|
Post by ground on Oct 19, 2018 9:25:37 GMT -5
Found something interesting in the depth of the internet a while back but only now got around testing. No interesting scores today, but this hard-disables the second QPI on a chip and gives ~3-5 MHz more BCLK headroom on dual QPI chips as far as I could tell. No difference in scores, heat output may be a little lower.
|
|
|
Post by ground on Oct 29, 2018 6:55:33 GMT -5
Quick update doing more air experiments, looks like the PCIe slot used can make a small difference in maximum bootable PCIe clock, just got my Rampage II to boot 119 by using the central slot, but its anything but reliable, maybe 1 in 5 boots succeeds.
|
|
|
Post by Vinster on Oct 29, 2018 11:46:39 GMT -5
ouch, that's an odd one. I always thought the top slot had a priority and that would help in these situations.
Vin
|
|
|
Post by ground on Oct 29, 2018 13:14:52 GMT -5
ouch, that's an odd one. I always thought the top slot had a priority and that would help in these situations. Vin Its not reliable enough to be worth anything, but the difference might be bigger on other boards. Just something to keep in mind.
|
|
|
Post by ground on Dec 18, 2018 2:53:33 GMT -5
|
|
|
Post by ground on Jan 10, 2019 22:11:42 GMT -5
researching another topic (1156, posting a thread with a link collection later) I stumbled over something that may be interesting. Gonna try throwing a BIOS for each of my boards in the editor and see if I can get anything out of it. www.xtremesystems.org/forums/showthread.php?252611-GIGABYTE-SUPERPOWER-H55M-UD2H-300-mhz-blck&p=4405718&viewfull=1#post4405718More to follow on this, but I've been doing some degradation (and death) testing on some of my reject C-Batch CPUs, particularly the 3104C403 (same batch as my good E5640). Gonna do some more testing with a mid 250 BCLK E5606 on Sunday. Degradation testing: - at 1.55V Vcore, 1.4V Uncore, 4.75 GHz Core, 3.9 GHz Uncore, temperatures in the high 80s under Intel Burn test, drops in stability were noticeable after about 3 days of normal use. - at 1.5V Vcore, 1.5V Uncore, 4.6 GHz Core, 4.15 GHz Uncore, temperatures in the high 80s under Intel Burn test, no drops in stability were noticeable after about 2 weeks of normal use. Onto the murderous part of the venture after deeming 1.5V VTT safe-ish for benching this batch: - repeated runs of 32m until a) chip gets unstable or b) dies. Clocks raised after every run. No slow mode used. No scores as they are not relevant to this: - at 1.64V Vcore, 1.4V Uncore, 5.1 GHz Core, 3.9 GHz Uncore, nothing noticeable, stability wasn't quite there all the way. Temps mid 30s load - at 1.7V Vcore, 1.5V Uncore, 5.1 GHz Core, 4.1 GHz Uncore, core starting to degrade, requiring more voltage after every run. Temps still mid 30s load - at 1.8V Vcore, 1.6V Uncore, 4.3 GHz Uncore really surprised the chip isn't dead yet, having read many claims of chips commonly dying above 1.4V VTT. Core still degrading quickly, thermals still under control. - at 1.9V Vcore, 1.7V Uncore - system crashes seconds into a run, core seems to be degraded badly. After rebooting it refuses to boot at all above 1.6V VTT. Uncore still seems to clock as before, but can't reach the old 255 BCLK any longer, no matter the voltage. Really surprised the chip lived at 1.9V on air. - at 2.0V Vcore, 1.6V Uncore - Fans spin up, chip dead. I'm sorry, it was for science. Gonna throw it in some dice over the weekend and see if it posts again afterwards though
|
|
|
Post by tapakah on Jan 18, 2019 18:02:41 GMT -5
|
|
|
Post by ground on Jan 18, 2019 18:21:54 GMT -5
Damn, thats some bad luck Though my best E5620 only does 256 BCLK bootable, but the core is the best I have seen myself. Its an 3201B759. I think it does 4.7 GHz 32m at 1.3V if my binning data is correct... Too bad its basically all the same batch, but I'll know to stay away from that one.
