[email protected]: 8x450 or 9x400?

With my ram I can run 4-3-3-8 timings at 400 at 2.0V, but at 450 I have to run 2.1V and CAS 5 (5-4-4-12) (though I can pull a 5-3-3-8 but doesn't really change much in performance).

8x450 is what I'm running right now since it stresses the ram worse than the 400 and tight timings are due to the raised mhz and voltage.

Which is the better option? Have any comparisons you are interested in?

I'll try testing out various things over the next week to see which shows better numbers, but so far I'm stable @ 3.6 >24 hours worth of prime at various ambient temps in a closed case ranging from 23C to 29C room ambients...

Things to compare...
3DMark06
Doom 3
FEAR (1600x1200)
SiSandra
SuperPi
HL2 Stress Test (CS:S)
HL2: Lost Coast HDR Stress Test

I think it has been said many times. There is no real difference in performance! I will do 9*400

i think 4x900 is best to go as it will not stress the northbridge as much.

I would like to see the board capable of going to 900MHz FSB!!! at the lower FSB I think it would be better though.

If you can run a stable 450x8 without blowing up your house I'll be very impressed.

I would think that the quad-cores would like as high an FSB as possible. They use the FSB as their only way of communication between the two pairs of cores. I could be wrong, but that's what I've gathered. I am running a similar motherboard to you, P5K Premium, and have the FSB at 476 (1904QDR) with only a slight bump in FSB voltage. All other settings stock.

The FSB doesn't limit their performance.

Go 400x9 and set 4:5 for 500MHz on the RAM if possible. No reason to add extra stress to the system for literally no benefit, and I mean literally no benefit at all - not even in benchmarks.

well I'm going to start the round of testing with the 9x400 first

I had it stable for 10+ hours at 9x450 so I'll gather I'm fine there, though some anomalies were reported by asus probe during that span, first I had a few spikes in temps to 65C, second I had my 12V rail drop inexplicably to 9V then back to normal.

I've also set ALL the voltages to their minimum (aside from CPU and RAM). let the testing begin, results will trickle in throughout the week (also playing TF2 and ET:QW not helping heh)

results not looking good for 9x400...

8x450
---------- RUN1PASS1.LOG
encoded 1749 frames, 185.95 fps, 1850.94 kb/s

---------- RUN2PASS1.LOG
encoded 1749 frames, 185.63 fps, 1850.94 kb/s

---------- RUN3PASS1.LOG
encoded 1749 frames, 184.71 fps, 1850.94 kb/s

---------- RUN4PASS1.LOG
encoded 1749 frames, 185.31 fps, 1850.94 kb/s

---------- RUN5PASS1.LOG
encoded 1749 frames, 185.02 fps, 1850.94 kb/s

---------- RUN1PASS2.LOG
encoded 1749 frames, 50.34 fps, 1829.44 kb/s

---------- RUN2PASS2.LOG
encoded 1749 frames, 50.34 fps, 1829.02 kb/s

---------- RUN3PASS2.LOG
encoded 1749 frames, 48.24 fps, 1829.16 kb/s

---------- RUN4PASS2.LOG
encoded 1749 frames, 50.32 fps, 1829.07 kb/s

---------- RUN5PASS2.LOG
encoded 1749 frames, 49.81 fps, 1829.19 kb/s

9x400
---------- RUN1PASS1.LOG
encoded 1749 frames, 171.88 fps, 1850.94 kb/s

---------- RUN2PASS1.LOG
encoded 1749 frames, 174.32 fps, 1850.94 kb/s

---------- RUN3PASS1.LOG
encoded 1749 frames, 173.79 fps, 1850.94 kb/s

---------- RUN4PASS1.LOG
encoded 1749 frames, 176.36 fps, 1850.94 kb/s

---------- RUN5PASS1.LOG
encoded 1749 frames, 174.32 fps, 1850.94 kb/s

---------- RUN1PASS2.LOG
encoded 1749 frames, 48.29 fps, 1829.15 kb/s

---------- RUN2PASS2.LOG
encoded 1749 frames, 48.35 fps, 1829.09 kb/s

---------- RUN3PASS2.LOG
encoded 1749 frames, 48.68 fps, 1829.54 kb/s

---------- RUN4PASS2.LOG
encoded 1749 frames, 47.38 fps, 1829.03 kb/s

---------- RUN5PASS2.LOG
encoded 1749 frames, 48.45 fps, 1829.48 kb/s

I am curious about your SiSoft memory bandwidth benchmarks at different FSB/RAM speed settings.

I would think that encoding and pure memory bandwidth tests would have the biggest differences with higher FSB settings. I would be curious to see both of these "best case" scenarios to see what the biggest possible difference would be.

