|
|
Post by Aleslammer on Oct 17, 2021 15:38:02 GMT -5
Someone on the Bot was talking up the board and that it would bclk clock Sky Lake Xeons, Being a Xeon fan boy picked one up, been sitting the top self gathering dust for awhile now in fact a lot of dust. It's a C232 chipset board and max RAM speed is DDR4 2133 so no where in line with a Z170. It also doesn't have a lot of bells & whistles but at under $150 new what do you expect. It will run the the RAM at DDR4 3200 if you get the bclk high enough but nothing over that would post. The first CPU I tried was a E3 1270 V5 with a multi of 36 and a max bclk of 129 producing a CPU freq of 4642 and a RAM speed of 2750 at 13-13-13-32 1T, timings were a little better than most I saw. When I got up this morning I decided to try and improve timings, if B dies will run 12s at 4k they got to do better than 13s at 2750. Spent a little time on line looking and found an article on Tweak Town talking about Z170 bclk and K chip clocking. www.tweaktown.com/guides/7481/tweaktowns-ultimate-intel-skylake-overclocking-guide/index.html#Overclocking-Flow-Chart-Non-K-SKUs-BCLK-and-MultipliersCut this image from a SS concerning what voltages I was interested in.  Using 2 sticks from a 4x8 kit of G. Skill F4-3600C16Q-32GTZSW and voltages as follows. CPU vcore 1.45v LLC, Level 1 DRAM voltage. 1.55v DRAM activating power supply, 2.7v PCH +1.0 voltage, 1.1v VCCIO voltage, 1.2v VCC PLL voltage, 1.3v VCCSA voltage, 1.25v That's the whole voltage adjustment menu in BIOS. Used the E3 1270 V5 set the bclk at 125 and started playing with the RAM at 2666 Moving the bclk to 129, fluid temp at 5c 13s from before  10s  Mostly primary's did change tREFI to 65535 max setting. Have never played with any DDR4 below 3200 before, so have no idea what to expect. |
|
|
|
Post by Aleslammer on Oct 17, 2021 15:38:43 GMT -5
Now for some odd things. This is the placement for the E3 1270 V3 on the Aida64 bench mark list, sorry haven't updated Aida64 for awhile and I never go online with my benching setups if I don't have to. Doesn't look to bad.  This is the score for HWBOT x265 1080P, same settings from the above screen shots first post.  This score is with the bclk set to auto, default speed for the CPU.  This also shows up in some of the Aida64 benches. CPU Zlib CPU SHA3 FPU Julia FPU Mandel FP 32 Ray Trace FP 64 Ray Trace After looking at results XTU is affected also but it is an Intel bench. (I knew this but forgot) Added: Y-Cruncher CineBench R20, if you get your CPU frequency high enough you can make up the lost and get a better score. Core temp sensors all show 100c in Aida64, the other CPU related sensors work fine. If bclk is set to auto core sensors read properly. Added: Max Vcore is 1.52v Added: 10/24/21 XTU 2.0 Allows bclk clocking on fixed multi Sky Lake CPUs |
|
|
|
Post by georgekokovinis on Oct 17, 2021 16:03:34 GMT -5
2666 cl10 is the best to be expected.
As a matter of fact, it is a miracle considering the rather average board.
|
|
|
|
Post by Aleslammer on Oct 18, 2021 11:04:39 GMT -5
Something to play with, just hoping the timings will hold up to 3200.