|
|
|
Post by ground on Feb 25, 2019 22:13:24 GMT -5
I was going through the leftovers of my 45nm testing, and decided to test one of my somewhat alright rejects for core clocks. I was pleasantly surprised by something I really didn't expect. A 45nm chip that clocks better then any 32nm chip I've ever had in my hands, and better then any 45nm chip I've ever heard off: 4.5 GHz 32m at 1.2V 4.83GHz/1.33V 4.0 GHz uncore at 1.27V with triple channel Guess this CPU will be BCLK limited on Dice
|
|
|
Post by ground on Feb 26, 2019 12:10:38 GMT -5
Binning some late W3530s: Appears not really worth to even bother. The 08-09 W3520s did far better on average.
|
|
|
Post by ground on Mar 3, 2019 12:52:42 GMT -5
Some cold testing: 32nm: E5620: 256->266 Dice E5606: 273.5->286+ Dice (likely board limited) E5649: 269.65->283+ Dice (didn’t test further yet) E5640: 272->287 Dice
45nm: W3520: 232->267 Dice W3530: 230->262 Dice i7 920: 230->263 Dice i7 920: 218->245 Dice
Of all these I’ll likely rerun the W3520 and all those that managed 280+ on Dice on LN2 at some point. All others didn’t perform as hoped and will not be tested much further, at least on cold.
|
|
|
Post by dr4g00n on Mar 4, 2019 10:05:18 GMT -5
In your testing did the Rev./stepping of X58 chips make any real difference in their bclk oc ability's?
|
|
|
Post by ground on Mar 4, 2019 12:13:31 GMT -5
In your testing did the Rev./stepping of X58 chips make any real difference in their bclk oc ability's? Nope, no difference basically. My Rampage II is a Rev 12 chipset and matches the one of a friend in max BCLK given a specific CPU. Its mostly CPU in this case. On a lot of boards I've tested / got tested by others the difference board to board is pretty minimal anyways (0-3 MHz given the same CPU unless the board is limiting in another way)
|
|
jcoc
New Member
Posts: 4
|
Post by jcoc on Mar 6, 2019 13:39:08 GMT -5
Hello Ground!! I wanted to stop by and say thank you for this awesome thread. It has a lot of X58 goodies in it and it helped me learn a lot about my i7-920. I did find a few things out that I would like to share. I tested two GPUs on my Rampage III Formula, here is the PCIe clock I was able to get out of them. R7 240 (DDR3) Once I set 127PCIe in OS, I could boot 127 PCIe everytime after that. Though it would not just cold boot 127 PCIe. 9500GT was good for cold booting 118 but only made it to 123 PCIe before the display gave out. The voltages used for testing PCIe were verbatim to the voltages you posted. During my LN2 run with this chip, there were a couple things I noticed on my board. I will post a hwbot sub for reference. hwbot.org/submission/4093490_I was not able to complete that run with out the following voltage changes -nothing past 1.6375V vcore would scale at -70C or at -80C in single channel memory config -CPU PLL from 1.812V to 1.9250V -QPI-Dram from 1.375v to 1.4250V -I had to enable to Load-Line Calibration -I also couldnt use any CPU differential amplitude for some reason. If i set anything at all, it wouldnt make it into the OS. -At fist I tried 127 PCIe but that didnt bring any stability. It wouldnt even pass 240 BLCK so I dropped it back down to 118 PCIe and 237 blck. And was able to get GB3 to pass there. I this was the only chip I have tested so I could have missed a something. Please, anyone, let me know if there is anything you noticed that I did wrong. X58 is so fun and I will definitely be doing some more ambient and LN2 testing.
|
|
|
Post by Mr.Scott on Mar 6, 2019 18:57:44 GMT -5
You need to pop in more often Jess.
|
|
jcoc
New Member
Posts: 4
|
Post by jcoc on Mar 6, 2019 19:08:30 GMT -5
You need to pop in more often Jess. Indeed I do broham. I just need to do some more looking around on this forum. I know there is a lot of good stuff in here!
