VOGONS


First post, by 386SX

User metadata
Rank l33t
Rank
l33t

Hi, i bought a Prescott 3,4ghz 1mb/800 socket 478 expecting to be faster than the 3,0 prescott model on the asrock p4i65g mobo
In linux result slower in benchmarks and quiet a lot less. Bios sees it correctly and so applications. But intel identification utility seems to read real time freq as 2.0ghz and 492mhz fsb...and it would make sense on the results. In a test called cpu blowfish with the 3.0 i got 11.9 points, now 17.9. In FPU raytracing before i got 70 with the 3.0 now 100 with the 3.4ghz...(lower better)
what could be wrong? Bios latest, intel speedstep disabled, thermal throttling too.
Thanks

Last edited by 386SX on 2018-10-15, 17:36. Edited 1 time in total.

Reply 1 of 16, by 386SX

User metadata
Rank l33t
Rank
l33t

Maybe I solved but still need to make some more test. I disabled the ICH Thermal Throttling on the bios and now it seems to run a bit faster than the 3.0Ghz. Cpu test gives 11.0 points vs 11.9 of the 3.0, and FPU Raytracing give 56 points vs 70 of the 3.0Ghz. It's strange cause the setting was already before to Auto with the 3,0Ghz.

Reply 2 of 16, by agent_x007

User metadata
Rank Oldbie
Rank
Oldbie

I recommend you better mount some BIG heatsinks on those MOSFETs.
3,4GHz isn't slow for Prescott... and using it on a mATX board with TWO phases (and no throttling enabled), 100% WILL kill your board.

Out of curosity : What MOSFETs are used on your board ?
I see two Low side, and one High side (per phase).

Last edited by agent_x007 on 2018-10-15, 18:02. Edited 1 time in total.

157143230295.png

Reply 3 of 16, by SW-SSG

User metadata
Rank Oldbie
Rank
Oldbie

That won't live very long, I don't think... especially now with those BIOS self-preservation measures (throttling, SpeedStep) disabled.

Reply 4 of 16, by 386SX

User metadata
Rank l33t
Rank
l33t
agent_x007 wrote:
I recommend you better mount some BIG heatsinks on those MOSFETs. 3,4GHz isn't slow for Prescott... and using it on board with T […]
Show full quote

I recommend you better mount some BIG heatsinks on those MOSFETs.
3,4GHz isn't slow for Prescott... and using it on board with TWO phases (and no throttling enabled), will kill your board.

Out of curosity : What MOSFETs are used on your board ?
I see two Low side and one High side per phase.

I read something like BA IS16.. or maybe BA 1516 something like that.. above there's what it looks like a data 0452 (52 week of the 2004?). there are four ones external and two ones close to the cpu.

Reply 5 of 16, by agent_x007

User metadata
Rank Oldbie
Rank
Oldbie

Just use a phone with decent camera (zoom/macro option), and turn the flash on (no pun on TV series intended) 😀
Auto focus should do the rest (it may work better at bigger distances).

Last edited by agent_x007 on 2018-10-15, 18:09. Edited 1 time in total.

157143230295.png

Reply 6 of 16, by 386SX

User metadata
Rank l33t
Rank
l33t
SW-SSG wrote:

That won't live very long, I don't think... especially now with those BIOS self-preservation measures (throttling, SpeedStep) disabled.

Do you think ich thermal was enabling itself cause the mobo can't work right with this cpu?

🙁

The mobo was new boxed never used. I was thinking being supported by the bios would mean it would last as usually with other cpus. At the wall the computer (with a Radeon HD 3450 passive, Oxygen HD, SSD, Sata controller pci and 2x1GB CL2 DDR and DVD sata) ask around 115 watts idle (maybe 5W less with radeon_dpm kernel module) and I think close to 200 watts during benchmarks.
For the mosfet I'd need to take a photo with a better camera cause my phone take bad low res low detail photo.

Reply 7 of 16, by agent_x007

User metadata
Rank Oldbie
Rank
Oldbie

I think it's VRM throttle (thank god for that), which can't be disabled in BIOS and is build into VRM controller chip.
From experience point of view :
I own P4i945, together with 3,2GHz E0 Prescott and that board has three phase VRM.
It still throttled sometimes on benchmarks (I used stock CPU cooling).
Only adding heatsinks on chokes solved my problems (and enabled some OC on top of that).

157143230295.png

Reply 8 of 16, by 386SX

User metadata
Rank l33t
Rank
l33t

About the mosfet it wasn't the date above it's the model. They all are D452 BA1S16 mosfet.

I'll look for some heatsinks even if the usual ram heatsink would be too big for them.

