Memory Timing Benchmark (CL/tRCD/tRP/tRAS/CR/tRC/tRFC/tREF)

[namegt]

Hey, DgLee
Many people refer to memory timing will be good.
Very good.
Welcome to XS.

곱등이 대근이 환영!

[DGLee]

Hey, DgLee
Many people refer to memory timing will be good.
Very good.
Welcome to XS.

곱등이 대근이 환영!

Hi bro lol
I'm not a 곱등이

[aslanwas]

ôw, thanks for this nice job, all done on your own ? you get

[DGLee]

ôw, thanks for this nice job, all done on your own ? you get

Yup Thanks for appreciation

[Formula350]

Nice, and very thorough! Someday you'll have to go more indepth and see what different combinations provide for performance.

One thing I've noticed was that, at least with my Micron D9 (JNL I think) OCZ Reaper HPC 1600 7-7-7-30 1.9v sticks is the tWRWR (Write to Write Delay) and tRDRD (Read to Read Delay) needs be the same for best memory performance. If they are even one apart it will drop my Copy and Write bandwidth down about 200-300MB/s.

Just did a quick run of a few different sub-timings, from within Windows using CPU-Tweaker 1.5 and MaxxMEM² Preview (Note: I'm running 1333 because my BIOS is weird and some mystery setting/s bump it down from 1600 to 1333, even though 1600 is selected. Also the MaxxMEM² sys info does not update unless you restart the program. Run-to-run variance is about ±20MB/s Copy, ±1MB/s Read, ±8MB/s Write)
Specs:
ASRock 890FX Deluxe3
PhII X2 555BE @ 3200MHz, 2400MHz CPU-NB (CPU-NB brought the score up from 7.7GBs)
OCZ Reaper HPC 2x2GB 1600 7-7-7-30 1T. tRFC@90ns (any others just ask)

Row 1: (assuming 1680x1050 heh)
Image 1 - Base run (Maximum Read Latency@60)
Image 2 - Maximum Read Latency dropped to 55
Image 3 - CPU speed bumped up 100MHz via multiplier (0.5x increase), Max Read Lat@60



Row 2:
Image 1 - CPU +100MHz, Max Read Lat@55
Image 2 - CPU +200MHz, Max Read Lat@60
Image 3 - CPU +200MHz, Max Read Lat@55

Note: No clue why 3.4GHz dropped performance so much compared to 3.3Ghz

[DGLee]

Thanks a lot for your info, Formula350
Frankly I'd never thought of them - tWRWR & tRDRD. Really impressive that they affect performance that much.
Maybe I'll retry on them sooner or later

[Sanousie]

Welcome to XS

Very nice thread, good info, effort well done.

If you would be kind enough to post some info on other subtimings, Im having trouble getting these sticks to run at 2000mhz, maybe with some input from your subtimings i can get them to work, Thanks.

E.g here is a cut and paste from aida/everest under motherboard/chipset/Northbridge AMD K10 IMC

Memory Controller
Type Dual Channel (128-bit)
Active Mode Dual Channel (128-bit) - Unganged

Memory Timings
CAS Latency (CL) 8T
RAS To CAS Delay (tRCD) 8T
RAS Precharge (tRP) 8T
RAS Active Time (tRAS) 26T
Row Cycle Time (tRC) 27T
Command Rate (CR) 1T
RAS To RAS Delay (tRRD) 4T
Write Recovery Time (tWR) 8T
Write To Read Delay (tWTR) 4T
Read To Precharge Delay (tRTP) 4T
Four Activate Window Delay (tFAW) 18T
Write CAS Latency (tWCL) 6T
Refresh Period (tREF) 7.8 us
DRAM Drive Strength 1.0x
DRAM Data Drive Strength 1.0x
Clock Drive Strength 1.5x
CKE Drive Strength 1.5x
Idle Cycle Limit 16

Error Correction
ECC Supported, Disabled
ChipKill ECC Supported, Disabled
RAID Not Supported
DRAM Scrub Rate Disabled
L1 Data Cache Scrub Rate Disabled
L2 Cache Scrub Rate Disabled
L3 Cache Scrub Rate Disabled

[Formula350]

Thanks a lot for your info, Formula350
Frankly I'd never thought of them - tWRWR & tRDRD. Really impressive that they affect performance that much.
Maybe I'll retry on them sooner or later

I had a friend change them on his DDR2 and didn't show any benefit in Sandra for RDRD and WRWR(For some reason it won't install on my computer so I can't compare that way, but I'm going to reinstall 7 soon anyways), but dropping down just the Max Read Latency by 2ns resulted in a 250-300mb/s improvement!

Welcome to XS

Very nice thread, good info, effort well done.

If you would be kind enough to post some info on other subtimings, Im having trouble getting these sticks to run at 2000mhz, maybe with some input from your subtimings i can get them to work, Thanks.

