|
Ostensibly this is for a 486 overclocking project I'm working on, I started messing around with peltier stuff. Power supply is from a older dell poweredge jumpered to run the fans slow, the thermocouple is only 70w but at 12 volts it's 6 amps. I'm guessing this old G860 is running around 15-20w at idle? It has a max TDP of 65w. The CPU i'm going to put it on only draws like 4 lol. ![]() ![]()
|
![]() |
|
![]()
|
# ? Jul 9, 2025 11:12 |
|
Little test POSTing![]()
|
![]() |
|
*sniffs* Mmmm, that vintage 1995 BIOS
|
![]() |
|
blown any caps yet
|
![]() |
|
Palladium posted:blown any caps yet Nope. I had an issue with some frost melting tonight and getting some water ingress, going to let it dry out overnight and check it in the morning. ![]() ![]()
|
![]() |
|
How does Duke3d run?
|
![]() |
|
Lockback posted:How does Duke3d run? I'll find out soonish I think. I'm focused on getting it set and memory timings. I think I had a run of doom at like 67 FPS so far.
|
![]() |
|
I'm going to resurrect this thread - after having been on Intel for multiple builds and 20 years I'm picking up a 7800x3d. Last time I had an AMD CPU it was a Phenom 955 black or something. Anyone have a good resource for learning about overclocking AMD stuff? Infinity Fabric? Thanks.
|
![]() |
|
I have the 5800x3d, the first 3d cache chip I think So lately at least on AMD you don't overclock at all anymore, but you can still get a minor boost out of under clocking (sounds counterintuitive, but true) I can manage a -30 on all cores which is the max and core cycler says it's stable, there have been several additional bios versions but I'm too scared to upgrade because generally it takes a long time to stability test and I won. Most people start with -25 all cores or 20 for real bad silicon samples maybe and when a core fails after awhile or overnight bump that one up 1 and try again. The 3d cache chips do not let you damage them by over clocking, that 3d cache is like a lid over the chip, but it will boost to max if you can keep the temps down with and under clock as long as you can do it stable to stay from crashes. Oh look into under clocking your GPU too if it's from the last 3-4 years it makes those run like 10-20 Celsius cooler and you can get like 95-100% of the speed or trade more for even cooler Quaint Quail Quilt fucked around with this message at 06:54 on Jul 14, 2024 |
![]() |
|
Quaint Quail Quilt posted:I have the 5800x3d, the first 3d cache chip I think Thanks, that's really interesting and I'll try to find some videos about this. I do know about underclocking in the GPU space but haven't because I'm using a custom water loop and because I really need all the FPS I can get for VR sim racing. When you say -30 do you mean CPU multiplier? Or do you mean mV and it's actually an undervolt not an underclock? Seeing as I shouldn't have any issues with cooling I don't know if there's a benefit in my use case except for economy with using less power.
|
![]() |
|
You can get better performance from an x3d with less voltage, it's true. Look up Ryzen curve optimization.
|
![]() |
|
Quaint Quail Quilt posted:I have the 5800x3d, the first 3d cache chip I think A negative curve offset is an overclock. You are increasing the clock at a given voltage on the V/F curve. It's actually exactly the same as a positive clock offset, which is how Nvidia frames the same thing. The only thing you can't do on X3D chips is increase the fmax, or maximum possible clock, but you aren't hitting that outside of strictly single threaded workloads. In practice, negative curve offsets are an all-core overclock.
|
![]() |
|
BurritoJustice posted:A negative curve offset is an overclock. You are increasing the clock at a given voltage on the V/F curve. It's actually exactly the same as a positive clock offset, which is how Nvidia frames the same thing. Cool, do you still have to adjust vCore and LLC etc? Does better performant RAM (both CL and overall speed) generally allow you just to use more of an offset?
