It looks like the 800 MHz FSB P4's WILL be OK for overclocking

[THUGSROOK]

Originally posted by: oldfart

I doub't it will be worth doing it if you do not keep the CPU and ram synced up

Nope. It will be very well worth it. I dont see why this is so hard to understand.

Maybe another way..

Lets call DC DDR 1:1 "100%" BW and compare mem numbers of different chipsets.

875/865/7205 DC DDR 1:1
100%

845PE 3:4
75%

845PE 4:5
66%

845PE 1:1
50%

875/865 5:4
80%

We've all seen benches of DC DDR amd 845PE running 4:5 ratio. Usually, there is a 3 - 6 % difference in perfomance between the two. Maybe a little more on some apps. Some have no difference at all. This is with 66% of the BW of a DC DDR board.

The 875/865 has 80% of a 1:1 DC DDR. If a drop of 33% mem BW amounts to a 3 - 6% drop in performance, how much will a drop of 20% cause? Very little.

What do you gain by dropping the mem ratio? Large increases in FSB and CPU speed. Unless you have some serious quality memory, you will hit a wall in short order when overclocking 800 MHz P4 "C" CPUs. Drop the mem ratio, lose a tiny bit of mem performance to gain very large increases in performnace due to higher FSB and CPU speeds.

ok,
who's gonna go 1st?

 

[Budman]

Originally posted by: THUGSROOK

Originally posted by: oldfart

I doub't it will be worth doing it if you do not keep the CPU and ram synced up

Nope. It will be very well worth it. I dont see why this is so hard to understand.

Maybe another way..

Lets call DC DDR 1:1 "100%" BW and compare mem numbers of different chipsets.

875/865/7205 DC DDR 1:1
100%

845PE 3:4
75%

845PE 4:5
66%

845PE 1:1
50%

875/865 5:4
80%

We've all seen benches of DC DDR amd 845PE running 4:5 ratio. Usually, there is a 3 - 6 % difference in perfomance between the two. Maybe a little more on some apps. Some have no difference at all. This is with 66% of the BW of a DC DDR board.

The 875/865 has 80% of a 1:1 DC DDR. If a drop of 33% mem BW amounts to a 3 - 6% drop in performance, how much will a drop of 20% cause? Very little.

What do you gain by dropping the mem ratio? Large increases in FSB and CPU speed. Unless you have some serious quality memory, you will hit a wall in short order when overclocking 800 MHz P4 "C" CPUs. Drop the mem ratio, lose a tiny bit of mem performance to gain very large increases in performnace due to higher FSB and CPU speeds.

ok,
who's gonna go 1st?

Send me a mobo and I will be glad to test stuff .

 

[oldfart]

I'll do it!

All I need is a Canterwood and a P4 "C" CPU. Somebody please send me one and I'll get started right away. I think the 2.6C will be the hot setup.

2.6C
13 x 275
3.57 GHz
DC DDR 440 (5:4)

 

[oldfart]

Wanted to update the BW comparison percentage numbers. Had them wrong the first time. Here are the correct numbers

875/865/7205 DC DDR 1:1
100%

845PE 3:4
66%

845PE 4:5
62%

845PE 1:1
50%

875/865 5:4
80%

875/865 3:2
66%

As you can see, running a 875/865 @ 5:4 has more BW than any SC DDR setup. Even dropping down to 3:2 has the mem BW of a 845PE running 3:4 and is higher than 845PE 4:5.

 

[THUGSROOK]

underclocking for overclocking eh?

isnt that like turning off turbo so you can run cas 222?

 

[oldfart]

No. You are missing the point entirely.

I cant fathom why this is so hard to understand.

Give up 1 - 2 % of memory performance for huge overclocking gains.

So, that must mean the 845E setups you were running all that time sucked? They had less mem BW than an 875 @ 5:4. Your own benches showed very little performance gain going from 845E 3:4 to DC DDR.

Look at the math. It works and makes perfect sense.

There are already people running 2.4Cs @ over 300 MHz FSB. Please explain how you would do that without using a mem ratio? Do you have a source of DDR600+ ram?

Do you think the tiny performace hit from slightly less mem BW is reason not to get a 100 MHz FSB, 50% overclock?

No, its not.

 

[jose]

Oldfart

Which memory would you buy in anticipation of building a p2.6C + canterwood mobo, if you were building a system soon ?

