This should really be fix :( thats alot of time making errors and not send back good data.

It is good to know that there are Errors being generated other than on my machine! I note with interest that those with i7 processors (I'm running an OC 2.66 920 @ 3.66Ghz and stable at 53C, so no core temp issues, OS W7Pro) seem to be having more errors - there may be something in this - as I am also running Grid on a 2 Core (3.2Ghz no OC and OS XP Pro) and this is churning out perfect HPF WU's every time - the i7 is getting about 50-60% - the reminder drop out in less than 5 minutes, often less than 2 mins, so no real loss of personal productivity - but is there a loss of efficiency overall? I am running 5 different projects in the usual random mix of WU's, nothing is exclusive to either machine - is there pattern of HPF Errors on i7's running W7 I wonder? Does anyone have anything to add to this theory????

Oh Yes - I've also noticed that the points credited per WU (doesn't seem to matter which project) on the Core 2 is about 15-20% higher than the points claimed - on the i7, the reverse happens, the points claimed per WU is almost always higher than the points credited, again by 10-15% or so - is there a reason for this?

I have no i7 to check it, but if it is the same as for my P4 HT the benchmark in HT mode is far too high for the work done. With regards to the actual job throughput (30 % better in HT mode) the benchmark should be about 65 % of what it is when not using HT, but it is only slightly smaller than the single-task one, so this HT device is overclaiming (longer runtime time with almost the same benchmark) and that gets corrected when validating.

But not as much as it should as your percentages show.

Cheers. Jean.

How do your quad manage to run 8x more than a single-core?

More accurate would be:
A quad-core running 100% will do maybe 3.9x A wu/day, and should get 3.9x points/day. **


A quad with HT will run 8 threads, not 16.

As for performance, this can be heavily influenced by the application, whatever other things you're doing, and so on.

But, if you're really unlucky, so you're running a memory-bandwidth-limited application, you'll maybe land on 3.8x A wu/day, and should get 3.8x B points/day, then you're running 8 threads on a quad-HT.

More commonly, you'll land somewhere between 4x and 5x A wu/day, and should get 4x - 5x B points/day, then you're using HT.

Exactly where you lands depends on the mix of applications and so on, it's possible running 8x wu of one application will give no benefit, while running 8x wu on another application gives maybe 25% increase. Or, running a mix of 8 different applications is the best...


** I've not tested it on i7 or similar, but some applications often has a "smp-penalty", there running N instances on N physical cores gives longer times per wu, than running 1 instance on 1 physical core. Also, how much, if there is an effect, depends on application...

A good example on the "smp-penalty" is with my long-since retired p3-dual-cpu, there running 1 seti-wu used N hours, while running 2 seti-wu simultaneously used 1.1x N hours, meaning a "penalty" of 10%. The total wu/day was of course much higher running 2 at once, but by no means a linear increase.

HT is only to trick the application to think it has more then one core or more spreading the performance across the board. It does not actually have the strength of extra cores. this was my understanding.