or pushing unstable overclocks, which'll error out randomly; and that just doesn't interest me.
With great power comes great stability
Last edited:
Yup , pretty much why when I found the sweetspot I just left it after dialling in some numbers from the internets and didn’t pursue alternates too far once stability tested, if more volts can’t stabilise I’m moving on quickly, got games to be playing rather the rebooting and watching the mobo LED
I have been using a Dota2 benchmark to see if there are CPU improvements with mem tune, nice and repeatable, quick using a batch file and very low GPU usage even at 4k, so many hundreds of FPS
CO on my 9950x is a bit tragic, I get low load hard locks if I try -20, I have at least one core that doesn’t want to play ball, I really can’t be bother to tune it per core.
My 5080 VRAM OC has a lot of instability with temperature, cranking up the fans to near max to keep it near 60C it was fine but noisy, water block solved that. now it runs the core at over 3Ghz all day and +3Ghz on the VRAM
I have been using a Dota2 benchmark to see if there are CPU improvements with mem tune, nice and repeatable, quick using a batch file and very low GPU usage even at 4k, so many hundreds of FPS
CO on my 9950x is a bit tragic, I get low load hard locks if I try -20, I have at least one core that doesn’t want to play ball, I really can’t be bother to tune it per core.
My 5080 VRAM OC has a lot of instability with temperature, cranking up the fans to near max to keep it near 60C it was fine but noisy, water block solved that.
now it runs the core at over 3Ghz all day and +3Ghz on the VRAM
Last edited:
- Joined
- 26 Apr 2004
- Posts
- 9,941
- Location
- Milton Keynes
Honestly thats the one area where my Nitro does surprisingly well - yesterday I was throwing maximum settings 1440p Oblivion Remastered, with full RT effects out the wazoo, gpu running 100% for a while, and the card remained considerably quieter than my system fans (which admittedly even tuned arent quiet - I need to break away the front fans and set them to static), but main GPU temp was hitting around 58, with the hotspot at 85. Considering the old 3080 ran at ~85 core average, and the memory ran very hot 90+ easily; this seems very cool and collected; just unfortunately my VRAM doesnt seem to have much headroom in the tank.
Going back to the original topic though, I'm extremely glad I picked up this cooler for £20. Its allowed me to keep my ram below the 60 degree safety threshold, and its usually considerably below that unless I'm hard stressing the machine; and it's allowed me to push the voltage I needed to stabilise my settings, without risking long term damage or degredation.
At 80% static its hitting a little over 3000RPM, but I cannot hear them at all over the rest of the system fans, where the front P14 Pros are by far and away the loudest fans in the system.
Going back to the original topic though, I'm extremely glad I picked up this cooler for £20. Its allowed me to keep my ram below the 60 degree safety threshold, and its usually considerably below that unless I'm hard stressing the machine; and it's allowed me to push the voltage I needed to stabilise my settings, without risking long term damage or degredation.
At 80% static its hitting a little over 3000RPM, but I cannot hear them at all over the rest of the system fans, where the front P14 Pros are by far and away the loudest fans in the system.
Last edited: