And Without Further Delay.... 2.4C IC7-G PC3500 Winbond CH-5

[Kunal]

Hey guys - finally picked up my RAM from the parcel depot Friday morning, although havnt had a chance to play with the system until today! Had my final exam on Friday, so I've spent the entire weekend out celebrating with friends :) What better way to nurse a hangover than a bit of OC'in while watchin a Back to the Future marathon :p

Anyhow, it's all up and running now, just wanna start with some benchies and info of the system at stock.

Intel Pentium 4 2.40Ghz 800-HT (SL6WF L310)
Abit IC7-G Bios 1.3
2x 512mb PC3500 Winbond CH-5 Timings 2-5-2-2
Maxtor 120Gb SATA
Antec TrueBlue480
SLK900U w/ 80mm Pabst (8412NG 40.6cfm, 3100RPM, 32dBA)

CPU - 1.525v
RAM - 2.6v
Performance Mode - Ultra (guessing I should set it to this over Turbo and Fast)

http://www.developerx.net/benchmarks/cpuz-1.gif

http://www.developerx.net/benchmarks/sandra-cpu1.gif

http://www.developerx.net/benchmarks/sandra-mem1.gif

http://www.developerx.net/benchmarks/superpi-1.gif

Flashed from 1.0 straight to 1.3 - although I don't think I ever had the temp problem - reported temps were 37degC idle and 45degC load, before and after the Bios flash.

CPU-Z reports RAS to CAS as 3, while it's set to 2 - should I believe CPU-Z or the Bios? I've run Prime95 for a coupla hours at 2-5-2-2 appears to like it.

Also, isn't the CPU bein undervolted quite a lot? Set to default 1.525, although reading in CPU-Z as 1.504v and 1.50 in Winbond Hardware Doc.

Do you guys n gals think those sandra mem and cpu score look right for the setup and config?

[theotherphil]

The Sandra RAM looks a little low. Have you installed the chipset drivers? I would expect around 5.3GB/s at that fsb / timings.

[Kunal]

I thought the RAM scores we're a bit low as well - I've installed the Intel(R) Chipset Software Installation Utility v5.00.1012, that came on the Abit CD, first thing I did when Win2K booted for the first time. Installed everything on the CD.

Anyone have any idea's as to what's holding my RAM scores back?

(Just updating win2k, although I don't see that making a difference ;) )

[theotherphil]

Ack, ignore me....these Canterwood boards are very confusing in how they get their bandwidth. For example, at 200/200 I get similar mem benches to yourself (4.6GB/s - I just tested). Now at 250fsb with the 5:4 divider (ram at 200fsb) you would think that you'd get the same performance, no?

Wrong...at 250/200 with the exact same timings as 200/200, I get:

http://homepage.ntlworld.com/phil.harling/images/P4/p4@3gmem.gif

Where the hell did this extra bandwidth come from when the timings and RAM speed are the same? :confused:. Never mind, I don't really care but it rocks!

[texasfit]

Sandra benches are affected by the cpu speed - also dual channel bandwidth is affected.

Kunal
You are about right for the 200/200 mem scores
I would trust the cpu_z readings.
Set the vcore in the bios to where it shows default 1.525v idle in cpu_z and then start to oc the fsb.
Set the mem ratio to 3:2 and lower the timings
Increase the fsb to see what you get at default. It should be somewhere around 270+fsb.

[theotherphil]

And to confuse even more, at 300/200 I can't even post with the timings at 2, 5, 2, 2. I have to drop to 2.5, 7, 3, 2 and still beat the 250/200 mem bench:

http://homepage.ntlworld.com/phil.harling/images/P4/300mem.gif

[theotherphil]

Originally posted by texasfit
Sandra benches are affected by the cpu speed - also dual channel bandwidth is affected.


Now that shows what a good benchmark Sandra really is ;)

[TC]

Your scores at 200 are low for some reason. My P4G8X was getting close to that at 150FSB. My P4C800 gets 5000 at 200fsb.

[xtrmeocr]

the p4 is the bottle neck thats why. it cant deliver the full 6.4gb/sec which is advertizes

[Kunal]

All very strange heh :)

Tried to push it at 2-5-2-2, although didn't get very far. Didn't expect to though.

SPD for the ram is 2.5-7-3-3, although it's advertised as, and used by many as 2-7-3-3.

I've set it to the advertised default of 2-7-3-3 (got loadsa strange Bios erros between 2-5-2-2 and 2-7-3-3). Still at 2.6v, 217mhz/434mhz.

Gonna see how high I can push it on the 1:1 divider, with mem voltage set to default.

BTW, I've left the NB Strap in the Bios as "By CPU", should I change it to PSB800? Or just leave it.