Congrats on the OC, too!

Slade, you on air?

I'd also like to try 9x400 w/ Ultra 120 Extreme cooling.

Currently stable w/ 9x333 using 1:1 on RAM.

This is on air... man my room is toasty... I'll do the sisandra comparison tonight

here's a shot of the 8x450, it was running for 4 hours and I left it to run overnight to 12 hrs+



here's a shot of 9x400, today's stability test has prime holding well past the 15 hr mark..

SiSoftware Sandra Results

SiSoftware Sandra

Processor Arithmetic
9x400
Dhrystone ALU 66612 MIPS
Whetstone iSSE3 43258 MFLOPS

8x450
Dhrystone ALU 66619 MIPS
Whetstone iSSE3 43303 MFLOPS


Processor Multi-Media
9x400
Multi-Media Int x8 iSSE3 390726 iit/s
Multi-Media Float x8 ISSE2 304975 fit/s

8x450
Multi-Media Int x8 iSSE3 390867 iit/s
Multi-Media Float x8 ISSE2 305465 fit/s

Memory Bandwidth
9x400
Int Buff'd iSSE2 Memory Bandwidth 6862 MB/s
Float Buff'd iSSE2 Memory Bandwidth 6862 MB/s

8x450
Int Buff'd iSSE2 Memory Bandwidth 7641 MB/s
Float Buff'd iSSE2 Memory Bandwidth 7644 MB/s

Memory Latency
9x400
Memory (Random Access) Latency 82 ns
Speed Factor 97.00

8x450
Memory (Random Access) Latency 75 ns
Speed Factor 88.60

My thoughts on these benchmarks.

FPU performance increased with the the memory bandwidth, in theory games that are more FPU dependent will show an improvement in overall score. Everything else should stay relatively the same.

As for gaming, there's no point to testing low res to do fsb comparisons, you won't play at those settings and by nerfing the detail it doesn't accurately show if a certain part of the cpu may gain benefit hint FPU calc...

So surprise surprise, FEAR and HL2: Lost coast showed a marked improvement in performance, a gain of 10fps in a few runs. I suspect DX10 games and future games that add effects like HDR will show a marked improvement. Even video encoding showed a deal more fps as well.

My observation so far is that there is a difference in performance between FSB speeds, but you have to test things or use programs that rely on RAM access.

Doom 3
1920x1200 High 2xAA

9x400 8x450
129.8 129.9
130.2 130.3
130.2 130.2

FEAR
1600x1200 High Soft Shadows Off AS4x FSAA Off

9x400 8x450
Min 54 60
AVg 72 81
Max 90 103

Min 53 59
AVg 71 81
Max 89 103

Min 53 59
AVg 71 80
Max 89 102

3D Mark 06
9x400 8x450
12624 12654
12665 12622
12614 12645

Counter Strike Source: Video Stress Test
1920x1200 Max Settings No AA
9x400 8x450
279.34 279.04
275.83 280.42
279.44 277.69

Half Life 2: Lost Coast
1920x1200 Max Settings No AA
9x400 8x450
139.01 148.15
137.67 147.55
138.65 147.05

Slade said:

Memory Bandwidth
9x400
Int Buff'd iSSE2 Memory Bandwidth 6862 MB/s
Float Buff'd iSSE2 Memory Bandwidth 6862 MB/s

8x450
Int Buff'd iSSE2 Memory Bandwidth 7641 MB/s
Float Buff'd iSSE2 Memory Bandwidth 7644 MB/s

Click to expand...

Now, if I remember correctly, you are scaling the actual memory speed proportionally with the FSB when you use the higher FSB. I notice here you get ~90% of the difference you would expect by just multiplying the 400FSB score by 450 then dividing by 400 to get the perfectly proportional amount for a theoretical 450FSB. Not bad. Not a bad result at all.

I have found however, with my e6600, that even maintaining the memory clock at 800 and increasing only the FSB, both from 1066.7 to 1333.3 and from 1333.3 to 1466.7, gives me about that same 90% of the purely proportional increase in score. I found that odd. Again, all three of those FSB speeds were running with the ram at 800Mhz. I went from 5341 to 6513 to 7122, just by changing bus speeds from 1066.7 to 1333.3 to 1466.7. Really odd. Obviously my CPU scores went up as you would expect.

Especially since going from 800 mem to 1066.7 mem at the 1466.7 FSB only got me to 7604. (All these numbers are for iSSE2 Integer, but the proportions are the same with the iSSE2 Float numbers). That is less than a 7% increase in score for a 33% increase in actual memory clock, only 20% of the purely proportional difference. Also note that the mem timings for all the above tests were 5-5-5-15. At 800 or at 1066.7. Really, really odd.

---