Regards,
Jose

 

[oldfart]

Tough question since we don't know how far a 2.6C will overclock yet. The easy answer is get the best memory you can afford! The Kingston HyperX has been getting good reviews.

Lets say in theory, the 2.6 can overclock to ~ 3.4 GHz
13 x 261 = 3.4 GHz

Memory requirements:
1:1 = DDR522 (Good Luck!)
5:4 = DDR417 (Easily done with good PC3200)
3:2 = DDR344 (The cheap stuff will work)

So, going by a theoretical 3.4 GHz overclock, some good PC3200 will do.

 

[Thor86]

Originally posted by: oldfart
I'll do it!

All I need is a Canterwood and a P4 "C" CPU. Somebody please send me one and I'll get started right away. I think the 2.6C will be the hot setup.

2.6C
13 x 275
3.57 GHz
DC DDR 440 (5:4)

This would be shhhwweet.

Here's to hoping my Corsairs can reach those memory speeds even with relaxed timings.

I'll have to agree with you, the higher the fsb/cpu speeds, the more raw performance you get. I noticed this playing Rallisport Challenge, and the difference is night and day between 176 and 177 on my GB setup.

 

[JDLT1]

Where can you buy one of those 800fsb chips?

 

[oldfart]

Zroc over @ [ H ] forums has tested this out and confirms what I was saying aout the 5:4 and 3:2 mem ratios:

Wow. I found this one really interesting...and breaking it out into percentages really made it hit home. The first percentage row shows the individual differences on each benchmark when dropping from 1:1 to 5:4. The Composite figure is simply those seven percentages divided by seven.
That's right, folks...the drop from 1:1 to 5:4 is an average of 2%. Overclockers everywhere should be smiling about right now.
In the third score column, I dropped to 3:2, and the following percentage columns are the percentage drop of 3:2 from 1:1 and 5:4, respectively.
3:2 is a 4.91% drop from 1:1. And 3:2 is a 2.85% drop from 5:4.
This whole chart actually shocked me...but in a good (GREAT) way. I was expecting MUCH greater hits. I'm SO glad to be wrong on that one
Good, good stuff...let's roll on...

Like I said in my earlier posts in this thread, maybe lose 2 - 6 % perfomance from the mem ratio, but gain a whole lot more from higher FSB and CPU speed. Its a small tradeoff.
Read more here.....

 

[ErikaeanLogic]

Thanks for the enlightenment, oldfart, that is really good news!

(. . .I just wish it were good enough for me to want to upgrade from Granite Bay

)

 

[jhites]

Originally posted by: oldfart
Like I said in my earlier posts in this thread, maybe lose 2 - 6 % perfomance from the mem ratio, but gain a whole lot more from higher FSB and CPU speed. Its a small tradeoff.
Read more here.....

Thanks for the heads up. I was curious so I
checked some numbers on my setup.
It did come out about the same percentages, also.

I still want the 1:1 DDR540 though.

 

[oldfart]

I still want the 1:1 DDR540 though.

Yeah, that would be nice. Kind of hard to find ram that will do that speed right now

 

[jhites]

Originally posted by: oldfart

I still want the 1:1 DDR540 though.

Yeah, that would be nice. Kind of hard to find ram that will do that speed right now

HyperX PC3500 out for UPS delivery right now.
Did order some TwinMos PC3700 this morning also.
4.3ns chip instead of the 5ns. Should be interesting to
see how they compare with timings and speed.

 

[THUGSROOK]

Originally posted by: ErikaeanLogic
Thanks for the enlightenment, oldfart, that is really good news!

(. . .I just wish it were good enough for me to want to upgrade from Granite Bay

)

1mb cache chips will make it worth wild

 

[amoralist]

don't forget that not all applications/programs are CPU bound, many are IO/MEM bound. CPU bound means that even if you boosted the IO/MEM bandwidth the comp wouldn't be able to execute the program any faster. IO/MEM bound means that even if you boosted the CPU bandwidth, the comp wouldn't be able to execute the program any faster. things like video encoding, code compilation are CPU bound. things like OS bootup, program startup, file transfer are IO/MEM bound. you don't want to lose the %2 mem performance if you never run CPU bound programs. so i mean it's a little zealous to say that it's dumb to care about the mem ratio. just like any good computer architect knows, performance means eliminating the bottleneck, and where the bottleneck is depends on how the machine is used, and that depends on the user. of course, since i do video encoding and code compilation than i'd prefer a faster fsb/cpu. but if it were my mom's computer, then i'd give here a 1:1 ratio. i'm pretty sure she never gets here CPU above 50% utilization.

