|
Post by mllrkllr88 on Jan 3, 2022 2:02:24 GMT -5
Day 2 of my X58 journey comes to a close for me. With PSC, I am about 1 second behind what appears to be top Hyper effi? At least 1035 6-9-6 is fully waza stable never fails...makes these long 32m runs easier to take...unlike fail loop 24 shiz on modern gens I am done with Giga AOC and going to move on to R3E/R2E Attachments:
|
|
|
Post by ground on Jan 3, 2022 2:29:46 GMT -5
Strong Score!
I was aiming for 2:2:1 for easy comparability, I'm sure with higher memory clocks I can get even more out of it; my Hypers are capable of a fair bit more as I showed before. My Waza isn't perfectly tuned, 32M is still a tough nut of a bench for me as well. Tertiaries need a second look as well, I recently found a couple things that could probably improve that old score a fair amount. Might retry after work and see what I can squeeze out of it.
|
|
|
Post by dr4g00n on Jan 3, 2022 13:30:09 GMT -5
Well I figured out that it was the "turbo performance" setting messing it up, disabled it and uncore works as it should.
Can't get my psc to do 2000 though no matter what I try, just fails to post every time. Threw my d-die in and it runs just fine at 2000c8. Might be the auto timings running too tight I guess?
|
|
|
Post by mllrkllr88 on Jan 3, 2022 20:37:57 GMT -5
Well I figured out that it was the "turbo performance" setting messing it up, disabled it and uncore works as it should. Can't get my psc to do 2000 though no matter what I try, just fails to post every time. Threw my d-die in and it runs just fine at 2000c8. Might be the auto timings running too tight I guess? I ran into problems with R2E and R3E and PSC. I can train 2000c6-9-6 tight but both boards refuse to run anything with that memory setup. I set timings manually and used similar volt profiles. Asus is not optimized for PSC is seems...so perhaps EVGA is the same. These aren't even my chronic PSC either, random 3x 2200c7 Pi Giga AOC runs 2100 6-9-6 tight @1.87 32M waza, but gets a bit squirrelly after that.
|
|
|
Post by PolRoger on Jan 3, 2022 21:23:00 GMT -5
I'm trying to improve the efficiency of my 32M scores but I'm finding it to be a tough slog, I've managed to get some scores below the 19min mark at 4GHz but I'm still many seconds behind the two Rabbitts score(s) of ~8m 43/44s (Nice work both of you). 1900C6 ~4GHz Dom GT Elpida Hyper: 2000C7 ~4GHz Dom GT Elpida Hyper:
|
|
|
Post by mllrkllr88 on Jan 3, 2022 21:43:06 GMT -5
I'm trying to improve the efficiency of my 32M scores but I'm finding it to be a tough slog, I've managed to get some scores below the 19min mark at 4GHz but I'm still many seconds behind the two Rabbitts score(s) of ~8m 43/44s (Nice work both of you). 1900C6 ~4GHz Dom GT Elpida Hyper: View Attachment2000C7 ~4GHz Dom GT Elpida Hyper: View AttachmentFirst task is to get setup on XP, it's quite a bit faster for this old gen. Next you'll need to do waza. Barbo has a good video + OS that does it all for you. You can make your own XP like I've done...I can help if you want
|
|
|
Post by PolRoger on Jan 3, 2022 22:04:50 GMT -5
I'm trying to improve the efficiency of my 32M scores but I'm finding it to be a tough slog, I've managed to get some scores below the 19min mark at 4GHz but I'm still many seconds behind the two Rabbitts score(s) of ~8m 43/44s (Nice work both of you). 1900C6 ~4GHz Dom GT Elpida Hyper: <button disabled="" class="c-attachment-insert--linked o-btn--sm">Attachment Deleted</button> 2000C7 ~4GHz Dom GT Elpida Hyper: <button disabled="" class="c-attachment-insert--linked o-btn--sm">Attachment Deleted</button> First task is to get setup on XP, it's quite a bit faster for this old gen. Next you'll need to do waza. Barbo has a good video + OS that does it all for you. You can make your own XP like I've done...I can help if you want I was running with W7 32-bit today. I guess I'll probably need to get a drive going with XP then. I was also looking into copy waza today. Does Vista 32-bit also run faster than W7 32-bit? Will I need to load drivers on XP for x58 parts? Link for the Barbo video? Thanks for offering to help.
