|
|
Post by Aleslammer on Mar 24, 2023 9:32:21 GMT -5
CPU in use limited to DD4 1600, been running 6s and 7s DDR4 1866 so gave 5-5-5-10 a try. Had to bump RAM voltage to 1.7, from 1.65v to get 4Gs stable. Did run y-cruncher at 6s full mem load. Don't know if it's real but what the heck.   |
|
|
|
Post by Bones on Mar 24, 2023 12:09:55 GMT -5
Try it with both timing sets/tweaks in Super PI at the same speed to see if it's real or not.
That's all I can suggest about it.
|
|
|
|
Post by Aleslammer on Mar 24, 2023 16:55:35 GMT -5
Bones Made me do a little testing, biased towards the 5s. Ran 5s once then used 7s three times to see if I could catch it. SP32 as sub'd with 5-5-5-10 timings, 16m 36.498. 105.9 bclk, all 20 cores & HT running, system lockup at 106 bclk. Tests, 105.7 bclk, 2 cores active, HT off, fresh boot for each test, GB3 ran before SP32 Best @ 7-7-7-14, 16m 34.124s, runner up 16m 36.352s, last 16m 36.739s Base @ 5-5-5-10, 16m 32.186s Do see a difference between 7s and 5s, so I'd say 5s work but with out 6s still have some gray area. Would of used 6s but from what I'm seeing, glad I didn't, would need more than 4 runs to see a difference. Best, 7-7-7-14  Base. 5-5-5-10 
|
|
|
|
Post by MachineLearning on Mar 24, 2023 17:53:20 GMT -5
Wow, I never thought any DDR4 could do those kinds of timings at any frequency... let alone with effect.  |
|
|
|
Post by Mr.Scott on Mar 24, 2023 18:01:13 GMT -5
My question is, how does cl5 @ 1600 fair against CL12 @ 4000 ?
|
|
|
|
Post by georgekokovinis on Mar 24, 2023 19:04:14 GMT -5
My question is, how does cl5 @ 1600 fair against CL12 @ 4000 ? It is impossible to reply. DDR4 b-die mems (A1) in this case, are not meant to be used on Haswell. Kudos to Brad, but only if there was a cpu capable to run 800/5,5,5,10 and 4000/12,11,11,24 we might be able to reach a conclusion. X99 platform does not support these 2 scenarios. Maybe x299 could. It does, I ran quad channel 4000/12,12,12,28 on it. But I do not have X299 any more,and even if I did, I don't have the time. |
|
|
|
Post by georgekokovinis on Mar 24, 2023 19:28:01 GMT -5
Btw, I just recalled running 2800/8,8,8,10 and 2400/7,7,7,9.
1.88v was required.
I come to think that 1600/5,5,5 would be slower. For sure.
In any case these are experimental cases.
Hybrid combinations of HW that was never designed for such use.
You did great Brad. Bravo.
|
|
|
|
Post by Bones on Mar 24, 2023 20:43:24 GMT -5
Intriguing! I'm wondering what my sets can do, and there is only one way to find out - Just not now, it will have to be later when I do.  |
|
|
|
Post by Aleslammer on Mar 25, 2023 6:07:58 GMT -5
Playing here based on CPU limits, this CPU just happened to be the first one I've seen with a DDR4 1600 limit. Have had the board post a few of the DDR4 1866 CPUs at 2133 with all auto settings but once manual set became a no post, even spent a little time trying. Do have three i7 Broadwells upcoming so could do a little comparison. IMO George's statement will ring true if comparing highest to slowest mem frequency, might be a mid-point where tight slower wins out but wouldn't bet the house on it.