|
|
|
Post by ground on Mar 6, 2019 21:00:38 GMT -5
Binning some late W3530s: Appears not really worth to even bother. The 08-09 W3520s did far better on average. The one where I didn't add the batch # was a 3020A571. I've since tested it on cold and noticed it (as well as both of my 920s) didn't scale beyond 1.5V. I think if I continue binning 45nm stuff I'll make max voltage it scales to a relevant criteria. Looking through subs on hwbot chips scaling to 1.8V exist, but most subbed ones end up scaling to around 1.6V. I've previously had a 920 C1 that scaled to 1.9V, but since it needed those for 4.64 GHz that doesn't really help here... Hello Ground!! I wanted to stop by and say thank you for this awesome thread. It has a lot of X58 goodies in it and it helped me learn a lot about my i7-920. I did find a few things out that I would like to share. I tested two GPUs on my Rampage III Formula, here is the PCIe clock I was able to get out of them. R7 240 (DDR3) Once I set 127PCIe in OS, I could boot 127 PCIe everytime after that. Though it would not just cold boot 127 PCIe. 9500GT was good for cold booting 118 but only made it to 123 PCIe before the display gave out. The voltages used for testing PCIe were verbatim to the voltages you posted. During my LN2 run with this chip, there were a couple things I noticed on my board. I will post a hwbot sub for reference. hwbot.org/submission/4093490_I was not able to complete that run with out the following voltage changes -nothing past 1.6375V vcore would scale at -70C or at -80C in single channel memory config -CPU PLL from 1.812V to 1.9250V -QPI-Dram from 1.375v to 1.4250V -I had to enable to Load-Line Calibration -I also couldnt use any CPU differential amplitude for some reason. If i set anything at all, it wouldnt make it into the OS. -At fist I tried 127 PCIe but that didnt bring any stability. It wouldnt even pass 240 BLCK so I dropped it back down to 118 PCIe and 237 blck. And was able to get GB3 to pass there. I this was the only chip I have tested so I could have missed a something. Please, anyone, let me know if there is anything you noticed that I did wrong. X58 is so fun and I will definitely be doing some more ambient and LN2 testing. Hey jesse! Regarding the R7 240 and the 9500 GT - did you test max boot or max in setfsb? Both Rampage III Formula I had tested by others so far managed 136+ in setfsb if I remember correctly. If you have AM3 on hand - can you check if either of those can boot 150+ PCIe (or at least a significantly different value)? Thats how I tested it for my 8600GT (which also does 138 consistently on gigabyte boards, yet my Rampage II appears to max out bootable around 118-119 on ambient and 120 with the CPU on cold). I've recently started playing around with higher voltages on the IOH and ICH and found some stuff. - ICH voltage (as well as ICH 1.5/ICH PCIe) appears to help with PCIe a tiny bit if set to obscene levels (1.55V+), though scaling is fairly minimal. - IOH voltage helps immensely with max BCLK without turning on slow mode (with slow mode it doesn't matter at all, neither of these is a big surprise, just something I hadn't bother testing yet). I've managed 240-ish without slow mode on ambient at safe voltages on my baseline E5606 and I'll likely try more once I have a board I consider low enough value to risk. However, going too far without slowmode can bug out the BCLK (and all clocks) completely, producing results such as this (more is possible, I saw as high as 650 MHz though it wouldn't surprise me if 700+ was doable given spamming F8 for long enough). I haven't checked (I don't have an oscilloscope sadly), but I wouldn't be surprised if it was actually all over the place since the system slowed to a crawl (significantly worse then slowmode) and Setfsb crashed upon opening after causing this bug. ALL clock measuring tools I tested freak out with this (CPUz, hwinfo64 for sure, need to test more). Lsdmeasap has observed this a long time ago too. I need to test my theory further, I may have to bin all my W3520s and W3530s a third time to confirm my theory, but binning 45nm CPUs for this platform may be a little more complicated then I initially thought. For 45nm binning for 4.5/1.4V or better for 32m first, then max BCLK (if relevant), max scaling voltage as well as uncore may be quite relevant - out of the 4 chips I had