E.g here is a cut and paste from aida/everest under motherboard/chipset/Northbridge AMD K10 IMC

Memory Controller
Type Dual Channel (128-bit)
Active Mode Dual Channel (128-bit) - Unganged

Memory Timings
CAS Latency (CL) 8T
RAS To CAS Delay (tRCD) 8T
RAS Precharge (tRP) 8T
RAS Active Time (tRAS) 26T
Row Cycle Time (tRC) 27T
Command Rate (CR) 1T
RAS To RAS Delay (tRRD) 4T
Write Recovery Time (tWR) 8T
Write To Read Delay (tWTR) 4T
Read To Precharge Delay (tRTP) 4T
Four Activate Window Delay (tFAW) 18T
Write CAS Latency (tWCL) 6T
Refresh Period (tREF) 7.8 us
DRAM Drive Strength 1.0x
DRAM Data Drive Strength 1.0x
Clock Drive Strength 1.5x
CKE Drive Strength 1.5x
Idle Cycle Limit 16

What brand and model of DDR3 are you running? With out knowing the factory specs it's hard to say what to really adjust.

I won't pretend to know what I'm talking about with the DDR2/3 setups since I upgraded from S939 with DDR1, BUT...

General rule of thumb is: tRAS + tRP = tRC, so for you 26+8=34. You have it set at 27. I don't think that it's preventing you from hitting 2000MHz, but it's a start.

A few other things I'd be try changing to:
CL-tRCD-tRP - 9-9-9
Command Rate - 2T
Write CAS Latency (tWCL) - 8T (If you can change this)

Increase the voltage of the memory a little
Increase CPU-NB voltage (that is the memory controller)
Increase the actual NB voltage (have heard it helps a little)

That's all I can think of w/o knowing you memory model. GL!

[Sanousie]

I had a friend change them on his DDR2 and didn't show any benefit in Sandra for RDRD and WRWR(For some reason it won't install on my computer so I can't compare that way, but I'm going to reinstall 7 soon anyways), but dropping down just the Max Read Latency by 2ns resulted in a 250-300mb/s improvement!




What brand and model of DDR3 are you running? With out knowing the factory specs it's hard to say what to really adjust.

I won't pretend to know what I'm talking about with the DDR2/3 setups since I upgraded from S939 with DDR1, BUT...

General rule of thumb is: tRAS + tRP = tRC, so for you 26+8=34. You have it set at 27. I don't think that it's preventing you from hitting 2000MHz, but it's a start.

A few other things I'd be try changing to:
CL-tRCD-tRP - 9-9-9
Command Rate - 2T
Write CAS Latency (tWCL) - 8T (If you can change this)

Increase the voltage of the memory a little
Increase CPU-NB voltage (that is the memory controller)
Increase the actual NB voltage (have heard it helps a little)

That's all I can think of w/o knowing you memory model. GL!

Patriot 2000mhz LL kit, rated at 8 8 8 26 , but thats on a p55 platform, with a Trc of 90 in the XMP profile, I found out the ram does not like low TRC, its been a night mare tr ying to get the sy stem stable with this ram, I cant find nothing on google lol.

I can get 1800mhz 7 7 7 20 stable, but anything past that, nothing seems to work, it cant be the cpu-NB its running at 1.475 at 3.4ghz, Im on SS phase.

All i would like is some sub timings info, for some reason i can never find anything on the net about sub timings, So im led to believe that everyone uses the Auto settings?

[STEvil]

Nice info.

You could add a memory benchmark specific graphics test like Dragothic Low detail (game 2 test 1) from 3dmark 2001 to see FPS changes. I wouldnt use all the benchmarks of 3d2001, thats a lot of running time lol. Although if you do, be careful with the low detail car test (game 1 test 1) because it has "good runs" and "bad runs" due to the way it was made.

[Formula350]

For anyone interested... Using MaxxMEM (I really need to figure out why Sandra errors when loading...) and all the same timings and CPU speed, here is something to consider:
CPU 3.2GHz, CPU-NB 2400MHz, DDR3 @ 1333 6-6-6
Mem Copy: 11405MB/s
Mem Read: 9889MB/s
Mem Write: 7971MB/s
"Reached Memory Score: 8.93GB/s"
Latency: 57.0ns

CPU 3.2GHz, CPU-NB 2000MHz, DDR3 @ 1600MHz 6-6-6
Mem Copy: 12020MB/s
Mem Read: 9548MB/s
Mem Write: 6712MB/s
"Reached Memory Score: 8.13GB/s"
Latency: Forgot to check, but it would've only been around 55ns


So, with thus far being able to run at a lower memory voltage, I've decided to ditch 1600 for now till I work the quirks out with my system :\

[NoGuru]

Nice review. I'm going to do a little testing soon to see how I compare. I will try and give results if I remember.

[bassplayer]

Good bump Guru. This is very helpful when trying to figure out how to cut down Pi times!

[nvmc]

Nice to see someone has the time to do the benches we all want to see.
This certainly flies in the face of what I previously thought to be true - tRCD being a (relatively) major factor in performance on AM3 platforms. Interesting.

Maybe I won't have to go and drop some money getting a SuperTalent Hyper kit after all.

[Okda]

nice review

on a Sandy bridge system, will the results be the same or nearly the same , or not ?

[kikicoco1334]

nice info!