Picked up a E3 1240L V5 has a 21 multi, only one submission for the cpu, CPU freq , had the bclk a touch over 195 going to see if I can catch it.
|
|
|
|
Post by Aleslammer on Oct 18, 2021 23:46:30 GMT -5
Had to raise RAM volts to 1.625 but got to 1522 mem running two cores @ 5K for PF. Mem read little under 4.9, Aida64 would lock the system up when I tried to open it any higher.  |
|
|
|
Post by georgekokovinis on Oct 19, 2021 7:18:30 GMT -5
Something to play with, just hoping the timings will hold up to 3200. Picked up a E3 1240L V5 has a 21 multi, only one submission for the cpu, CPU freq , had the bclk a touch over 195 going to see if I can catch it. 3200 - 10/10/10/28/1T ? Well... not impossible, since I saw Splave do 3600 /11/11/11/24/1T. But it was on an Apex X, with an I7-8700K under LN2, mem sticks cooled with LN2, 2.2V on mems and 1.50VccIO and 1.65VccSA. To be more clear. The assumption that " if they can do 4000 / 12, then they can do 3600 / 11 and 3200 / 10 " is absolutely incorrect. Not your assumption Brad  I have heard and read it many times. Books have been written on how mem speed is tied to timings. The fact is efficiency. I have been multiple times thru the years surprised, getting in SPI32m ( reference benchmark ), better results with a tad looser timings and a bit lower speed. Pressing too much can and will have adverse effects and efficiency starts dropping. Anyway, I would be very satisfied running 2800 / cl10. Good luck ! |
|
|
|
Post by Aleslammer on Oct 19, 2021 10:59:05 GMT -5
Something to play with, just hoping the timings will hold up to 3200. Picked up a E3 1240L V5 has a 21 multi, only one submission for the cpu, CPU freq , had the bclk a touch over 195 going to see if I can catch it. 3200 - 10/10/10/28/1T ? Well... not impossible, since I saw Splave do 3600 /11/11/11/24/1T. But it was on an Apex X, with an I7-8700K under LN2, mem sticks cooled with LN2, 2.2V on mems and 1.50VccIO and 1.65VccSA. To be more clear. The assumption that " if they can do 4000 / 12, then they can do 3600 / 11 and 3200 / 10 " is absolutely incorrect. Not your assumption Brad I have heard and read it many times. Books have been written on how mem speed is tied to timings. The fact is efficiency. I have been multiple times thru the years surprised, getting in SPI32m ( reference benchmark ), better results with a tad looser timings and a bit lower speed. Pressing too much can and will have adverse effects and efficiency starts dropping. Anyway, I would be very satisfied running 2800 / cl10. Good luck ! I agree with all you said, really didn't think I'd do better than 12s. Your commit about efficiency got me wondering so did a little looking on the Bot, found a bclk clocked i7 6700 searching PF scores 6700 hwbot.org/submission/4448741_digg_de_pifast_core_i7_6700_12sec_590msE3 1240 V5 hwbot.org/submission/4837723_aleslammer_pifast_xeon_e3_1240_v5_12sec_660msI'm not going to complain. |
|
|
|
Post by Aleslammer on Oct 20, 2021 9:24:44 GMT -5
DDR4 3200, CL 10 ain't going to happen, best I could get, ram voltage @ 1.725 everything else the same.  Best I can do on the primaries, tRAS is as low as the BIOS will allow. Asrock's memory configurator doesn't work on the board, least the one I tried. |
|
|
|
Post by georgekokovinis on Oct 20, 2021 10:04:01 GMT -5
THAT is amazing. I am very happy for you |
|
|
|
Post by Aleslammer on Oct 27, 2021 16:53:21 GMT -5
Been thinking about posting this so why not as XTU is all about mem on the locked CPUs
Using XTU from the Intel Version 7.5.3.3 and a E3 1240L V5 CPU
5c Fluid
Patriot Viper Elite DDR4 2133 2x4 14-14-14-32-278 1T (XMP)
Average of 5 = 1026.0 with a high of 1032 in XTU
G. Skill Trident Z @ DDR4 2133 2X8 10-10-10-28-180 1T (Stock is DD4 3600 C16)
Average of 5 = 1062.6 with a high of 1066
Should of stopped here!!!!!