considered worth testing further only 1 scaled beyond 1.5V. PLL - I've yet to see any scaling from raising this. Sadly it can't be lowered on the Rampage II, so even on Dice I kept it on the lowest value the board allows. QPI - (mostly) irrelevant for BCLK, but quite relevant for RAM and Uncore. My good W3520 scales to at least to 5400 MHz uncore (short benches only) at 1.6V QPI/VTT from my dice testing, though I didn't test more voltage yet. Gonna save that for LN2 tests.CPU differential amplitude - another thing I need to test further, but maxing it usually helped a tiny bit in my testing while having the chip run a bit hotter. I've heard optimising PCIe clock for the actual BCLK may be helpful (running it at the (targetBCLK-221)*0.445+100=PCIe clock) in some rare cases where it helps stabilising the BCLK. Another thing that needs further testing. Nothing I noticed you missing in particular, looking forward what else you can do with that setup Hope you have fun on this platform, its currently my all-time favourite! been doing some forum-diving recently (gone through ~400 xtremesystems threads as well as all the relevant madshrimps threads) and the most interesting thing I found was this one, some interesting findings. If anyone else has anything - links, their own testing etc - please contribute it. Having as many of these in one place as possible would be great. A link collection of various things can be found in the first post of this thread. If someone has an unmodded EX58-Extreme/UD4/DS4 etc (the gigabyte boards that max out around 101 PCIe without hardmods) Coldwove and I may have reverse engineered the modification they need to work at 120+. Since we aren't sure it works and doesn't brick the board I'm not publishing it, but upon request via PM I'll forward the information we found. Sadly its a different issue causing the max bootable BCLK on the Rampage II Extreme which I've yet to figure out. I'm getting another board - MSI x58 Pro-E - pretty soon. This board is particularly interesting because I've seen it tested all the way up to 180 MHz PCIe, though I'm quite unsure how that has been achieved. I'll try getting it to work and see if I can apply it to other boards.
|
|
jcoc
New Member
Posts: 4
|
Post by jcoc on Mar 6, 2019 22:22:49 GMT -5
Regarding the R7 240 and the 9500 GT - did you test max boot or max in setfsb? Both Rampage III Formula I had tested by others so far managed 136+ in setfsb if I remember correctly. If you have AM3 on hand - can you check if either of those can boot 150+ PCIe (or at least a significantly different value)? Thats how I tested it for my 8600GT (which also does 138 consistently on gigabyte boards, yet my Rampage II appears to max out bootable around 118-119 on ambient and 120 with the CPU on cold). I was using TurboV to set PCIe and BLCK in OS for both of the GPUs. I am going to work with a i7-950 tonight and will use setFSB. I completely forgot that setFSB worked on X58 to be honest. Lol I also want to worth the BLCK/PCIe formula you posted too. (targetBCLK-221)*0.445+100=PCIe clock). As I was running 118 PCIe when trying to work with 240 instead of 108/109 (108.445) PCIe like the formula calls for. If do get better results I will you guys know.
|
|
|
Post by ground on Mar 31, 2019 19:13:48 GMT -5
|
|
|
Post by ground on Apr 27, 2019 13:45:25 GMT -5
Doing some more testing today to confirm a theory I got in the last session. Apparently an uncore multi of 14+ helps a lot stabilising high BCLK, which also helped in the run linked above. Currently testing at 289-290 BCLK on my E5640 on Dry ice. It has a lower coldbug then said E5606...
|
|
|
Post by ground on Jul 1, 2019 18:22:04 GMT -5
tested a new lot of xeons today, pretty terrible results: Xeon E5620 3038B033 237 Xeon E5620 3038B033 232 Xeon E5620 3038B033 238 Xeon E5620 3038B033 229 Xeon E5640 3038B179 243 Xeon E5640 3038B179 236 Xeon E5640 3040C001 238 Xeon E5620 3041B410 238 Xeon E5640 3043B673 243 Xeon E5640 3043B673 240 Xeon E5620 3107B255 230 Xeon E5620 3107B255 223 Xeon E5645 3128B445 236 Xeon E5645 3131B156 233 Xeon E5649 3132B256 228 Xeon E5649 3132B256 230 Xeon E5645 3138B541 226 Xeon E5645 3140B548 236 Xeon E5649 3228B504 doa? Xeon E5649 3228B504 245
|
|
|
Post by Vinster on Jul 1, 2019 18:38:24 GMT -5
^ All of those are still better than my bests... lol
I think I need to start blaming myself or the board...
