Follow along with the video below to see how to install our site as a web app on your home screen.
Note: This feature may not be available in some browsers.
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Oh I've by no means 'properly' stress tested like I would have done many moons ago. But I don't need to due to my usage, gaming, browsing and logging onto a VM for work is all this machine does.
Single core max is 4,191mhz (1mhz added to BCLK)
Multi core is 4,090mhz
So I'm happy that it easily runs at those speeds indefinitely. Boost isn't going to go any higher regardless of what I do (excluding more on the BCLK)
Gone are the days where I could/want to spend time tweaking every little setting, I enjoyed it at the time but nowadays my usage just doesn't justify it. Not sure if that's a good or a bad thing!
Not overthinking it. I was just sharing some things I knew from haunting forums and my own tinkering experience.
I haven't done negative CO on 5700X3D yet. I have on a few 5900X and 5800X3D.
They gained performance in synthetic benches, I didn't have time to check in games benchmarks. If I'm getting more performance in a synthetic benchmark, CPU must be boosting/sustaining higher or else where does the performance come from? which is overclocking.
The boost algorithm is opportunist. If it's see headroom, may that be temperature, voltage, power usage, etc, it's going to boost/sustain higher clocks. By tweaking CO we are creating headroom.
From my own considerable testing in games and synthetic benchmarks, in both Windows and Linux, I saw no appreciable difference in performance when testing -30 on a 5700x3d. Only a difference in power-draw and temperature. The clock speed is hard-locked unless you want to go messing with BCLK, and they happily run at lower than stock power.
They can’t be compared to unlocked 5900x’s which of course do respond favourably to a negative offset; or indeed even 5800x3d’s and their higher clock speed.
I'm using MSI's Kombo Strike options in the bios. The highest level '3' was mostly stable but failed on runs of Cinebench 2024.
Level 2 works fine on all workloads.
I'm only guessing that level 3 = -30 and 2 = -20, but seems likely.
I'm using MSI's Kombo Strike options in the bios. The highest level '3' was mostly stable but failed on runs of Cinebench 2024.
Level 2 works fine on all workloads.
I'm only guessing that level 3 = -30 and 2 = -20, but seems likely.
Correct, Kombo Strike is just presets for the curve optimiser. If you turn Kombo Strike off and set a -20 Curve Optimiser, it should be doing about the same thing.
Not sure which one takes precedence if both Kombo Strike and a manual Curve Optimiser number is set.
From my own considerable testing in games and synthetic benchmarks, in both Windows and Linux, I saw no appreciable difference in performance when testing -30 on a 5700x3d. Only a difference in power-draw and temperature. The clock speed is hard-locked unless you want to go messing with BCLK, and they happily run at lower than stock power.
They can’t be compared to unlocked 5900x’s which of course do respond favourably to a negative offset; or indeed even 5800x3d’s and their higher clock speed.
Stock PPT EDC TDC is the same for 5900X, 5800X3D and 5700X3D, but the 5900X has more cores plus higher frequency, 5800X3D has higher frequency.
I loaded each core with Statuscore, used HWINFO to monitor the specifics sensors I wanted to see for VID, polling interval 250ms. The 5700X3D didn't differ from other CPUs mentioned, it exhibited differing VID per core, but the variance was low, ~17mV. Where as I have seen upto ~40mV on 5800X3D and upto ~50mV on 5900X. Boost frequency per core was identical, where as I've had some 5800X3D where the variance is upto about ~75Mhz between best and worst, but on 5900X I've seen variance of even upto ~270Mhz. Here is album of 5700X3D Statuscore VID per core testing.
Next was some CO tweaking.
