AI7 maximum fsb

[mcoleg]

i wonder, do sinks on mosfets help ai7 reach higher fsb? they don't seem to run too hot but then again maybe i am not touching the right ones... any ideas?

 

[Spacepiston]

I put sinks on my mosfets and they worked quite well. I also lapped my northbridge heatsink and put some arctic 3 on it.

Now the NB stays under 35c @ full load, and the PWN stays under 75c.

The sinks on the mosfets get quite warm. But there 1 mosfet close to the p4 power connector that I could not get a sink on beause it was too tight. I have to rig a smaller one. Either way my stabiliity did not change, but my memory is holding me back.

 

[eva2000]

I'm surprised at your unbuffered Eva, 3100-3200 at DDR448. Hmmm, big dropoff due to SS sticks and 2-2-2-6? I haven't had my board up for awhile but I still had humongous bandwidth on the 800 strap, 5:4, 2-2-2-5 with dual 512 MB XMS3500, due to the SR-E-A-D-D enabling partial PAT. I got just over 3600 at 289, 5:4, 2-2-2-5.

Even at 300, 3:2, 2-2-2-5 or DDR400 I got 3150.

So far I've run 300+, 3:2, 2-2-2-5 with the Epox, Asus, and Abit boards and I get the highest 3DMark2001 scores with the Abit, due to the slightly higher unbuffered at around DDR400-408. The IS7 even beats the Asus P4C800-D, an 875P board, but then it beats the IC7 too by a tad. Can't run anything but A-A-A-D-D on the IC7 with PAT already enabled.

yeah 256MB SS sticks will reduce bandwidth alot by at least 350-400MB/s

 

[Clevor]

yeah 256MB SS sticks will reduce bandwidth alot by at least 350-400MB/s

Are you seeing that too with BH-5? I know with PC4000, yes, a huge difference of 300+ Sandra marks, e.g., 2900 versus 3200, same fsb, 1:1, but it didn't seem that bad when I ran 2x256 MB SanMax BH-6 versus 2x512 MB XMS3500.

I recently ran 283, 1:1 on the P4C800-D and I broke 3700 on the first mark. DS of course. That happened on the Epox 4PDA2+ too. I broke 3700 on the DFI LanParty once at 273, 1:1. I was surprised there.

 

[eva2000]

yeah even with BH-5 ram

here's 2x 512MB OCZ PC3500 Plat LE BH-5 @ 250mhz 2-2-2-7 at 3.18v vdimm on AI7

NB Strap 667fsb
GAT = AAADD

unbuffered

 

[eva2000]

How you manage 283mhz 1:1 with 3700mb/s unbuffered

2.8C @ 280fsb
2x 512MB OCZ PC4400 Performance series @ 280mhz 3-4-4-7 at 2.85v vdimm
P48C00-E 1016 bios

unbuffered

 

[Clevor]

Eva, were you 3D stable at 250, 1:1, 2-2-2-5?

I don't bother with screenies if it isn't even Memtest clean. All my unbuffered scores tested 3D stable. I could show some really high stuff otherwise.

With my better XMS3500 sticks, I broke 3700 at only 245, 1:1, 2-2-2-5 on the 4PDA2+, Rev. 2.0, with aggressive memory settings to turn on PAT. 3D stable.

How you manage 283mhz 1:1 with 3700mb/s unbuffered

I don't know, but I showed the screenie over at Xtreme. Really, it's not about doing 4000 unbuffered at maybe 270, 1:1, 2-2-2-5 horribly unstable. It's how high unbuffered you can get RELATIVE to the fsb, and is it stable??? Some guys are showing screenies of BH-5 at 265, 1:1, 2-2-2-5 or whatever but they are pumping 3.6 volts through the sticks. Stability? Hah! It is more of an accomplishment to do say, 250, 1:1, 2-2-2-5 at 3.1 volts, two 512 MB sticks, 3D stable. And can it even clear Sandra Cache and Memory?

When I ran 283, 3700+ on the Asus board I was at 2.5-4-4-7 but I had the VDIMM mod for extra volts. On the Epox I was running 2.9-3.0 volts or something. On the DFI, I was running 273, 1:1, 2.5-4-3-6.

I have not tested my new Vitesta sticks which will do maybe 5-8 mhz higher than the ram above!