|
![]() |
|
If I understand the concept correctly, it's the voltage curve that you're offsetting so you are tuning vCore/LLC already by adjusting it. RAM is a little more complicated because at a certain point the Infinity Fabric (CPU-chipset interconnect, which once upon a time when the chipset contained the memory controller would have been called a "front-side bus") can't keep up anymore and has to step back to a 1:2 ratio with the RAM, which hurts performance. The officially supported RAM clock only ever got up to 3200MT/s on AM4 - though you can go a bit higher, and are generally recommended to if you care. For Ryzen 5000-series I think 3600MT/s was usually a safe sweet spot to aim for, with some folks able to get up around 4000. I started out with a 3700X which wasn't consistently able to hit the 3600 rating on my kit so I had to run it at 3200 or 3466 for a long time, but when I stepped up to a 5800X3D I just set it to use XMP and had zero issues. Looks like 7000-series may be similar - the supported figure is 5200, but a lot of folks are recommending 5600 or 6000MT/s kits. Eletriarnation fucked around with this message at 14:39 on Jul 14, 2024 |
![]() |
|
VelociBacon posted:Cool, do you still have to adjust vCore and LLC etc? Does better performant RAM (both CL and overall speed) generally allow you just to use more of an offset? You don't need to think about VCore, that is what the curve is for, and the default LLC should be fine. RAM is unrelated to curve optimisation. The long and short of RAM for AMD is that unless you want to manually tune it, you should just target DDR5-6000. You want to make sure you buy Hynix memory, which will be the only memory sold in 6000C30 kits. This will let you run Buildzoids easy hynix timings, which are basically conservative tuned timings that should work with all Hynix memory and Ryzen CPUs. Small correction on BZ's timings, tRC should be equal to tRAS + tRP. BZ put down 68 assuming all XMP kits have 38 tRP, but this isn't strictly the case. You can get better results by going further than BZ timings but it's obviously a lot of work to do manual memory OC. If your BIOS is up to date, DDR5-6000 should run with the memory controller in 1:1 mode (3000MHz for both UCLK and MCLK). Note that you won't be able to run the fabric synced 1:1 to the memory controller in this instance, as 3000MHz is impossible for fabric. There is still a large penalty for running fabric unsynced from the memory controller like with past Ryzen CPUs, but there is a new 2:3 sync mode that brings the latency down. So you want to have your fabric clock at 2000MHz. Buildzoid's easy timings video says that 2033 showed faster results, but this is due to the earlier AGESA versions enforcing the 2:3 ratio, and that change actually increasing the memory to 6200. On newer AGESA's it'll just let you run unsynced, harming performance if you do 2033MHz.
|
![]() |
|
Word on the street is the next X3D chips will be more unlocked for overclocking.
|
![]() |
|
You can actually overclock X3D chips past their stock fmax right now but it's a little cursed, you have to use a motherboard with an external clock generator and do ECLK overclocking. I run my 7950X3D at 5.43 on the cache cores and 5.95 on the frequency cores (103.55MHz ECLK), which needs positive curve offsets. Requires a very expensive motherboard, lots of fiddling, overkill cooling, and increases power draw significantly. But it's fun.
|
![]() |
|
Thanks everyone. I already had ordered the Trident Z5 Neo (F5-6000J3038F16GX2-TZ5NR, it's Hynix) RAM that's EXPO rated to 6000mhz with CL30. Can I get a quick check on the order of operations for 7800x3d overclocking then? If I'm understanding it right you basically set your RAM timings either manually or with EXPO, seems like the EXPO timings are 99% as good as what BZ is recommending for my RAM. Assuming the system boots properly with the EXPO timings I wouldn't waste time doing memtest runs for stability. Next you set the fabric clock to 2000mhz (is this automatically done on 6000mhz RAM with EXPO enabled?) and start adjusting the voltage offset to effectively undervolt the CPU, which will adjust it's boost clocks/voltage automatically based on tjmax. Sounds like actually adjusting the CPU multiplier is a lot of work with little reward, where does a 7800x3d limit it's own clocking? 5ghz regardless of temp? I get the feeling I'll hit the max clock and have temp to spare, which is interesting. Appreciate the explanations thus far. e: for stress testing, is it safe to use p95 small FFTs? I'm from the era where the AVX instructions could potentially be dangerous to CPUs of the time. Thanks VB - Bruere fucked around with this message at 21:04 on Jul 14, 2024 |
![]() |
|
VelociBacon posted:Thanks everyone. I already had ordered the Trident Z5 Neo (F5-6000J3038F16GX2-TZ5NR, it's Hynix) RAM that's EXPO rated to 6000mhz with CL30. Can I get a quick check on the order of operations for 7800x3d overclocking then? EXPO only sets the primary timings and voltage, buildzoids timings change the secondary and tertiary timings. Despite the name, there is actually significantly less performance gains in changing primary timings compared to the other timings (which is why BZ says to straight up not bother changing them from the EXPO timings). So you enable EXPO, then set the rest of buildzoids timings. There can be 10% of CPU limited gaming performance from going from EXPO to BZ's timings in memory bound games, it's worth doing. Ryzen CPUs do technically still support the old style of overclocking, setting the multiplier and voltage, but the X3D CPUs don't. So you don't need to worry about it. The fmax for the 7800X3D is 5.05GHz, but you will be voltage/temperature limited in multi-threaded situations. AMD for some reason set the stock PPT (power limit) to be really high, 162W, so you will literally never be power limited. Enabling PBO and setting it to motherboard limits will raise the current limits as well though, which you can hit in certain extremely stressful loads (like all core AVX-512). So PBO is still worth enabling. It's safe to use anything for stress testing, the worst that will happen is that it will crash. AMD CPUs are very careful with boosting to avoid any damage, the only PBO setting that will meaningfully contribute to electromigration is PBO Scalar. The boost algorithm tracks something called FIT (FailuresInTime) which is basically a model for how a given voltage/current/temperature/duration combination contributes to long term failure, which is how it knows to avoid things like higher voltages at higher temperatures. PBO Scalar modifies the "FITness" of boosting to allow it to degrade the CPU faster. It provides a meaningful increase in performance in certain loads but it's really only for if you aren't planning on using your CPU forever, lol.