[texasfit]

Run PCMark2002 IMHO you will get a better comparison

[AZN]

Originally posted by theotherphil
And to confuse even more, at 300/200 I can't even post with the timings at 2, 5, 2, 2. I have to drop to 2.5, 7, 3, 2 and still beat the 250/200 mem bench:

http://homepage.ntlworld.com/phil.harling/images/P4/300mem.gif

tell me u are not serious. even thou the ram speed is still at 200mhz the cpu speed at 250fsb is 600mhz more and at 300fsb its 600mhz more then that or 1200mhz more then stock so the increase in cpu speed will affect the overall memory bandwidth. comon are u joking with us?

[Rio71]

Originally posted by AZN


the increase in cpu speed will affect the overall memory bandwidth.
no, no!
i have absolutely the same sandra mem-bench with 845pe.

p4m @ 2400 4:5 --> 500ddr
or
2.4/533 @ 3380 3:4 --> 500ddr

memtest86 3.0 shows me the same, independence of cpu-speed.

[texasfit]

Originally posted by AZN


tell me u are not serious. even thou the ram speed is still at 200mhz the cpu speed at 250fsb is 600mhz more and at 300fsb its 600mhz more then that or 1200mhz more then stock so the increase in cpu speed will affect the overall memory bandwidth. comon are u joking with us?

u are joking, right? :eek:

LOL

[CrazyP]

what he is saying is the more bandwidth the p4 puts out, the more the memory takes in, up to a certain point. the memory wont be maxed out in bandwidth unless the p4 can feed it. thats why amd numbers are lower all the time in these type of tests

[AZN]

Originally posted by Rio71

no, no!
i have absolutely the same sandra mem-bench with 845pe.

p4m @ 2400 4:5 --> 500ddr
or
2.4/533 @ 3380 3:4 --> 500ddr

memtest86 3.0 shows me the same, independence of cpu-speed.
the mobile chip will have a much higher bandwidth (fsb) then the B chip so it should be about the same range in memory bandwidth
the p4m is at 200fsb right? and the 2.4b is at 187fsb so

p4m @ 2400 4:5 --> 500ddr
12x200fsb 4:5=ddr500

2.4/533 @ 3380 3:4 --> 500ddr

18x187fsb 3:4=ddr500

so even thou the cpu speed is different u have the same ddr speed cuzz of the differrent ratio u used and should be gettign the same memory scores.

[texasfit]

Originally posted by AZN

the mobile chip will have a much higher bandwidth (fsb) then the B chip so it should be about the same range in memory bandwidth
the p4m is at 200fsb right? and the 2.4b is at 187fsb so

p4m @ 2400 4:5 --> 500ddr
12x200fsb 4:5=ddr500

2.4/533 @ 3380 3:4 --> 500ddr

18x187fsb 3:4=ddr500

so even thou the cpu speed is different u have the same ddr speed cuzz of the differrent ratio u used and should be gettign the same memory scores.

Actually, I thought you were joking.
In Sandra the scores will be different due to the overall memory usage that the program tests even though the ram is running at the same speed ie: DDR400.
Couple of examples for you:
In that c#@p Sandra Program.
The bus speed is 400mhz more at 250 and does have impact on the overall memory bandwidth for the system.

Here are 2 different Mem Brands using the same timings:
****Corsiar TwinX3200LL PC3200****
P4C800-2.60 = . . . 4654 / 4659 . . DDR400(1:1) . 200fsb . . . . 2600Mhz
P4C800-2.60 = . . . 5421 / 5387 . . DDR400(5:4) . 250fsb . . . . 3250Mhz
****Kingston HyperX KH3500 PC3500****
P4C800-2.60 = . . . 4662 / 4639 . . DDR400(1:1) . 200fsb . . . . 2600Mhz
P4C800-2.60 = . . . 5421 / 5402 . . DDR400(5:4) . 250fsb . . . . 3250Mhz

[Rio71]

Originally posted by AZN

the mobile chip will have a much higher bandwidth (fsb) then the B chip so it should be about the same range in memory bandwidth
the p4m is at 200fsb right? and the 2.4b is at 187fsb so

p4m @ 2400 4:5 --> 500ddr
12x200fsb 4:5=ddr500

2.4/533 @ 3380 3:4 --> 500ddr

18x187fsb 3:4=ddr500

so even thou the cpu speed is different u have the same ddr speed cuzz of the differrent ratio u used and should be gettign the same memory scores.
yes, correctly.
but theotherphil has different memory bandwidth at even 200ddr. (similar to my 500ddr)
only changing with cpu-speed.
this can't be.
same memory-fsb = same memory bandwidth

[AZN]

i took both blue and red pills so dont listen to me i really dont know what im talking about most the time :)

[Rio71]

oh damn, german is much easy'er. :beer:

[markodude]