The highest VID was on core 3, 6, 7, setting them each to -3 brought them close to the lowest VID on a core. Next I did cores 2, 4, 5, setting -2. Next I went -1 -1 -3 -4 -3 -3 -4 -4, lowest to highest VID request per core ~6mV variance. As I further tweaked CO, it just became core 0 needed to be a notch higher than core 1. Cores 2 to 7 all needed same value, which was 1 notch more than core 1, and 2 notches more than core 0. The 5700X3D wasn't doing as a 5900X or 5800X3D would do, for example a poorer boosting core gain MHz. The tweaking of CO values wasn't going the way it did on 5900X/5800X3D, it suddenly became just start rolling with a global setting. Which made sense, as core to core boost/VID variance was less than 5900X/5800X3D.
Then I benched -4 -5 -7 -7 -7 -7 -7 -7 in CB23. Gains in CB23 multi core were the lowest I'd seen vs 5900X and 5800X3D. I then just jumped to -30 all cores, performance loss wasn't there, all I noted was somewhat the milder CO tweak had slightly higher results consistently than stock or global -30 CO.
Next got wall plug power meter readings for total rig, idle is ~105W inc screen, screen is ~30W.
Dark base 900, with mesh mod to front, top panel behind mesh removed plastic forming to improve exhausting of air.
Cooling is 2x Magicool G2 Slim 360mm in series, EK D5 G2 PWM pump, Bykski CPU block, distilled water with ~20% Mayhems XT-1 clear concentrate, loop order pump>GPU>CPU>rad>rad. Arctic cooling F12 fans on rads, front rad fans push/pull, top rad fans push only. Case rear exhaust fan Silent Wings 3@1000RPM, RAM area cooled by another Silent Wings 3@1000RPM.
Crosshair VIII Dark Hero, Crucial Ballistix Sport LT 2x16GB 3200MT/s C16, XFX Merc 7900XTX (Alphacool block), 2x WD SN770 2TB, 2x 870 EVO 4TB, EVGA G6 850W, Dell S2721DGFA.
Stock CB23 13745 points ~200W, mild CO tweak 13906 points ~196W, -30 all cores CO tweak 13860 points ~180W. Here my jaw did drop .
Then I ran Y-Cruncher default stess test mode using CO tweak -30 all cores and compared it to an earlier stock setup run. CPU package power was lower between each plus temperature improvement. Ran CoreCycler overnight at global -30 CO, which passed 12 iterations and I stopped it.
To round up testing I also checked Statuscore all cores loading at stock, global -30, global -20, global -10.
Global -10 CO tweak I saw very little drop in power at wall plug, -20 ~13-15W lower than stock, -30 ~21-24W lower than stock.
Stock PPT EDC TDC is the same for 5900X, 5800X3D and 5700X3D, but the 5900X has more cores plus higher frequency, 5800X3D has higher frequency.
I loaded each core with Statuscore, used HWINFO to monitor the specifics sensors I wanted to see for VID, polling interval 250ms. The 5700X3D didn't differ from other CPUs mentioned, it exhibited differing VID per core, but the variance was low, ~17mV. Where as I have seen upto ~40mV on 5800X3D and upto ~50mV on 5900X. Boost frequency per core was identical, where as I've had some 5800X3D where the variance is upto about ~75Mhz between best and worst, but on 5900X I've seen variance of even upto ~270Mhz. Here is album of 5700X3D Statuscore VID per core testing.
Next was some CO tweaking.
The highest VID was on core 3, 6, 7, setting them each to -3 brought them close to the lowest VID on a core. Next I did cores 2, 4, 5, setting -2. Next I went -1 -1 -3 -4 -3 -3 -4 -4, lowest to highest VID request per core ~6mV variance. As I further tweaked CO, it just became core 0 needed to be a notch higher than core 1. Cores 2 to 7 all needed same value, which was 1 notch more than core 1, and 2 notches more than core 0. The 5700X3D wasn't doing as a 5900X or 5800X3D would do, for example a poorer boosting core gain MHz. The tweaking of CO values wasn't going the way it did on 5900X/5800X3D, it suddenly became just start rolling with a global setting. Which made sense, as core to core boost/VID variance was less than 5900X/5800X3D.
