F5F Stay Refreshed Hardware Desktop Check Ram schedules against speed ratings—Haswell-E 3200 at C22 versus 2666 at C15.

Check Ram schedules against speed ratings—Haswell-E 3200 at C22 versus 2666 at C15.

Check Ram schedules against speed ratings—Haswell-E 3200 at C22 versus 2666 at C15.

M
MrKryp
Senior Member
Find
643
02-04-2025, 10:02 PM
#1
I checked my performance with both settings and found that running at higher timings like 22 22 22 40 gives a noticeable boost compared to the older 15 15 15 35 at 2666mhz. It’s definitely better in terms of speed.
Reply
Reply
M
MrKryp
02-04-2025, 10:02 PM #1

I checked my performance with both settings and found that running at higher timings like 22 22 22 40 gives a noticeable boost compared to the older 15 15 15 35 at 2666mhz. It’s definitely better in terms of speed.

Reply
Reply
S
Slyseade
Member
Find
51
02-07-2025, 05:40 PM
#2
Are you certain it won't affect performance at 3200MHz? I understand the memory controller on Haswell-E is problematic, but 3200MHz CL16-19-19-39 should work on most systems and chips. These seem to be auto-timings, as suggested by CL22. The latency at 3200MHz CL22 is quite high compared to 2666 15-15-15-35, though it should offer more bandwidth. Consider running additional memory-intensive tests like SuperPi 32M, Y Cruncher, PyPrime, etc., to confirm any gains at this frequency. If 3200MHz CL22 is your limit, it might be wiser to move down to 3000 and focus on improving timings significantly. Often, if stability suffers that much at higher speeds, lowering the speed multiplier and tightening timings further is a better approach.
Reply
Reply
S
Slyseade
02-07-2025, 05:40 PM #2

Are you certain it won't affect performance at 3200MHz? I understand the memory controller on Haswell-E is problematic, but 3200MHz CL16-19-19-39 should work on most systems and chips. These seem to be auto-timings, as suggested by CL22. The latency at 3200MHz CL22 is quite high compared to 2666 15-15-15-35, though it should offer more bandwidth. Consider running additional memory-intensive tests like SuperPi 32M, Y Cruncher, PyPrime, etc., to confirm any gains at this frequency. If 3200MHz CL22 is your limit, it might be wiser to move down to 3000 and focus on improving timings significantly. Often, if stability suffers that much at higher speeds, lowering the speed multiplier and tightening timings further is a better approach.

Reply
Reply
T
tmc00
Member
Find
180
02-10-2025, 02:25 PM
#3
2666 cl15 promises improved mathematical performance. Aim for the highest number of clocks per cycle—2666 divided by 15 equals 177.73, and 3200 divided by 145.45 gives 177.78.
Reply
Reply
T
tmc00
02-10-2025, 02:25 PM #3

2666 cl15 promises improved mathematical performance. Aim for the highest number of clocks per cycle—2666 divided by 15 equals 177.73, and 3200 divided by 145.45 gives 177.78.

Reply
Reply
J
Jotta
Member
Find
61
02-16-2025, 10:44 AM
#4
It's odd that C18 starts up, but only six out of eight sticks work—two seem to disappear without a trace. Memtest64 reports issues, C20 shows some errors but fewer than C18, and C20 itself has no errors.
Reply
Reply
J
Jotta
02-16-2025, 10:44 AM #4

It's odd that C18 starts up, but only six out of eight sticks work—two seem to disappear without a trace. Memtest64 reports issues, C20 shows some errors but fewer than C18, and C20 itself has no errors.

Reply
Reply
E
Einhorn_Kevin
Member
Find
94
02-16-2025, 05:21 PM
#5
I exchanged the settings for 15 15 15 30 and the Windows version feels more responsive. I’ll stop here now; even Cinebench was 100 points behind.
Reply
Reply
E
Einhorn_Kevin
02-16-2025, 05:21 PM #5

I exchanged the settings for 15 15 15 30 and the Windows version feels more responsive. I’ll stop here now; even Cinebench was 100 points behind.

Reply
Reply
R
Roe56
Member
Find
102
02-27-2025, 03:35 PM
#6
Sure, that matches X99. Adjusting to 3000 and exploring the possibilities. 3200MHz with eight DIMMs puts a strain on the memory controller, and when it hits its limits on Intel processors it often loses memory channels.
Reply
Reply
R
Roe56
02-27-2025, 03:35 PM #6

Sure, that matches X99. Adjusting to 3000 and exploring the possibilities. 3200MHz with eight DIMMs puts a strain on the memory controller, and when it hits its limits on Intel processors it often loses memory channels.

Reply
Reply
S
Sindyion
Member
Find
203
02-27-2025, 05:31 PM
#7
I understand! It looks like you might need to adjust the BLCK setting to reach 3000MHz, which would turn off Intel's power management. You mentioned you've already boosted the CPU to 4.6GHz at 1.25v and want to avoid constant full turbo so it stays around 300MHz from RAM.
Reply
Reply
S
Sindyion
02-27-2025, 05:31 PM #7

I understand! It looks like you might need to adjust the BLCK setting to reach 3000MHz, which would turn off Intel's power management. You mentioned you've already boosted the CPU to 4.6GHz at 1.25v and want to avoid constant full turbo so it stays around 300MHz from RAM.

Reply
Reply