|
![]() |
|
BurritoJustice posted:EXPO only sets the primary timings and voltage, buildzoids timings change the secondary and tertiary timings. Despite the name, there is actually significantly less performance gains in changing primary timings compared to the other timings (which is why BZ says to straight up not bother changing them from the EXPO timings). So you enable EXPO, then set the rest of buildzoids timings. There can be 10% of CPU limited gaming performance from going from EXPO to BZ's timings in memory bound games, it's worth doing. Hey thanks for this great post. I'll look up some of this stuff and yeah I'll put in BZ's other timings after setting EXPO, I just have such a low understanding of memory overclocking that I wouldn't know what to adjust if it's unstable. I suppose I can just run only the EXPO stuff if it's unstable. Looking forward to playing with this stuff as always. So fun to see how your chip binned out but I guess the more performant 7800x3ds are 79xx chips.
|
![]() |
|
Got the system going and just finished setting up the ram at EXPO with BR's timings - thanks everyone for the pointers. Right now seems stable. Looking at my ram it seems I ended up with M die (trident neo rgb). Waiting for some other stuff to come in so I can get thermal management finished before I start stress testing or applying underclocks but I'll let you all know how that goes. Appreciate all the help.
|
![]() |
|
Sorry for a triple post but what's the most punishing stress test these days? I ran p95 small FFTs with AVX512, it completed without even increasing my coolant temp more than a few degrees to 31C. I downloaded OCCT and I'm currently running the combined test on 'extreme' (it's pulling 450w) and having a hard time getting serious heat into anything. My coolant temp wasn't getting over 35C with my radiator fans at 75% so I dropped them to 50% and it's not getting past 38C. It's 25C in my room. I'm trying to get a good steady state heat load in so I can start playing with pump speed to see how much of a difference it might make and if there are any obvious inflection points.
|
![]() |
|
My understanding of OCCT is it's a stability test, it doesn't so much push the high end limit as it tries to to trigger spinning up to speed under less than ideal conditions. If you can get the max negative core offset and OCCT passes overnight or a whole day or whatever you've won and can be done The amd chips only seem to be able to squeeze like 2-5% gains (while running cooler) It ain't the era of the i-5 2500k where I juiced it up like 1ghz If a core fails it should tell you and you bump the voltage up some (on that core alone) or if you experience a crash in a game or whatever. I'm all -30 which was the max for 5800x3d but I bought mine within a month or 2 or release and online people claim those were some of the best samples. You can get more benefit out of juicing your ram timings nowadays than the processer.
|
![]() |
|
VelociBacon posted:Sorry for a triple post but what's the most punishing stress test these days? I ran p95 small FFTs with AVX512, it completed without even increasing my coolant temp more than a few degrees to 31C. I downloaded OCCT and I'm currently running the combined test on 'extreme' (it's pulling 450w) and having a hard time getting serious heat into anything. My coolant temp wasn't getting over 35C with my radiator fans at 75% so I dropped them to 50% and it's not getting past 38C. It's 25C in my room. I'm trying to get a good steady state heat load in so I can start playing with pump speed to see how much of a difference it might make and if there are any obvious inflection points. 38C is very high coolant temperature. You really don't want to go over 35 if you can, the heat transfer capabilities drop drastically as the coolant temperature increases. If you're at 40C you're gonna have something like thermal runaway.