|
|
|
Post by ShrimpBrime on Jan 3, 2022 22:53:48 GMT -5
Here's CL6 as promised 1000mhz - Mixed kit 2x Vipers and a Dom. NVM that time though. I gotta work on the time, but CL6 baby! Edit... I have no idea what voltages to use. This is 1.80256v DRAM
|
|
|
Post by mllrkllr88 on Jan 3, 2022 23:15:19 GMT -5
First task is to get setup on XP, it's quite a bit faster for this old gen. Next you'll need to do waza. Barbo has a good video + OS that does it all for you. You can make your own XP like I've done...I can help if you want I was running with W7 32-bit today. I guess I'll probably need to get a drive going with XP then. I was also looking into copy waza today. Does Vista 32-bit also run faster than W7 32-bit? Will I need to load drivers on XP for x58 parts? Link for the Barbo video? Thanks for offering to help. www.google.com/amp/s/barbonenet.wordpress.com/2017/04/19/xp-32bit-for-spi32m-on-z270/amp/OS link is there. There was a video explaining how to use his OS, below edit: barbonenet.wordpress.com/2016/11/19/how-use-my-xp-spi-os/ @ ShrimpBrime Good work!! getting 1000c6 working!
|
|
|
Post by mllrkllr88 on Jan 4, 2022 0:39:03 GMT -5
I have a question for you guys about X58. As I said I am brand new to this platform and I'm seeing something weird. When I raise FSB in OS with SetFSB, it will raise up to about 50 MHz only. Going beyond +50 MHz the multi ratio will drop. The more I riase up BCLK, the more the ratio drops, so the clock stays within about 50MHz of the bootup clock speed. Anyone know the cause?
|
|
|
Post by ground on Jan 4, 2022 1:34:19 GMT -5
I have a question for you guys about X58. As I said I am brand new to this platform and I'm seeing something weird. When I raise FSB in OS with SetFSB, it will raise up to about 50 MHz only. Going beyond +50 MHz the multi ratio will drop. The more I riase up BCLK, the more the ratio drops, so the clock stays within about 50MHz of the bootup clock speed. Anyone know the cause? old cpuz bug; you can use CPUz 1.66 or older which don’t have it. I reported it to cpuz dev but they never got back to me on this issue. Clock is still going up, confirmed with multiple boards. Makes validations on locked chips an x58a-oc only endeavor, since you can simply boot lower multi on that board. RE psc on asus: I had no problem up to ~2300 with psc on asus, but DR really hits a hard wall in the outermost channel. inner two do 2700 DR with my best IMC.
|
|
|
Post by PolRoger on Jan 4, 2022 9:03:47 GMT -5
I have a question for you guys about X58. RE psc on asus: I had no problem up to ~2300 with psc on asus, but DR really hits a hard wall in the outermost channel. inner two do 2700 DR with my best IMC. Just so I'm clear about which DIMM slot you are referring to... I find that sometimes the slot closest to the CPU socket (A1) won't train or can drop out and not be recognized when booting a newly installed memory kit. Is this what you are calling the outermost channel? Or do you mean C1 out by the edge of the motherboard? I always use the A1, B1, C1 slots. If I try to use A2. B2, C2 alone the board doesn't post? Those DIMM slots only seem to work when a stick is already installed in A1, B1, C1.
|
|
|
Post by ground on Jan 4, 2022 9:42:16 GMT -5
C1 drops out at high frequency. A1 can drop out with insufficient pressure on the CPU; this could be caused by ILM slowly loosening up over time or over hundreds of CPUs swaps, I usually bend the ILM arm a little bit to mitigate that, haven't had the inner channels drop on any of my asus boards that had issues in the past. *2 slots can be ignored for benching anyways, the behavior of them only working with dimms installed in *1 is normal.
|
|
|
Post by PolRoger on Jan 4, 2022 12:28:20 GMT -5
C1 drops out at high frequency. A1 can drop out with insufficient pressure on the CPU; this could be caused by ILM slowly loosening up over time or over hundreds of CPUs swaps, I usually bend the ILM arm a little bit to mitigate that, haven't had the inner channels drop on any of my asus boards that had issues in the past. *2 slots can be ignored for benching anyways, the behavior of them only working with dimms installed in *1 is normal. With your experience testing so many 1366 CPU(s) I would think that you now have a method of binning a new untested CPU as far as getting a quick idea of the potential OC margin that a particular chip might have for either ambient or cold overclocks? Something like maybe 50x @ ~1.5v? I don't have very many samples and my current 980X that I'm running with now seems to be topping out at ~49x on ambient with just light single core loads (pifast, SuperPi 32M). If you were binning, what criteria would you use for determining something like a poor/average chip vs. a good/very good one?
|
|
|
Post by ground on Jan 4, 2022 13:11:08 GMT -5
Multi core goals for me are typically Cinebench R15 > 4500 at 1.35V or 1.4V, depending on whats easy to cool. Best ambient 6 core chips (I mostly binned quads) I have are the following: W3680 4640 1.4V E5649 4760 1.4V
Quads I've had as high as 4900/1.4V in a couple E5630s and E5620s, though not as important since those chips are always BCLK limited on cold, since good chips on BCLK are 99% of the time also good on core, and in practice BCLK matters most on those chips.