Then I benched -4 -5 -7 -7 -7 -7 -7 -7 in CB23. Gains in CB23 multi core were the lowest I'd seen vs 5900X and 5800X3D. I then just jumped to -30 all cores, performance loss wasn't there, all I noted was somewhat the milder CO tweak had slightly higher results consistently than stock or global -30 CO.
Next got wall plug power meter readings for total rig, idle is ~105W inc screen, screen is ~30W.
Dark base 900, with mesh mod to front, top panel behind mesh removed plastic forming to improve exhausting of air.
Cooling is 2x Magicool G2 Slim 360mm in series, EK D5 G2 PWM pump, Bykski CPU block, distilled water with ~20% Mayhems XT-1 clear concentrate, loop order pump>GPU>CPU>rad>rad. Arctic cooling F12 fans on rads, front rad fans push/pull, top rad fans push only. Case rear exhaust fan Silent Wings 3@1000RPM, RAM area cooled by another Silent Wings 3@1000RPM.
Crosshair VIII Dark Hero, Crucial Ballistix Sport LT 2x16GB 3200MT/s C16, XFX Merc 7900XTX (Alphacool block), 2x WD SN770 2TB, 2x 870 EVO 4TB, EVGA G6 850W, Dell S2721DGFA.
Stock CB23 13745 points ~200W, mild CO tweak 13906 points ~196W, -30 all cores CO tweak 13860 points ~180W. Here my jaw did drop .
Then I ran Y-Cruncher default stess test mode using CO tweak -30 all cores and compared it to an earlier stock setup run. CPU package power was lower between each plus temperature improvement. Ran CoreCycler overnight at global -30 CO, which passed 12 iterations and I stopped it.
To round up testing I also checked Statuscore all cores loading at stock, global -30, global -20, global -10.
Global -10 CO tweak I saw very little drop in power at wall plug, -20 ~13-15W lower than stock, -30 ~21-24W lower than stock.
So in summary, -30 CO seems to work well on all cores?
So far Ive had that and have not noticed any issues with gaming, browsing, idling and some stress testing (but not to the degree you have gone to have to admit!)
So in summary, -30 CO seems to work well on all cores?
So far Ive had that and have not noticed any issues with gaming, browsing, idling and some stress testing (but not to the degree you have gone to have to admit!)
If CPU has no stability issues in use cases, global -30 CO is going to give the most power/temperature drops without a performance penalty.
Some more testing is going on . Plan to also try 4 dimms and see if IMC can run 3800MT/s.
*** edit ***
Looking at the total rig power figures I posted 20W may seem like ~10% lower power usage between stock and global -30 CO. When you look at it % based on just CPU, which we can take from HWINFO CPU package power it is even larger %.
*** edit 2 ***
Be great if another 5700X3D owner can try the Statuscore per core VID test as I was doing. We could see how variable these CPUs are.
*** edit 3 ***
My rig had not been off since 29th Nov, today whilst I looked over all the data in post #192 and constructed post on another PC. The 5700X3D was given a break and powered down, the global -30 CO and 3800MT/s setup had passed ~20min Y-Cruncher, ~9hrs CoreCycler. So later I tried Realbench stress mode 1hr, which passed. Then tried Kahru RAM Test, this failed. The RAM profile prior to CO tweak had passed mixed testing of ~3 days continuous uptime. I can only assume the global -30 CO tweak is is causing instability.
In my experience most mobo companies will provide a compatibility list with appropriate bios that marries up with your board. So there isn't any need for guesswork.
In my experience most mobo companies will provide a compatibility list with appropriate bios that marries up with your board. So there isn't any need for guesswork.
Probably, only reason I prefer to have a direct check, is simply some boards can have a board specific issue that company can fix specifically in bios update.
I would check the compatibility list for your board specifically, just because you can.
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.
.
.