Any way to disable AXP Barton Speed Throttling?

[mordrid52]

I just bought a Mobile 2400+ Barton processor that I put in my Shuttle SN45 XPC. It runs fine at 200x12 at 1.8v. However, when it goes under heavy load (like running a distributed computing project or whatnot), the CPU throttling kicks in and scales the speed back 200-300 MHz. The CPU temperature never goes above 54-55 C, so it's not like the CPU is overheating. I've even had the throttling kick in when the temp was only 52 C.

The SN45 bios does not have any options to adjust the throttling like some nforce 2 boards have. Is there any way to modify the chip or the bios to disable the cpu speed throttling? Or am I just screwed? Thanks.

 

[sandorski]

Originally posted by: mordrid52
I just bought a Mobile 2400+ Barton processor that I put in my Shuttle SN45 XPC. It runs fine at 200x12 at 1.8v. However, when it goes under heavy load (like running a distributed computing project or whatnot), the CPU throttling kicks in and scales the speed back 200-300 MHz. The CPU temperature never goes above 54-55 C, so it's not like the CPU is overheating. I've even had the throttling kick in when the temp was only 52 C.

The SN45 bios does not have any options to adjust the throttling like some nforce 2 boards have. Is there any way to modify the chip or the bios to disable the cpu speed throttling? Or am I just screwed? Thanks.

Check out Power Management in Windows. Dunno for sure if there's anything there, but likely there is.

 

[loafbred]

I've been looking for an answer to the performance dropoffs I get with the same cpu (on Abit NF7) at certain multipliers and bus settings. Multis 10 and 11 work fine, as long as fsb doesn't go over 211. Other multipliers I've tried cause huge performance drops, regardless of bus speed. Using CPUMark99, I get score of 216 at 200*11. At 200*12, it scores ~180. 3DMark2001 goes from ~18000 at 200*11 to ~13000 at 200*12. I'm planning to take time on Saturday to use pin wires to lock the cpu at 12 multi and 166 bus to see if it has any impact. If it fixes the problem, I'll start a thread about it. If not, I'll try to let you know in this thread.

 

[dnuggett]

My DFI Lanparty NFII Ultra B has an option in the BIOS to turn this on/off. Check there.....

 

[myocardia]

Why not just cool it better?

 

[loafbred]

Originally posted by: myocardia
Why not just cool it better?

Speaking for myself, it's not temp related... I can set mine to 12*180 and CPUMark99 takes a nosedive to 175. 10*210 scores ~205 consistently. 10*212 drops score to ~175. My temps at 11*200 are 38-40*C idle and ~46* running Prime95. 12*200 w/1.90v core runs at 51*C in Prime95. I formatted another drive and saw no change with a clean install of WinXP. CPU interface and memory timings also have no effect. I've also tried D10 and D21 BIOSes, and tictac's D10 beta 4. I'm using standard D18 BIOS now.

I was planning to buy watercooling, but if I can't get this sorted out, I'll wait. I'm happy with 200*11 at 1.65v core, but I'd like to get rid of some fan noise, and would take the water cooling plunge if I coud get a performance increase at 2.3 to 2.4 GHz.

 

[WebDude]

Originally posted by: loafbred
I've been looking for an answer to the performance dropoffs I get with the same cpu (on Abit NF7) at certain multipliers and bus settings. Multis 10 and 11 work fine, as long as fsb doesn't go over 211. Other multipliers I've tried cause huge performance drops, regardless of bus speed. Using CPUMark99, I get score of 216 at 200*11. At 200*12, it scores ~180. 3DMark2001 goes from ~18000 at 200*11 to ~13000 at 200*12. I'm planning to take time on Saturday to use pin wires to lock the cpu at 12 multi and 166 bus to see if it has any impact. If it fixes the problem, I'll start a thread about it. If not, I'll try to let you know in this thread.

Did you turn off "cpu thermal-throttling" in the bios?

 

[loafbred]

Yep, I turned off thermal throttling and tried about fifty combinations of anything which I thought could be remotely related. I suspect that the PowerNow function is causing it, and hopefully locking the multiplier will stop it.

 

[loafbred]

I tried shorting pins according to this chart, and it didn't lock the multiplier, nor did it make it boot at 12 multi. It booted at 100*6. It did, as the instructions stated, prevent a multiplier lower than the one configured (12). It didn't make any change to the drops in performance. I triple checked the wire placements, and they were still in place when I took the cpu out again to remove them.

 

[loafbred]

Here's an article showing how to enable/disable the "mobile" functions by an L5 bridge. Unfortunately, it will have to be cut to disable it. I'll have to think this one over a bit.

[L=The Article]http://www.cpuheat.wz.cz/html/AXP_multiplier/AXP_Multiplier.htm[?L]

 

[loafbred]

Well, curiosity got the better of me, and I cut the L5 bridge. Now I have max 166 bus. Some multipliers no longer work. 10, 15 and 18 do work, but I didn't try many. 11, 12, 13 and 14 don't work. What I have to do now is to change all the bridges to match a specific cpu, and I'm going to make it either a 2500+ or 3200+ (input appreciated). Until I get that done, I have a 2200+ I can live with. This is a lot of work for an extra 200 MHz. If anybody else gets excited enough about this to change their bridges, here is a good guide.

 

[loafbred]

I sat down three hours ago with conductive paint and Xacto knife, prepared to reconnect a few bridges to make it a 3200+. Using a stronger magnifying lens than I used when cutting the L5 bridge, I noticed that I had cut into two traces right next to that bridge. I patched them with the paint and thought it best to give it a try before going further. It behaves exactly as it did before I cut the L5 bridge... in other words, I'm running at 200*11 with 1.65v. These new grid packages are a whole new ballgame when it comes to connecting the laser-cut bridges. The little gold dots are no longer connected to the bridge (according to the author of the guide I linked to above), and you have to scratch away some of the surface to expose the bridge traces (the surface is transparent, so you can see the traces) before you can connect them. I decided it's just too hard to do without something to hold the cpu steady, and am calling it quits for now.