|
|
|
Post by mllrkllr88 on Jan 4, 2022 15:02:41 GMT -5
old cpuz bug; you can use CPUz 1.66 or older which don’t have it. Thank you, it was driving me nuts. I will stick to newest CPUz for this comp (should be required IMO)
|
|
|
Post by mllrkllr88 on Jan 4, 2022 15:23:00 GMT -5
Multi core goals for me are typically Cinebench R15 > 4500 at 1.35V or 1.4V, depending on whats easy to cool. Best ambient 6 core chips (I mostly binned quads) I have are the following: W3680 4640 1.4V E5649 4760 1.4V Quads I've had as high as 4900/1.4V in a couple E5630s and E5620s, though not as important since those chips are always BCLK limited on cold, since good chips on BCLK are 99% of the time also good on core, and in practice BCLK matters most on those chips. So my 990X is looking strong then it seems? 4970 geek3 and 4800 x265 4k I'm blasting volts with a fat water loop, but normal 75F ish ambient.
|
|
|
Post by Mr.Scott on Jan 4, 2022 15:36:09 GMT -5
So my 990X is looking strong then it seems? 4970 geek3 and 4800 x265 4k I'm blasting volts with a fat water loop, but normal 75F ish ambient. I would say so, yes.
|
|
|
Post by georgekokovinis on Jan 4, 2022 16:49:56 GMT -5
Multi core goals for me are typically Cinebench R15 > 4500 at 1.35V or 1.4V, depending on whats easy to cool. Best ambient 6 core chips (I mostly binned quads) I have are the following: W3680 4640 1.4V E5649 4760 1.4V Quads I've had as high as 4900/1.4V in a couple E5630s and E5620s, though not as important since those chips are always BCLK limited on cold, since good chips on BCLK are 99% of the time also good on core, and in practice BCLK matters most on those chips. So my 990X is looking strong then it seems? 4970 geek3 and 4800 x265 4k I'm blasting volts with a fat water loop, but normal 75F ish ambient. How much voltage are you using Davide for geek3 4970 ? Thank you.
|
|
|
Post by PolRoger on Jan 4, 2022 17:32:31 GMT -5
A cold front came through Atlanta yesterday. I moved my setup outside this afternoon where the ambient temps ~51F. Over the next couple of days, we might see lows at night/early morning in the low to mid 30's. I'm going to try and do some multi-core benchmarks with the cooler temps. EDIT: Are we going with the 32-bit version of Geekbench 3? My old 4-pipe Noctua cooler is having trouble handling the heat output from Geekbench 3 at 1.5v(+)! I'm having to back the clocks down below 4.8GHz. I think I may have to hook up some big water-cooing to try and help tame the multi-core bench heat. I also realized that I need an internet connection so I'm going to have to run a Lan cable from my router to the outside table.
|
|
|
Post by mllrkllr88 on Jan 4, 2022 19:24:00 GMT -5
So my 990X is looking strong then it seems? 4970 geek3 and 4800 x265 4k I'm blasting volts with a fat water loop, but normal 75F ish ambient. How much voltage are you using Davide for geek3 4970 ? Thank you. 1.55v core set from bios. Temps are 65c max, geek is very weak stress on this gen, even DFFT test was light load.
|
|
|
Post by ShrimpBrime on Jan 4, 2022 19:30:00 GMT -5
C1 drops out at high frequency. A1 can drop out with insufficient pressure on the CPU; this could be caused by ILM slowly loosening up over time or over hundreds of CPUs swaps, I usually bend the ILM arm a little bit to mitigate that, haven't had the inner channels drop on any of my asus boards that had issues in the past. *2 slots can be ignored for benching anyways, the behavior of them only working with dimms installed in *1 is normal. With your experience testing so many 1366 CPU(s) I would think that you now have a method of binning a new untested CPU as far as getting a quick idea of the potential OC margin that a particular chip might have for either ambient or cold overclocks? Something like maybe 50x @ ~1.5v? I don't have very many samples and my current 980X that I'm running with now seems to be topping out at ~49x on ambient with just light single core loads (pifast, SuperPi 32M). If you were binning, what criteria would you use for determining something like a poor/average chip vs. a good/very good one? My chip is about the same clock speed. Can bench in the 4.9ghz area. Feels like pain 1.65 v-core. Looks like a bit of v-droop but that v-core is true. It would never post that speed at 1.55v. THIS has seemingly been an issue. Further on triple channel, running Cl6 2000 mhz dropped two memory sticks reporting only 2gb and the way it feels actually in single channel even though cpuz reads triple. Ive pulled a lot of gray hairs out the past few days lol.
