Sign in with
Sign up | Sign in
Your question

OC E4300 - Tradeoff between CPU clock and memory bandwith

Last response: in Overclocking
Share
April 16, 2007 6:00:31 PM

I'm working on overclocking an E4300. Unfortunately, I probably bought the wrong memory - A-Data Vitesta DDR2 667 (PC2 5300). I'm thinking that the memory is limiting my ability to overclock. The spec on the memory just says "CAS 5", but it appears to work ok at 680 MHz with timings of 5-5-5-12. The memory was cheap, so I bought it with hopes that it would work well - now I have it, so I have to make it work as well as possible.

I've found two scenarios that appear to work well (pass MemTest, pass Orthos, appear to work ok with Windows):

1) Core freq of 320 MHz, with memory clocked at 800 MHz (ratio of 5:4). This yields a CPU frequency of 2.88 GHz and memory bandwidth (reported by MemTest) of 4.7 GB/s.

2) Core freq of 340 MHz, with memory clocked at 680 MHz (ratio of 1:1). This yields a CPU frequency of 3.06 GHz and a memory bandwidth (again from MemTest) of 2.7 GB/s.

So, assuming the memory bandwidth reported by MemTest is meaningful, it appears that scenario '1' could be the overall winner even though the FSB is clocked a little slower.

Any ideas? If no definitive yes/no answer, what test should I run to decide which is the better overall configuration?
April 16, 2007 6:18:17 PM

#2. Noticeably higher clockspeed and the benefits of running memory faster than 1:1 are slim. It's best to run at 1:1 and tighten the timings if you can. Desktop tasks generally don't need that much bandwidth anyway.
April 16, 2007 6:33:07 PM

If you can run this memory at 400 MHz - you can run it stable at ANY speed below. If not - the mobo is limiting factor. What is it? Secondly - what do you use your PC for? Gaming? Office? Multimedia? Knowing this makes the discussion about bandwidth more sensible.
April 16, 2007 7:05:30 PM

Application: desktop apps, video editing, some light gaming

Hardware:
Asus P5B-Deluxe WiFi motherboard
E4300 w/ Scythe Ninja Plus heatsink
2 GB A-Data Vitesta DDR2-667 (5-5-5-15? Spec only mentions CAS)
EVA GeForce 7600 GS 256MB PCIe
three RAID arrays (0,1,0) on 4 SATA-II disks (2 500 GB, 2 160 GB)
Antec P180 case
Envision EN500W EPS power supply
Lite-On LH-20A DVD writer

The Asus BIOS is somewhat cryptic, but here's what I've seen so far:

1) You set the core clock frequency and the BIOS gives you a choice of memory speeds (starting at 1:1, then going to 5:4, so on - about 3-5 choices)

2) You can modify the VCore and VDIMM voltages (among others). I have them raised very slightly above nominal in the 3.06 GHz case, nominal in the 2.88 GHz case. I can provide details if it helps.

3) There are lots of parameters for memory timing, some of which have obscure (to me) names. As I went above the 2.88 GHz case, I took these out of the "Auto" state (determined by SPD?) and started tweaking to tighten them. For the 3.06 GHz case, I ended up with timings of 5-5-5-12, with TRFC at 35 (instead of 42).

4) My testing sequence for any new configuration went like this:
a) does it boot at all?
b) does MemTest8+ (bootable) run for an extended period (several hours)
c) can I get into Windows?
d) can I run Orthos for several hours?

Issues I currently have:

A) I'm running a little hot. It is hard to tell which monitoring program is giving the right answers, but I think my overall CPU temp (as opposed to junction temp) is running a little over 50 deg-C under load. I'm concerned that as ambient temp goes higher in the summer I'll overheat. I've ordered some Shin-Etsu 23 TIM to replace the AS-5 I started with - when it comes in I plan to lap the heatsink and try it.

