boinc is at it again weird benchmark scores

[-Acid-]

hi all now my p4 northwood is under scoring in benchies
P4 2.4c northwood @ 3ghz
12x250 1.65v
mem at 250 7-3-3-2.5 2.8v
number of CPUs : 2
1276 floating point MIPS (Whetstone) per cpu
1321 interger MIPS (Drystone) per CPU

what the hell is going on over there ????

anyone else running a p4 northy waht scores you getting ?

[[XC] itznfb]

i have a P4 2.4C @ 3ghz with a 512 stick of gskill value ram. i run the benches on that.

edit:
1371 whetstone
1200 dhrystone

[-Acid-]

is that with or with out ht enabled
i just oced it more to 260 x11 3122mhz and got
1337 whetstone
1257 drystone

[[XC] itznfb]

with HT enabled

[Martijn]

You should bench with HT off, or it will run 2 threads on a single CPU, which sucks. Then turn it on when you start working :)

[-Acid-]

with ht off i get 1671 whetstone and 3218 drystone

how do i turn ht on once the computer has loaded????

[[XC] hipno650]

as far as i now it is bios only (enable/disable) so don't worry about it just turn ht on to get te most of your cpu and the quorum system will make it so you lose no points because of it.

[[XC] riptide]

Jaysus ACID.,.. the oldest trick in the Boinc book. When benching... in teh Bios turn off HT. When updated your bench results restart said machine and in the BIOS turn HT back on. Some would say its cheating (actually Vapor had hinted in Rosetta that he considered it.... borderline.... ) but other than that its a winner. :)

[-Acid-]

i,m new to boinc tho :( cheers riptide

you can't teach an old dog new tricks, but they can show you some :)

[meshmesh]

This is probably a stupid question. As I understand it, HT allows the CPU to be split into two thus appearing to the OS as two cores. Ok, fine then it will run two WUs on a single core. But since there is just one core anyway, it will take twice the time. So why enable it at all then? I understand that it would be good if processes are waiting for I/O then the second process can run, etc... But these boinc processes are pure CPU and no I/O (except at rare occasions of initial data access, checkpoints, and the like....).

Is there any data that shows that there is a substantial measurable benefit to running two faahs (for example) concurrently on a single core? Is the machine actually able to do more over time (say a week)?

If it is of benefit, why did Intel abandon it then? Thanks.

[[XC] riptide]

This is probably a stupid question. As I understand it, HT allows the CPU to be split into two thus appearing to the OS as two cores. Ok, fine then it will run two WUs on a single core. But since there is just one core anyway, it will take twice the time. So why enable it at all then? I understand that it would be good if processes are waiting for I/O then the second process can run, etc... But these boinc processes are pure CPU and no I/O (except at rare occasions of initial data access, checkpoints, and the like....).

Is there any data that shows that there is a substantial measurable benefit to running two faahs (for example) concurrently on a single core? Is the machine actually able to do more over time (say a week)?

If it is of benefit, why did Intel abandon it then? Thanks.
Mesh, in context to this HT 'tweak' for Boinc, the reason its used, in plain all to see English, is that Boinc is so retarted, that when it is disabled, your bench is on the single logical core, rather than having 2 benchs operate at the same time on each HT core ie. lower score. Boinc will then apply the high bech mark to the 2 WU's completed on each HT core.

in general terms, HT can be very effective. I've found that when running SOB plus any BOINC project, that the output of SOB does not half, but drops to around 60-65%. (So my HT enabled chip would be giving me 115%DC effort. Most would intuitively expect it to drop to 50%. this is not the case.

I thought HT will be back again in other Core2 chips in the future. Anyways, having a dual core machine does not merit having HT aswell.

[meshmesh]

Mesh, in context to this HT 'tweak' for Boinc, the reason its used, in plain all to see English, is that Boinc is so retarted, that when it is disabled, your bench is on the single logical core, rather than having 2 benchs operate at the same time on each HT core ie. lower score. Boinc will then apply the high bech mark to the 2 WU's completed on each HT core.

in general terms, HT can be very effective. I've found that when running SOB plus any BOINC project, that the output of SOB does not half, but drops to around 60-65%. (So my HT enabled chip would be giving me 115%DC effort. Most would intuitively expect it to drop to 50%. this is not the case.

I thought HT will be back again in other Core2 chips in the future. Anyways, having a dual core machine does not merit having HT aswell.
Thanks for the plain English explanation. Even I understood it.:)

Now theoretically speaking, does this mean that this disable/enable process needs to be repeated every five day cycle?:eek:

[[XC] riptide]

Thanks for the plain English explanation. Even I understood it.:)

Now theoretically speaking, does this mean that this disable/enable process needs to be repeated every five day cycle?:eek:
Theoretically, AND pratically. Its annoying... but.....

Another real way of driving home the benefit of HT, particularly in Rosetta, where benchs were king, was that my 3.4 P4 Northwood, when benched at 4+ghz, and using said tweak, was taking home the same as AMD 3800 X2's Using the optimizations supplied by crunch3r.

[[XC] gomeler]

As said before, HT splits the logic core into 2 cores. Works well for some applications such as webserving as I remember hearing reports of apache and mysql seeing huge benefits with HT enabled chips. HT will be re-introduced into the generation after next Intel processors iirc. I think it is 4 unified cores with 4 HT cores, I can't wait :p: In a very few cases HT has shown to actually destroy performance but you'd notice this as your times to completion would shoot through the roof. Just spend 3 minutes a week for a huge increase in points, or send the rig to me :)

[joshd]

Suppose you have to be careful with that though... up the benchmarks too much and it will invalidate units... ?