 

[eva2000]

Clevor for you

here's 17 bios with 250mhz 2-2-2-7 at 3.18v vdimm

3dmark2001
128MB Gigacube 9800PRO Extreme @ stock 398/351x2 speed
Cat 4.3

compare url = http://service.futuremark.com/compare?2k1=7860709

note 17 bios increased my unbuffered scores @ 250mhz 2-2-2-7 at 3.18v to 3828/3801 -> http://www.ocforums.com/showthread.php?s=&threadid=299646

 

[batboy]

I've had my AI7 running at a stable 283 FSB with a 2.4C using the 5:4 ratio and tight timings (RAM with BH-5 chips of course). I was able to boot at 287 FSB with slightly relaxed timings (but haven't stability tested or benched at that speed yet). I haven't really tried pushing the vdimm past 2.8v yet to see what happens, just wanted to make sure it did at least as good as my IC7 before torturing it.

Update: recently I was able to reach 294 FSB with my 2.4C on the AI7 mobo. My BH-5 RAM was cruising along using the 5:4 ratio with 2,2,2,5 timings at 3.2v (DDR470). The rig feels very fast at this speed. I got 6.4k buffered Sandra memory bandwidth.

 

[Clevor]

Eva, where's the 3DMark2003 score? How about 3DMark2000 too?

One thing I like about Win98SE/ME is you can check 3DMark2000 too. I seen overclocks pass both 3DMark2001 and 2003, but fail 2000. The AMD boards will run 3DMark2000 too even in WinXP. Aquamark is also another toughie. I have one 3.2C at 3.92 gig that passes everything I can throw at it, SuperPI 16M too, but REFUSES to pass Aquamark! Even tried bumping the VCORE way up, no dice. Put in a different CPU and max that out, no problem with Aquamark.

Gotta try the -17 BIOS. I only ran my AI7 once about 5 months ago, heh. I did 300 fsb on air, stable in everything (you know the story). I was a newbie to 300 fsb then so I got more tricks up my sleeve now, plus some water.

I got my AI7 early from Newegg. Curious if the later boards are a bit different, a la the NF7-S. Some say they got better, some say worse.

 

[eva2000]

more 3d mark benchies for Clevor

@ 250mhz 2-2-2-7 at 3.18v vdimm 1:1
AI7 17 bios
CAT 4.3

128MB Gigacube 9800PRO Extreme @ 486/805 = 22,205

here's stock video card speed @ 398/702 = 6,425

 

[microfire]

eva, why do you use cycle time 7 ? I thought it was you that said over at the abit forums that 6 was the best setting.

 

[eva2000]

2-2-2-6 is faster than 2-2-2-7, but @ 250mhz 2-2-2-7 is more stable for me than 2-2-2-6 since only have 3.18v vdimm to work with With 3.3v vdimm i'm sure 250mhz 2-2-2-6 would be ideal

no use having fast if it ain't stable

 

[number 138]

I have a 2.4c that boots at 295 in my IC7, the AI7 will not even POST at 290 5:4 or 3:2. Also it is finicky at 3:2 at some lower FSb's as well. The best I have done is 285 5:4. 285 3:2 does not POST either. Weird.

 

[yelgnid]

How low of a FSB is it finicky at with 3:2? I've got a 2.4C 1.675v with OCZ PC3200 rev 2 at 2.7v at 269 3:2 fairly stable (lasts about 4 hours on Prime95). Been trying to get it up to 275. But can't get it prime95 stable even with CPU @1.75v/ Mem @ 3.1v. Would I have better luck at a higher FSB?

 

[Fletchs]

Think I max'ed out my AI7 at around 290 FSB...

anyone have link/guide for vagp/vdd AI7 mod ?

Fletchs

 

[microfire]

I just got a hold of a 2.4c in the last few days. No matter what I do, I cannot get it to boot at 275fsb.

This 2.4c must be good for more, the board does 266fsb(also memtest stable for 12hours) with ease, stock volts. So ive added more vcore when going for higher fsb just in case thats the problem, but nothing. That rules out the cpu imo.

This only leaves it to the board, I don't think this board can high fsb, about 272fsb is where it will or may not boot. So far ive relaxed timings, pumped vdimm, use 5:4 and 3:2 dividers, change the sets of memory around the slots, even run single channel.

At this point I think that I can only try my other AI7(3 days old) in the fastest rig. Not that I want to swap it around after I just done an install, and the vcore is more stable on the newer board.

I know this 2.4c has heaps leaft in it. Its old pack date 4/4/03, batch L309A586, SL6WF. Im sure this particular cpu I have is a good one.

Is this board at its maximum fsb ?

 

[Jimbob7]

another vote for the ai7 stock cooling


Sounds like your chip is max'd try using one stick of ram or some generic stuff in 3:2 that normally rules out the ram.

 

[microfire]

Tryed the BH-5 and posted 301fsb so far, didnt load windows.

CH-5 wont post 275 no matter what modules slots, single, or whatever.

My guess is the cpu is good for 3.6ghz with lots of voltage and wicked water setup.

 

[Jimbob7]

hooah , sounds good is the bh-5 yours? sorry i havent read the whole thread.