Pulling the handle 15 more times on the G. Skills, reboot after every 5
Average of 20 = 1064.1 with a high of 1075
Two scores stand out a 1072 and 1075 the rest fell between 1058 & 1067
Raising fluid to 20c & vcore to 1.52v produced a 64c load temp ( I'm happy with this though I was hoping for more)
Average of 20 = 1063.3 with a low of 1055 and a high of 1070
All information above is:
Long /Short Duration Power Limit = 4095
Long Duration Maintained = 1 second (LDM).
Going back to 5c fluid
Testing all the LDM settings in XTU, Long & Short Duration Power Limits @ 4095
Based on 5 run averages for LDM, no reboots between changes
Lowest score 12 seconds, average of 5 = 1057.8 with a high of 1060
Highest score 8 seconds, Average of 5 = 1065.0 with a high of 1068
28 seconds as it seems popular, Average of 5 = 1061.6 with a high of 1066
128s used as a constant, Average of 5 = 1063.6 with a high of 1068
Not knowing if the settings within XTU were actually doing anything I manual set them in BIOS.
8 seconds, Average of 5 = 1065.4 with a high of 1073 ( second highest score thus far)
12 seconds, Average of 5 = 1064.8 with a high of 1066
128 seconds, Average of 5 = 1062.8 with a high of 1069
Next I looked at the Long / Short Power Limits using the 8 second LDM, I'm getting board so only two pulls per setting.
3095 = 1068 & 1066
2095 = 1078 & 1075 (highest score)
1095 = 1058 & 1061
Going back to the 2095 setting never saw anything close, all falling below 1066 (probably pulled the handle 30 plus times on more than a few reboots.
In the end all scores fell between 1055-1070 with fliers at 1072, 1073, two 1075 and 1078, (300 plus pulls on the handle).
From what I'm seeing all those settings in XTU really don't do anything, fliers are fliers. never liked this bench cause I sucked at it.
|
|
|
|
Post by Aleslammer on Oct 28, 2021 15:26:06 GMT -5
Put the E3V5 away, switched to a ASRock Z170 OC Formula, best at 10s so far is DDR4 2900. XTU 2.0, RAM at DDR4 2800 only been working on the primaries so far. Good for first for 4 core, sure it won't last long.  |
|
|
|
Post by Aleslammer on Oct 31, 2021 7:56:48 GMT -5
Thread sort of became about tight timing's low mem speed. Playing on a ASRock Z170 OC Formula CPU vCore 1.32v Boot Dram Voltage 1.82v Eventual Dram Voltage 1.78v VCCIO 1.25v VCCSA 1.26v 9s  XTU, new version a lot faster so no real comparison.  Little slower than 10s on most things but have only played with the primaries. The Geekbenches do not like it. Tops out about DDR4 2880 |
|
|
|
Post by Aleslammer on Oct 31, 2021 11:20:35 GMT -5
Used an image from the memory threads that George posted for Lila. Aida saying the board is adding a little extra voltage than BIOS.  |
|
|
|
Post by Aleslammer on Nov 1, 2021 10:05:24 GMT -5
Little faster, tight as I could get.  |
|
|
|
Post by freeagent on Nov 1, 2021 17:46:39 GMT -5
That's pretty tight.. Nice pi.
|
|
|
|
Post by Aleslammer on Nov 3, 2021 5:33:34 GMT -5
Internet was down yesterday, so for something to do I picked up the Z170 manual while waiting during runs and discovered memory was in the wrong slots (head slap moment, knew this forgot). Didn't think about it when I switched from the E3V5 and put the sticks in the A/B 1 slots. Switching to the A/B 2 slots, 9s @ 2800, 10s @ 3200, 11s @ 3466 (will post @ 3600, no windows), and 12s @ 3866, 4K was no go but didn't expect it to post. So for comparison I tightened up the 9s at 2800, ran the 6100T up to match the 6320 (from above), as close as I could get and got this.  Latency's are comparable but the rest suck! Might try to get 3200 running on the A/B 1 slots just to see if it scales higher than the E3V5 @ 3100. |
|
