(cf)Eclipse
Freelance Overclocker
- Joined
- Feb 18, 2003
- Messages
- 30,027
I've had this kit sitting around for a while, but between finals week, going home, moving into a new apartment and starting new classes, I've had depressingly little time to dedicate towards overclocking. After much frustration and excitement, I have finally completed the testing for this 2x2GB kit from Mushkin.
The rated specification is 400MHz at 5-4-4-12 with 2.0v, and they come with blue heatspreaders with Mushkin's signature loops on the top for some additional cooling. How well these loops work, I am unsure, as the memory ran pretty warm under my 120mm Delta as I approached 2.5v, but it's certainly better than nothing.
(click for larger pic)
Blue is my favorite color...
Overclocking:
So, when overclocking, a few interesting and frustrating things were found. With 4GB of memory, things like superPI 32M don't work very well. It's only using like 1/10th of the total space. However, to tweak the timings, I needed a quick test that would fail repeatably to give me a good baseline to work from and note any improvements on. Since the best thing was usually 3 instances of windows memtest for >10 minutes, a lot of the tuning was done through guess and check, until the transition between stable and unstable improved enough to fail consistantly in 32M.
However, it is absolutely imperative that you do proper tuning with this kit. I went from being stuck at around 540MHz with 5-4-3-12 with 2.5v, looser not helping to doing just north of 600MHz at 5-4-5-12 with the same voltage. Drive strengths are most important, if you have control over them. A low "normal" drive was ideal for me. In the DFI NF590's BIOS, "normal" was always used, with a drive level near 10, depending on the timings and voltage.
For those who wonder, the system used for testing was:
AMD Athlon64 4000+ (F3)
DFI LP NF590 SLI-M2R/G
7600GT
Powerstream 420
Bunch of fans
The maximum stable speed (pass 32M, memtest for >30 min and 3dmark05) at each significant timing level and a bunch of voltages:
I thought that the difference between tRCD 3 and tRCD 4 is interesting, since they all seem to start at the same MHz, plus or minus a few, with really low voltage. Then depending on the other timings used, it scales better or worse to hit some final value at 2.5v. For testing purposes, I felt that going over 2.5v was probably unwise for the time being. Further testing at higher voltages showed 2.6-2.65v to be ideal for suicide runs:
http://valid.x86-secret.com/show_oc?id=211113
And slightly less than that was ideal for benching, though with real cooling, heat would not be quite as much of a problem, and higher may work better.
CAS 3 isn't up there because it was basically useless, getting in the low to mid 200MHz range. CAS 6 wouldn't boot at 480MHz, with 2.3v. Either it needs less voltage, or CAS 6 simply fails
Some random screens I took during testing:
545.5MHz @ 5-4-4-12, 2.29v in BIOS
562.2MHz @ 5-3-4-8, 2.5v in BIOS
602.8MHz @ 5-4-5-12, 2.5v in BIOS
Doing low 24 min in 32M is actually pretty good for my setup at 3GHz. I suspect that the extra banks in the 1Gbit chips (8 vs the 4 in 512Mbit chips) allow for more interleaving, thus slightly higher memory efficiency at a given timing set and speed. Speaking of chips, some interesting guys in this kit. A quick look at micron's site says that they are 333MHz, 5-5-5 binned chips, based on the revD die, but it is not in any of the technical documentation. Through some somewhat obvious math, we know that the chips are a 128x8 configuration to give the required 1Gbit density and
enough width to make a double rank module, no more or less. Being revD, which is the same as D9GMH and DKX, I'm sure they are fine for 24/7 use in the 2.2-2.3v range with adequate cooling. However, ONLY with adequate cooling, as they put out a nice amount of heat
Note: I'll be putting a full review up on my site in a few days, for those who want some extra jibberish with performance comparisons as well
The rated specification is 400MHz at 5-4-4-12 with 2.0v, and they come with blue heatspreaders with Mushkin's signature loops on the top for some additional cooling. How well these loops work, I am unsure, as the memory ran pretty warm under my 120mm Delta as I approached 2.5v, but it's certainly better than nothing.
(click for larger pic)
Blue is my favorite color...
Overclocking:
So, when overclocking, a few interesting and frustrating things were found. With 4GB of memory, things like superPI 32M don't work very well. It's only using like 1/10th of the total space. However, to tweak the timings, I needed a quick test that would fail repeatably to give me a good baseline to work from and note any improvements on. Since the best thing was usually 3 instances of windows memtest for >10 minutes, a lot of the tuning was done through guess and check, until the transition between stable and unstable improved enough to fail consistantly in 32M.
However, it is absolutely imperative that you do proper tuning with this kit. I went from being stuck at around 540MHz with 5-4-3-12 with 2.5v, looser not helping to doing just north of 600MHz at 5-4-5-12 with the same voltage. Drive strengths are most important, if you have control over them. A low "normal" drive was ideal for me. In the DFI NF590's BIOS, "normal" was always used, with a drive level near 10, depending on the timings and voltage.
For those who wonder, the system used for testing was:
AMD Athlon64 4000+ (F3)
DFI LP NF590 SLI-M2R/G
7600GT
Powerstream 420
Bunch of fans
The maximum stable speed (pass 32M, memtest for >30 min and 3dmark05) at each significant timing level and a bunch of voltages:
I thought that the difference between tRCD 3 and tRCD 4 is interesting, since they all seem to start at the same MHz, plus or minus a few, with really low voltage. Then depending on the other timings used, it scales better or worse to hit some final value at 2.5v. For testing purposes, I felt that going over 2.5v was probably unwise for the time being. Further testing at higher voltages showed 2.6-2.65v to be ideal for suicide runs:
http://valid.x86-secret.com/show_oc?id=211113
And slightly less than that was ideal for benching, though with real cooling, heat would not be quite as much of a problem, and higher may work better.
CAS 3 isn't up there because it was basically useless, getting in the low to mid 200MHz range. CAS 6 wouldn't boot at 480MHz, with 2.3v. Either it needs less voltage, or CAS 6 simply fails
Some random screens I took during testing:
545.5MHz @ 5-4-4-12, 2.29v in BIOS
562.2MHz @ 5-3-4-8, 2.5v in BIOS
602.8MHz @ 5-4-5-12, 2.5v in BIOS
Doing low 24 min in 32M is actually pretty good for my setup at 3GHz. I suspect that the extra banks in the 1Gbit chips (8 vs the 4 in 512Mbit chips) allow for more interleaving, thus slightly higher memory efficiency at a given timing set and speed. Speaking of chips, some interesting guys in this kit. A quick look at micron's site says that they are 333MHz, 5-5-5 binned chips, based on the revD die, but it is not in any of the technical documentation. Through some somewhat obvious math, we know that the chips are a 128x8 configuration to give the required 1Gbit density and
enough width to make a double rank module, no more or less. Being revD, which is the same as D9GMH and DKX, I'm sure they are fine for 24/7 use in the 2.2-2.3v range with adequate cooling. However, ONLY with adequate cooling, as they put out a nice amount of heat
Note: I'll be putting a full review up on my site in a few days, for those who want some extra jibberish with performance comparisons as well