 

[raddish]

It woudl be great to see a 4:3 divider.

That would allow a 2.4C @ 3.2 with 267 FSB while your DDR400 RAM runs exactly on spec.

 

[Acanthus]

testing with SiS655 and P4 2.4B@~2.9ghz

results soon.

Edit: what benchies you want? 3dmark? sandra? UT2003? Warcraft3? Planetside Beta?

Baseline Sandra: 2.4B @ 2.92ghz, 162mhz FSB, Memory 1:1 2.5/3/3/6/2T

CPU Arithmatic: 7436 MIPS, 3827 SSE2
CPU MultiMedia: 11558 Integer, 14579 Float
Memory Bandwidth: 3783 Int Buff, 3786 Float Buff
Cache and Memory: 8kb - 26030 MB/s, 1MB 2061 MB/s, 256MB 2100 MB/s

Baseline Warcraft 3 (using fraps to measure FPS, on a custom game, in Ice Crown, with 11 computer opponents on normal)
61 FPS

Baseline Warcraft 3 (using fraps to measure FPS, on a custom game, in Ice Crown, with 11 computer opponents on normal) 4XS AA, 8X Aniso
31 FPS

Baseline Planetside Beta (using the offline training session, 1st training, fps while facing default position)
46 FPS

Baseline Planetside Beta (using the offline training session, 1st training, fps while facing default position) 4XS AA, 8x Aniso
26 FPS

All settings are maxed on all games, and run at 1024x768x32, quincux AA, 8x aniso.

Rig:
P4 2.4B @ 2.92ghz 162fsb
1024MB Kingston Valueram 2700 (slowest timings its trash) Run in dual channel
Geforce 4 Ti4200 128MB run at default clocks
Western Digital 1200JB
Antec 430w Truepower PSU with tweaked pots
Using integrated sound and lan
Direct X 9.0a
43.45 detonators
latest sound drivers

 

[Acanthus]

Asymetric Sandra: 2.4B @ 2.92ghz, 162mhz FSB, Memory 6:5 (134mhz) 2.5/3/3/6/2T

CPU Arithmatic: 7487 MIPS, 3830 iSSE2
CPU MultiMedia: 11564 Integer, 14564 Float
Memory Bandwidth: 3278 Integer Buff, 3278 Float Buff
Cache and Memory: 8kb - 26034 MB/s, 1MB 1816 MB/s, 256MB 1842 MB/s

Asymetric Warcraft 3 (using fraps to measure FPS, on a custom game, in Ice Crown, with 11 computer opponents on normal)
61 FPS (0% change)

Asymetric Warcraft 3 (using fraps to measure FPS, on a custom game, in Ice Crown, with 11 computer opponents on normal) 4XS AA, 8x Aniso
31 FPS (0% change)

Asymetric Planetside Beta (using the offline training session, 1st training, fps while facing default position)
46 FPS (0% change)

Asymetric Planetside Beta (using the offline training session, 1st training, fps while facing default position) 4XS AA, 8x Aniso
25 FPS (2% performance loss)

Take the synthetic for what you will, the games i play arent effected by asymetric clocks at all (even with DDR2100 speeds!)
My conclusion: Dual DDR 400 cas2 will be MORE THAN ENOUGH to feed a 1GHZ fsb.

 

[claytoma]

I received my 2.6c P4 yesterday and over clocked it with ease to 3.25Ghz with the FSB at 250mhz (1000mhz effective) with the mem ratio at 5:4 (200 mhz). I got 18229 on 3DMark 2001SE with an overclocked Radeon 9700pro (360/670). The new CPUs run amazingly cool; it never gets over 48c on my IC7 where as my 533 2.66G P4 got up to 59c at 3.25Ghz with the exact same setup (these are loaded temps). Oh, I'm using the SLK900U for cooling.

 

[oldfart]

Nice overclock. I'm curious about your temps on the IC7. Its well known that temp readings are fairly fubar on this board. What do you get for an idle temp? What program are you using to read it?