|
|
|
|
Post by georgekokovinis on Mar 25, 2023 7:33:11 GMT -5
Playing here based on CPU limits, this CPU just happened to be the first one I've seen with a DDR4 1600 limit. Have had the board post a few of the DDR4 1866 CPUs at 2133 with all auto settings but once manual set became a no post, even spent a little time trying. Do have three i7 Broadwells upcoming so could do a little comparison. IMO George's statement will ring true if comparing highest to slowest mem frequency, might be a mid-point where tight slower wins out but wouldn't bet the house on it. Huge respect to you my friend. |
|
|
|
Post by Mr.Scott on Mar 25, 2023 8:31:42 GMT -5
All I needed to hear. Thank you George and Brad. |
|
|
|
Post by georgekokovinis on Mar 25, 2023 8:48:02 GMT -5
All I needed to hear. Thank you George and Brad. As always, you are most welcome my good friend P.S. I like very much what Brad does. Remembers me partially of myself. Following the path of others and hoping for a good result is the easy way. The challenge is to have the patience and a bright innovative mind to create your own path. That is the "scientific" part of this hobby, and if you ask me, the most interesting one. |
|
|
|
Post by Aleslammer on Mar 27, 2023 6:35:21 GMT -5
Little more testing. See screen shots for CPU speeds and uncore info, same for all, RAM voltage 1.80v except DDR4 3467, 1.85v Bench = SP32, Average of 3 5-5-5-10, DDR4 1600 414.989 seconds 9-9-9-18, DDR4 2800 405.492 seconds 10-10-10-20, DDR4 3000, usable if you can't get 3400, looking good compared to 3200 403.452 seconds 12-12-12-24, DDR4 3200 404.209 seconds 11-11-11-24, DDR4 3200, didn't try for 10s 403.499 seconds 12-12-12-24, DDR4 3400, most consistent 402.559 seconds 11-11-11-24, DDR4 3400, most used based on Bot screen shots 402.199 seconds 13-13-13-26, DDR4 3467, 12s would post but no more, didn't try to force it. Leeghooft is running 11's on this divider with a i7 6950X. 403.704 seconds DDR4 3600 was a no post with any timings tried. Best 5-5-5-10  Best overall  Best 13-13-13-26  |
|
|
|
Post by georgekokovinis on Mar 27, 2023 6:48:05 GMT -5
Brad, Since X99 has a by default limit at 3400 ram speed, your 3400/12-12 is MAGNIFICENT. Search no further. Mission accomplished |
|
|
|
Post by Mr.Scott on Mar 27, 2023 6:55:07 GMT -5
Exactly what I wanted to see. George, you were right on.  |
|
|
|
Post by Vinster on Mar 27, 2023 22:08:45 GMT -5
wow, Brad that is insane man... Master at work making it look like childs play... you are freaking awesome
Vin
|
|
|
|
Post by Aleslammer on Mar 28, 2023 9:02:40 GMT -5
Just parroting most of the higher speeds, lower speed timings were tried just because. Have seen 7s & 8s used on some of the Xeons. Played a little more this morning, added some info for 9-9-9-18 DDR4 2800 to the list above. Did make windows at 1.725v with everything on auto but RAM timings and voltage, used 1.80v for the test. Best 9-9-9-18 DDR4 2800, Uncore 39x as are the others above.  Effect of a lower uncore, 9-9-9-18 DDR4 2800, Uncore 35x, couldn't go higher. Average of three, 407.169 seconds.  Added: After looking through some of the i7 6950X results, the one I'm most impressed with. Maddmutt's DDR4 3354, 10-10-10-16, CPU running on the 125 bclk strap. |
|
|
|
Post by alpi on Mar 31, 2023 13:59:20 GMT -5
Little more testing. See screen shots for CPU speeds and uncore info, same for all, RAM voltage 1.80v except DDR4 3467, 1.85v Bench = SP32, Average of 3 5-5-5-10, DDR4 1600 414.989 seconds 9-9-9-18, DDR4 2800 405.492 seconds 10-10-10-20, DDR4 3000, usable if you can't get 3400, looking good compared to 3200 403.452 seconds 12-12-12-24, DDR4 3200 404.209 seconds 11-11-11-24, DDR4 3200, didn't try for 10s 403.499 seconds 12-12-12-24, DDR4 3400, most consistent 402.559 seconds 11-11-11-24, DDR4 3400, most used based on Bot screen shots 402.199 seconds 13-13-13-26, DDR4 3467, 12s would post but no more, didn't try to force it. Leeghooft is running 11's on this divider with a i7 6950X. 403.704 seconds DDR4 3600 was a no post with any timings tried. Cool session and it shows really clearly why higher mem clocks can be better even if the main timings are growing slightly higher. (percentage) Bdie is really awesome btw ! Such a crazy stuff. You can do anything with it. Unreal flexibility and ability ofc. Paired with the best voltage tolerance and scaling.  |
|