Hey Rio71 you getting an Abit IS7-G or something Dual Channel for that mobile chip?
I have an i865 Intel board waiting on me (genuine Intel ;-)) but I dont know if the P4M 2.2 will work in it, never tried in a real Intel board, hoping it will detect it as a 2.4c (12x multi)
The IS7-G and 2x Twinmos PC3700 with the mobile chip is my next move, then get a real C chip and pop it in later...
sorry for the thread hijack people :D

[theotherphil]

Originally posted by Rio71

yes, correctly.
but theotherphil has different memory bandwidth at even 200ddr. (similar to my 500ddr)
only changing with cpu-speed.
this can't be.
same memory-fsb = same memory bandwidth


The results I get are:

200fsb / 200mem (1:1) - ~4500/4500 (2,5,2,2)
250fsb / 200mem (5:4) - ~5300/5300 (2,5,2,2)
300fsb / 200mem (3:2) - ~5650/5650 (2.5,7,3,2) <- relaxed timings...won't post at 2,5,2,2!


Don't forget though that the 875P chipset has PAT that supposedly adjusts latencies on the fly etc for better performance. It just makes manually tweaking that more funky :eek:

[Kunal]

Ergh, had a problem with windows last night, some files got corrupted, havn't had time to format until today...

OK, bb in a little while, aiming for 230FSB on 1:1 - I've left performance mode on disabled, as it seems to give me a lot of bios errors no mater what the RAM timings or voltage are. Hopefully I can figure out what the problem is here and re-activate it later.

[Kunal]

OK, 225/450 @ 2.5-7-3-3 2.6v I can get into Windows, but it's not Prime95 / SuperPi stable.

Winbond Hardware Doctor reports memory as 2.59v and vcore flickering between 1.50 and 1.52 (still undervoltin)

Max I can set RAM voltage to is 2.8v (2.78v actual). I'm gonna reboot at 2.7v and see if Prime runs.

[Kunal]

OK, 2.7v DDR. 2.5-7-3-3, things appear to be SuperPi and Prime stable - although Photoshops acting a bit strange... keeps freezing up for 20seconds then coming back to life....

I was watching the voltages while Prime was running.... DDR flickers between 2.66 and 2.70, but here's the wierd thing.... Now at 2.7Ghz CPU, the vcore is reporting LOWER, and is flickering between 1.49-1.50 in Winbond Hardware Doctor.... CPU-Z reports it as 1.488v!!!!

http://www.developerx.net/benchmarks/sandra-mem2.gif

Mem scores are up at least lol.

I'm gonna push it slowly 1mhz at a time, before I'm forced to go 2.8v (Then start to lower timings to 3-7-3-3)

I might bump the vcore up a notch as that just doesnt seem healthy to me.... 1.488 is v low

[theotherphil]

The vcore on my IC7G reads a little low also. My vDIMM on the other hand reads 2.83 when set at 2.8v.

[Kunal]

Well here it is....

http://www.developerx.net/benchmarks/sandra-mem3.gif

This is as high as I can go :( I want 250! :o :)

Damn voltage restrictions :(

Time to hit the 5:4 divider!

[Kunaak]

wow-P4 2.4 C is kinda slow on the super Pi time.
I just gave a shot at my XP 2100 at default speed. with the same ram timings you show there, 2-3-3-5 and I got 1 minute 3 seconds. so I was 1 second slower.
I thought they were faster then that.

[theotherphil]

Try the Prime 95 (ftp://mersenne.org/gimps/p95v234.zip) Benchmark and then you'll see how fast the P4 can be in comparison to the AMD:


Intel(R) Pentium(R) 4 CPU 2.40GHz
CPU speed: 3600.94 MHz
CPU features: RDTSC, CMOV, PREFETCH, MMX, SSE, SSE2
L1 cache size: 8 KB
L2 cache size: 512 KB
L1 cache line size: 64 bytes
L2 cache line size: 128 bytes
TLBS: 64
Prime95 version 23.4, RdtscTiming=1
Best time for 384K FFT length: 9.857 ms.
Best time for 448K FFT length: 11.714 ms.
Best time for 512K FFT length: 13.299 ms.
Best time for 640K FFT length: 16.017 ms.
Best time for 768K FFT length: 19.491 ms.
Best time for 896K FFT length: 23.101 ms.
Best time for 1024K FFT length: 25.903 ms.
Best time for 1280K FFT length: 33.906 ms.
Best time for 1536K FFT length: 41.843 ms.
Best time for 1792K FFT length: 49.760 ms.
Best time for 2048K FFT length: 56.169 ms.

[Kunaak]

theres a benchmark for prime 95...

geez a year using it and I never even looked at that.

[theotherphil]

Originally posted by Kunaak
theres a benchmark for prime 95...

geez a year using it and I never even looked at that.


LOL...definitely a hardcore DC'er :D