|
![]() |
|
furmark is a gpu stress test; occt is a gpu stability test if you're worried about issues that are harder to suss out hanging in path of exile towns for five minutes is its own weird test that can crash most gpu's for some reason. at least switching from vulkan to dx12 made it stop happening for me
|
![]() |
|
Quaint Quail Quilt posted:My understanding of OCCT is it's a stability test, it doesn't so much push the high end limit as it tries to to trigger spinning up to speed under less than ideal conditions. Looking around online I saw a lot of people explicitly state that -30mV wasn't stable for a single person on the 7800x3d (in their communities). I should try it for completeness regardless but currently it's set for 85c target (never gets over 65c), PBO on, -20mV. I know the frequency on these chips isn't where you grab the most performance (as in you're getting the game performance from the cache and the infinity fabric/RAM timings) but I'd like to get the chip to 5ghz all core if only to see the performance difference. I definitely hear you about the realities of overclocking these vs the 2500k or similar. TheFluff posted:38C is very high coolant temperature. You really don't want to go over 35 if you can, the heat transfer capabilities drop drastically as the coolant temperature increases. If you're at 40C you're gonna have something like thermal CV runaway. Respectfully I disagree, allowing the coolant to get up to around 40c max allows you to take advantage of a larger gradient between the ambient temp and the coolant, meaning you get more relative performance in a per-db sense. I've always thought of 40c as being around the ballpark of where you wouldn't want to push it much farther but you are essentially at an inflection point of efficiency after which you experience diminishing returns. I believe most people with custom loops set up their curves for a ~40c max. My component temperatures are still very low at that coolant temp target! ![]() kliras posted:furmark is a gpu stress test; occt is a gpu stability test if you're worried about issues that are harder to suss out I've always found furmark to be not that stressful (out at least not that heat-generating). I didn't even install it this time around but I can certainly give that a shot again. I remember some build of occt actually had further built into it's suite, wonder if that's still the case. No PoE here but playing street fighter 6 at 4k with every setting maxed seems to generate more heat than anything synthetic I've tried so far.
|
![]() |
|
fwiw, poe is f2p, and the first town is 2m from the starting area. creating a user might take a second, though tbh the real gpu stress test is a main menu the devs forgot to cap the framerate on like every capcom game
|
![]() |
|
TheFluff posted:38C is very high coolant temperature. You really don't want to go over 35 if you can, the heat transfer capabilities drop drastically as the coolant temperature increases. If you're at 40C you're gonna have something like thermal runaway. 40C is a totally fine coolant temp. Depending on where you are in the world it might be inevitable just due to higher ambients. The hard limit to be concerned about is the max temp for your coolant and pump, which is 60C for most popular coolants and D5 tops. I target 40C because it lets my fans be much quieter, around 800RPM, and my component temps are still fantastic.
|
![]() |
|
TheFluff posted:38C is very high coolant temperature. You really don't want to go over 35 if you can, the heat transfer capabilities drop drastically as the coolant temperature increases. If you're at 40C you're gonna have something like thermal runaway. I disagree. My curve is basically 30% at 37 and 100% at 43 and it never even gets to 40 most the time. Lowering all of this just makes my fans louder for what amounts to maybe one to 2 to 3°C, which is a really poor trade-off. While we are on the topic of things people disagree about. https://www.youtube.com/watch?v=xNAMxZgvves Despite the entire Internet screaming at you, it’s actually better to use your radiator as intake than exhaust. The GPU doesn’t seem to care about the radiator as intake nearly as much as people think it would, but the CPU absolutely notices the difference. spunkshui fucked around with this message at 19:05 on Jul 31, 2024 |
![]() |
|
spunkshui posted:] Ah I was confused at first until I saw you're talking about an AIO. I'm running a custom loop with two quite large rads, one at the top as an exhaust and another in the front as an intake: ![]() I would prefer having a bit more room for more case intake fans but this is working well, especially now that I'm on the 7800x3d which doesn't seem to generate the same kind of heat as the 9900k. Don't want to derail the thread though, there is a water cooling thread. Going to spend some time in the bios today to see if I can find where to instruct the cpu to stay at ~5ghz instead of settling down at 4.5ghz Uber sustained load. I just want to see if it makes a difference.