Reply 9 of 16, by agent_x007

User metadata
Rank Oldbie
Rank
Oldbie

Good news : MOSFETs are "OK" for up to 55Amps 😀
Bad news : Each one has 25W of TDP while at 100C for that rating (LINK)
Basicly : Uncoolable (that's around laptop Core i3 level of TDP on MOSFET footprint FYI).
Throwing heatsinks on them is still your best bet.

Also, that two bare wire inductors/chokes - can't possibly be rated for 75 Amps.
So, let's assume they are 50 Amp ones, basic math : 2x 50A x 1,35V = 135W.
If you overload/saturate an inductor, it blows up after a while (sending 12V to whatever is connected on the other side).

PS. You get throttling on Load, so usage power value for it... probably is underrated a bit.

157143230295.png

Reply 10 of 16, by 386SX

User metadata
Rank l33t
Rank
l33t
agent_x007 wrote:
Good news : MOSFETs are "OK" for up to 55Amps :) Bad news : Each one has 25W of TDP while at 100C for that rating (LINK) Basicly […]
Show full quote

Good news : MOSFETs are "OK" for up to 55Amps 😀
Bad news : Each one has 25W of TDP while at 100C for that rating (LINK)
Basicly : Uncoolable (that's around laptop Core i3 level of TDP on MOSFET footprint FYI).
Throwing heatsinks on them is still your best bet.

Also, that two bare wire inductors/chokes - can't possibly be rated for 75 Amps.
So, let's assume they are 50 Amp ones, basic math : 2x 50A x 1,35V = 135W.
If you overload/saturate an inductor, it blows up after a while (sending 12V to whatever is connected on the other side).

PS. You get throttling on Load, so usage power value for it... probably is underrated a bit.

Thanks! Interesting!

I've seen that there're two different model of this cpu, one with "loadline A at 103W" and another "loadline B at 89W". Mine is a SL7E6 so core stepping D0 I don't know but I think it could be the 103W one, am I right? So I've still some free space on the wattage requirements to actually hope that frequency scaling would help it? Considering at idle the whole computer ask 115 watts at the wall (compared to 105W of the P4 3,0ghz).

Anyway I'll look for some compatible heatsink.

Reply 11 of 16, by agent_x007

User metadata
Rank Oldbie
Rank
Oldbie

It's an "A" : LINK
Interesting.bit is 80A max. on CPU, according to intel VRM10.1 Spec...
Better cooling is good, but remember to have direct airflow over VRM.

157143230295.png

Reply 12 of 16, by 386SX

User metadata
Rank l33t
Rank
l33t
agent_x007 wrote:

It's an "A" : LINK
Interesting.bit is 80A max. on CPU, according to intel VRM10.1 Spec...
Better cooling is good, but remember to have direct airflow over VRM.

Not really easy to put a fan directly over those mosfet.. need to study some custom made metal support for the fan.

Reply 13 of 16, by canthearu

User metadata
Rank Oldbie
Rank
Oldbie

I solved my Prescott problem (was a 3.2ghz Celeron D) by installing a Cedar Mill P4 😀

Unfortunately, not an option of 478 socket 🙁

Reply 14 of 16, by 386SX

User metadata
Rank l33t
Rank
l33t

For now and with the p4 kernel module for frequency scaling it's working fine now. At idle it's around 105-110W for the entire computer and I've not seen values above 180W in benchmark for the entire pc. Mosfet seems just warm but still, I'll try to put some heatsinks for safety. I remember the Athlon XP 3200+ build I tried some times ago, its values were from 100W to 130W for the entire pc same gpu so not so far considering how much faster it should be.

The 400 Mhz more of this Prescott surely does quiet some difference! Even if obviously modern kernel and applications would need much faster architectures, I'm sure it does its best considering that I imagine Hyperthreading and SSE1/2/3 to be helpful anyway. It's quiet amazing and still I've to enable gpu compositing that will help to speed up a bit and trying proprietary drivers for the gpu. Also maybe some more speed could come from the ram choice if I'll get some CL2 memories.

Reply 15 of 16, by canthearu

User metadata
Rank Oldbie
Rank
Oldbie

Just curious, why a P4 for this linux box rather than a core 2 or other newer/more efficient processor. Almost anything is better than a P4

Compatibility shouldn't be so much of an ugly issue with linux.

Reply 16 of 16, by 386SX

User metadata
Rank l33t
Rank
l33t
canthearu wrote:

Just curious, why a P4 for this linux box rather than a core 2 or other newer/more efficient processor. Almost anything is better than a P4

Compatibility shouldn't be so much of an ugly issue with linux.

I have faster components but It'd have been too easy to use a core 2 that is just on a different level. The P4 Prescott with HT and SSE3 seems still quiet "modern".

Before testing the Core 2 i will try the Pentium D.😀