|
|
|
Post by PolRoger on Jan 4, 2022 22:56:12 GMT -5
I switched over from an air cooler to a custom water setup and the temps during Geekbench 3 are now much better. The cool outside weather is also helping. Ambient air temps are now ~ 46F. Max temp during the bench on the hottest core was 69C. Time to get a LAN cable hooked up.
|
|
|
Post by mllrkllr88 on Jan 4, 2022 23:33:07 GMT -5
I cant come up with a suitable prize for the comp, so I'll just donate some PayPal. I will send $50 to ShrimpBrime this week and he will have full discretion about how the money is distributed for the various categories. It's not much, but it's better than a high-five I will be prize-exempt.
|
|
|
Post by ground on Jan 5, 2022 1:38:42 GMT -5
couple more things: -be more careful with volts guys I have done some testing on degradation on this platform in the past, and while that air testing might not translate 1:1 to water with window open, its still relevant I’m pretty sure. 1.55V, 80°C loaded, tweaked for daily lasts around 3 days before starting to lose stability. 1.7V is much harsher; here it only took me a couple minutes to notice clocks starting to drop. Keep in mind this was on aircooling, but it should still show where to be careful If there is interest, I’d love to throw in one of these E5630s I’m running here as a raffle price among those participating in at least 4 categories (as in - for example all 4 ambient single socket categories). Not sure which one, but they all do >5GHz 32M air + >=260 bclk air
|
|
|
Post by georgekokovinis on Jan 5, 2022 2:02:55 GMT -5
couple more things: -be more careful with volts guys I have done some testing on degradation on this platform in the past, and while that air testing might not translate 1:1 to water with window open, its still relevant I’m pretty sure. 1.55V, 80°C loaded, tweaked for daily lasts around 3 days before starting to lose stability. 1.7V is much harsher; here it only took me a couple minutes to notice clocks starting to drop. Keep in mind this was on aircooling, but it should still show where to be careful If there is interest, I’d love to throw in one of these E5630s I’m running here as a raffle price among those participating in at least 4 categories (as in - for example all 4 ambient single socket categories). Not sure which one, but they all do >5GHz 32M air + >=260 bclk air So true. I have a 990X which does cbr20 4.85ghz with 1.45v in bios. 5.5ghz on dice with 1.6v. One day I decided ( face palm ) to push under good water cooling. Got it up to 4.925 with 1.525v and temps around 65c. Stable. But not after 2 days. I immediately tuned it down. Now it can't do 4.9ghz with 1.55v. Oh well, I have a better one. Such stupidity after 22 years of experience
|
|
|
Post by ground on Jan 5, 2022 3:01:58 GMT -5
Multi core goals for me are typically Cinebench R15 > 4500 at 1.35V or 1.4V, depending on whats easy to cool. Best ambient 6 core chips (I mostly binned quads) I have are the following: W3680 4640 1.4V E5649 4760 1.4V Quads I've had as high as 4900/1.4V in a couple E5630s and E5620s, though not as important since those chips are always BCLK limited on cold, since good chips on BCLK are 99% of the time also good on core, and in practice BCLK matters most on those chips. One addition to this - scaling and 32M tolerance can vary quite a bit. My W3680 for example has 200MHz between pifast and 32M at same voltage (5GHz 32M 1.48V, 5.2GHz pifast 1.48V), while on other chips the gap can be much smaller. Scaling wise, typical chips are around 0.04V/100MHz; for 32M my example W3680 is a bit off from that, needing 1.35V for 4500 and thus ending up in the <0.03V range. There are also some chips which seem to absolutely hate Superpi 32M, though this seems to be much more the case the colder you get - my best E5640 runs pifast on dice at 5900 with ease, meanwhile 32M won't even attempt passing at 5200, no matter the voltage. One of the e5630s I currently have set up also seems to have an issue with waza, needing 0.05V more vcore to pass 32M with Waza, thus limiting clocks quite a bit.
|
|
|
Post by PolRoger on Jan 5, 2022 10:49:10 GMT -5
Thanks for voltage degradation tips/experience guys.
With the 980X that I'm testing now it seems to be scaling like this... CB R15 @4630mhz CPU 4000 Mhz Uncore with ~1.425v and CB R15 @4825mhz 4000MHz Uncore with ~1.5375/1.55v. It looks like ~4850Mhz Geekbench 3 is the ~ Max? for this chip. I can't seem to get a successful bench completion bumping up the speed. 4875/4900Mhz won't complete/pass even with ~1.575v.
|
|
|
Post by Mr.Scott on Jan 5, 2022 19:40:21 GMT -5
Very generous donations guys. Thank you.
|
|
|
Post by PolRoger on Jan 5, 2022 20:17:42 GMT -5
LOL... That air cooler setup TAGG is running is kind of nuts! Edit: Help a "Noob" to understand something? Why does TAGG's memory CPUZ screen show only 620MB when he is running 3x2GB sticks?
|
|