|
![]() |
|
VelociBacon posted:Ah I was confused at first until I saw you're talking about an AIO. I'm running a custom loop with two quite large rads, one at the top as an exhaust and another in the front as an intake: Just tossing the knowledge out here if anyone is cooling a CPU with an AIO doing this is helpful. Not only is it great for water temps its also just strictly better since the gpu doesn't suffer as much as you would think.
|
![]() |
|
Hey, not sure if this is the place for it, but: I have the same problem as this guy, where the PROCHOT setting on my... mobo? CPU? got improperly perma-triggered and keeps setting my CPU speed to the minimum 0.5 GHz (my actual temps are in the 30Cs). If I turn PROCHOT off through Ryzen Master the CPU instantly goes back up to 3.7 GHz and all is well, but I have to do that on every reboot and wake from sleep. That guy's solution doesn't work for me, because that program is apparently incompatible with my hardware (B550MH 3.0 Biostar mobo, Ryzen 5 5600X CPU) - trying to do anything with AMDRMCLI.exe gives me "Not Supported Processor!" and fails. Is there some other way to disable PROCHOT on Win10, or to use Ryzen Master from the command line so I can make a .bat to disable it automatically?
|
![]() |
|
Maha posted:Hey, not sure if this is the place for it, but: You can disable PROCHOT in the UEFI persistently, it should be buried somewhere under AMD CBS.
|
![]() |
|
Maha posted:Hey, not sure if this is the place for it, but: When I had this issue, the solution for me and my R7 1700 was to replace the biostar x370 motherboard with an MSI B450 (and of course in the meantime use ryzen master to bypass PROCHOT and never shut down except for windows updates). Biostar sucks and always has, though, so it's not a bad option to consider when other decent motherboards for AM4 are ~$90 or less. I think Biostar has a problem where it's reading temps wrong or something to trigger prochot regularly. I didn't actually replace the motherboard right away, I left it like that until the system got unstable and would crash every couple of days and then replaced the motherboard. It's pretty par for the course for Biostar to make bad stuff, though so it's not unsurprising. You'd think I'd know better than to buy Biostar, but I got the CPU and motherboard as a bundle for under $200 at the time and just needed something to replace my haswell system that had gotten unstable.
|
![]() |
|
Rexxed posted:When I had this issue, the solution for me and my R7 1700 was to replace the biostar x370 motherboard with an MSI B450 (and of course in the meantime use ryzen master to bypass PROCHOT and never shut down except for windows updates). Biostar sucks and always has, though, so it's not a bad option to consider when other decent motherboards for AM4 are ~$90 or less. I think Biostar has a problem where it's reading temps wrong or something to trigger prochot regularly. I didn't actually replace the motherboard right away, I left it like that until the system got unstable and would crash every couple of days and then replaced the motherboard. It's pretty par for the course for Biostar to make bad stuff, though so it's not unsurprising. Yup, I got mine for cheap and learned about their reputation later, oh well. BurritoJustice posted:You can disable PROCHOT in the UEFI persistently, it should be buried somewhere under AMD CBS. I just looked and couldn't find it, do you know a name it could be under other than PROCHOT? (Or a way for me to show you the menus that isn't taking 50 pictures of my screen?)
|
![]() |
|
Maha posted:Yup, I got mine for cheap and learned about their reputation later, oh well. It should be under AMD CBS > SMU Common but it appears that most motherboards hide it. Sorry for getting your hopes up.
|
![]() |
|
The board might have a faulty temperature sensor, if it's not too much of a pain in the rear end I would try RMAing it. Almost everyone that experiences biostar ends up swapping to something else, though. I don't think it's worked since Zen 1 (parts of these tricks basically got incorporated as OC features in future boards) but sometimes setting "CPU Temperature Threshold" (if it isn't called PROCHOT) to 0, 120, -1, or 255 will disable it if the other option is not exposed. You could also try the EFI Bootloader poo poo alongside ThrottleStop, I remember that being popular with a few generations of Intel a while back.
|
![]() |
|
You should be able to use Smokeless UMAF to access the hidden AMD CBS setting and disable it persistently.
|
![]() |
|
![]()
|
# ? Jul 9, 2025 11:12 |
|
my 9800x3d is observationally stable at -30 all core curve AND a -90mV Vcore undervolt as well as 6000c30 on only 1.085V Vsoc for 4 months now not sure if i struck the chip lottery
|
![]() |