B) I'm not sure what to do about the apparent tradeoff of "clock speed" versus memory bandwidth. If MemTest86 reports meaningful memory bandwidth (i.e., that results in perceptible difference in real apps), then maybe I should go for a lower core freq with a 5:4 ratio of core freq to memory clock. However, I've also heard that 1:1 is really more efficient, so maybe I should stick with that.

This thread is really about issue 'B' (but I'll take advice on 'A' too!). Is the tradeoff obvious, or should I be running a benchmark to test it? If a benchmark, which one would be most apppropriate?
April 16, 2007 9:39:41 PM

@A - NOTHING will put such thermal&electrical load on CPUs as Orthos (Priority 9, Small FFTs), except TAT. So having slightly more than 50 degrees on CPU (assuming core temps about 65-70C) is not critical. If your CPU is L2 stepping - it might be a way higher. Use CPU-Z to determine stepping, CoreTemp for temps, PCProbe for voltages.

@B - The difference between 2.7 & 4.7 GB/s is VERY meaningful, in my opinion. I can see 4.7 GB/s at 400 MHz and 4.0 GB/s at 401 MHz, because of CPU Strap change on Asus P5B series and even this is meaningful for me. But to test it, you should use really big video file and your preferred editor/encoder, to measure the difference in real time.
April 17, 2007 4:18:41 AM

Thanks (again).

Quote:
@A - NOTHING will put such thermal&electrical load on CPUs as Orthos (Priority 9, Small FFTs), except TAT. So having slightly more than 50 degrees on CPU (assuming core temps about 65-70C) is not critical. If your CPU is L2 stepping - it might be a way higher. Use CPU-Z to determine stepping, CoreTemp for temps, PCProbe for voltages.


CoreTemp shows about 71-72 deg-C, ASUS ProbeII shows about 61. I'm assuming that CoreTemp is showing the core junction temps and Probe is showing the temp at the heat spreader. All of above is running Orthos (P9, Small) as suggested - thanks. Does that makes sense? Reasonable (about 22 deg-C ambient)?

What's the deal with L2 stepping? I'm hearing not-so-wonderful things about it on the web (and, of course, that's what I have...).

Quote:
@B - The difference between 2.7 & 4.7 GB/s is VERY meaningful, in my opinion. I can see 4.7 GB/s at 400 MHz and 4.0 GB/s at 401 MHz, because of CPU Strap change on Asus P5B series and even this is meaningful for me. But to test it, you should use really big video file and your preferred editor/encoder, to measure the difference in real time.


Do you know of any canned benchmarks that would indicate likely performance when video editing? I haven't got anything fancy installed yet (on purpose, in case I trash the OS while trying to OC).
April 17, 2007 5:05:14 AM

use the memory bandwidth benchmark in everest, the one in memtest isn't all that good, its just a quick generalization
April 17, 2007 5:12:34 AM

Or you can use Sandra, which is synthetic, but I deem it somewhat accurate for these types of tests.
April 17, 2007 7:33:58 AM

Quote:
CoreTemp is showing the core junction temps
Correct.
Quote:
Probe is showing the temp at the heat spreader
Correct.
Quote:
Does that makes sense? Reasonable (about 22 deg-C ambient)?

What's the deal with L2 stepping? I'm hearing not-so-wonderful things about it on the web (and, of course, that's what I have...).
Me too, but it doesn't hurt us 8) It seems that L2 has 15C higher TjMAX than previous steppings. If it's true - all temp equations in CompuTronix's temps guide sticky should be corrected with this 15C difference. If so, max safe core temp should be 80C... Sounds not so cool, ehhh... but it's simple equation.

Quote:
Do you know of any canned benchmarks that would indicate likely performance when video editing?
http://www.mainconcept.com/site/index.php?id=4 You can download ANY of their standalone encoders as demo and check the speed of encoding. Use DV or uncompressed .avi as a source, DV .avi 1 GB at least, bigger=better.
!