Vin
|
|
|
Post by ground on Jul 11, 2019 7:15:17 GMT -5
So, took my new x58a-oc cold last weekend with T.Rex and it worked fairly well. Given that this was at -93°C Cascade and this chip has a cb of -160, and the R2E does 1-2MHz more then the x58a-oc I think there should be more in this chip. We also took his R3E cold, but it didn't work that well, above 280 it starts being really hard to handle, those last 4 MHz each took longer then the 291 on the x58a-oc total (incl mount + insulation). Also broke 6:30 32M on the 21x multi last weekend which is a milestone for me - next fastest at that Multi is still 4 seconds slower. Effortless benching at 5.5+ GHz with insanely low voltages - 5.53 needs 1.46V for 32m, 5.25GHz uncore need 1.57V VTT. basically BCLK limited since the cores don't seem to be the limiting factor. With LN2 more should be possible. Might try pushing the chip more next time I get around to taking it cold, no doubt that 5.55 32m is doable, but I might have to drop uncore down a multiplier.
|
|
|
Post by ground on Aug 4, 2019 12:26:40 GMT -5
Picked up a new i7 920... air testing. Cores are sitting around ~65°C running 32M at 1.21V...
|
|
|
Post by Vinster on Aug 7, 2019 20:44:21 GMT -5
wow, that's amazing.
Vin
|
|
|
Post by ground on Oct 1, 2019 15:20:04 GMT -5
|
|
|
Post by Vinster on Oct 1, 2019 15:58:40 GMT -5
oh that hurts....
Vin
|
|
|
Post by osmiumoc on Nov 30, 2019 22:10:34 GMT -5
You have done some intense research on this platform, may I ask a question or two about how X58 and BCLK behaves? I have an X5650 on a Rampage III Extreme board. It stops CPU initialisation when I go above 226 BCLK, is that more likely to be a CPU limit or do I hit a board limit? Is there a way to find out if a board is the limit or not? The next thing is, my X5650 will not run SuperPI at anything above 4,9GHz @ 213 x 23. It can boot and validate fine valid.x86.fr/7i74mt up to 5,2GHz tho. I tested the RAM and it is stable at that frequency, I can clock RAM and Uncore higher than the 5GHZ run while setting a lower multi and SuperPI works again, but once the CPU frequency is above 4,9 it BSODs instantly. It is a hard CPU frequency wall that only shows up in SuperPI and other PI benchmarks. Cinebench runs well. Is there a specific reason for this? It appears to me as if some system-bus gets overloaded by the PI-workload and collapses. I tried QPI slowmode with no success. Anything I can do to try and stabilize SuperPI? Its a shame that I can´t utilize the 300MHz more that the CPU appears to have in it.
|
|
|
Post by ground on Dec 1, 2019 3:18:24 GMT -5
226 wall - I’ve recently got a large lot (~190) L5640s, all 2010 batches and they are all pretty terrible for bclk. The best pair I’ve found so far (my R2E blew up again after about 100 of them) does 250 bclk post on air, maybe a dozen do 240 and the rest all sits below that. It appears the only way to reliably get >250 bclk is to buy a handful of 2011 batches, since I had about a 50% success rate there. As for that stability wall, I haven’t encountered that one in particular, though I mostly stopped doing ambient testing at high clocks/semi-high volts (if the chips can do it with <1.5V) go up to 5GHz for quick/single threaded benches or higher for valids (my E5649 validates ~5.3 air 1.5V if I remember right). I have, however, encountered a hard stability wall at some point, but going the other way with bloomfield. My 920 can run 32M all day at 4.5/1.205V, but at 4.5/1.2V it instantly crashes even if clocked up after boot.
|
|
|
Post by ground on Feb 7, 2020 18:51:37 GMT -5
Quick update doing more air experiments, looks like the PCIe slot used can make a small difference in maximum bootable PCIe clock, just got my Rampage II to boot 119 by using the central slot, but its anything but reliable, maybe 1 in 5 boots succeeds. Just got a R3E to temporarily replace my R2E and R3F and got stuck (couldn't even up in Setfsb) at 118 PCIe in the third PCIe x16 slot. Decided to try the other slots: x16 #1 - 120 boot, 134 max x16 #2 - 120 boot, 136 max x16 #3 - 118 boot, 118 max x16 #4 - 118 no boot, no further testing x4 - 118 boot, 119 max
|
|