p4436, p4437 missing & performance q

[Foxery]

Projects 4436 and 4437 are being assigned with somewhat high frequency to Windows Classic clients, but are not available on the Projects page for utilities such as FahMon. (Gromacs, 225 points like 4438 I would assume.)

Also worth noting, these only have 7885 atoms, but produce about 1/3rd the PPD on many different Core2-family CPUs compared to other recent Gromacs projects with 10x as many atoms. Seems a bit odd.

[Foxery]

Upon further consideration, perhaps the larger L2 cache is what gives Core2s such a performance gain on large proteins over the P4 Reference Machine. Being small could mean the gains come solely from IPC enhancements. Bummer. 2483-85 and 5113-5114 are wonderful on Core2s.

[PeterA]

I got a 4436 yesterday. FahMon doesn't recognize the project. Regardless, everything is running smoothly. Someone should type these up in the project summary.

[brityank]

If your Project Number is not reflected in the Project Summary Chart, select one of the Description Links, and put your WU Project Number in the window there on the bottom of that page. 4437 then shows up; but it still doesn't tell the point value (225) or Core (Gromacs). Some do, others don't - depends on the writer, I guess. Cheers.

[Teddy]

I have been getting plenty of p4436/p4437 & Fahspy doesn't know what they are worth either. Guess it is too much to ask for Stanford to assign a points value to these in the summary page so utilities like Fahspy & fahmon can work out what they are all worth PPD wise.

brityank where did you get the 225 point value from?

Teddy

[brityank]

brityank where did you get the 225 point value from?

Teddy

I too have had a bunch, and saw the value added through FAHmon updates on its three-hour boundary. Give them a break, I'm sure they have more important things to do, which is why we all help out here.

Merry Christmas and Happy New Year. Cheers!

[bruce]

Guess it is too much to ask for Stanford to assign a points value to these in the summary page so utilities like Fahspy & fahmon can work out what they are all worth PPD wise.

I know it's frustrating. The psummary data is generated automatically. That often means that when a project first appears, it can be assigned to you before it appears in psummary or that the server in question is not talking to the network when psummary is updated so the project disappears.

I also know that Stanford has rewritten the server code and that's being tested. I'm not sure which projects are on the test server(s) but getting the new code to talk to the psummary script correctly is being worked on.

[Teddy]

I guess it is human nature to be curious what these new proteins are worth, ah Yes the rewritten Server code thats in the wings. I imagine that is taking a fair few resources to get right. I hope that is going well?

Thanks for your post brityank, its hard for me to be that observant as I am running many clients here

Merry Xmas all..

Teddy

[DrBB1]

Happy New Year!

My last four WUs have been in the 4436-4447 range; each has resulted in less than half the PPD I had been earning--on average--over the past 18 months. My PCs both have single CPUs, so this may be an issue. But is it possible that the points per unit need to be revisited for these new units?

[bruce]

Yes, it is possible . . . and only the Pande Group can determine if rebenchmarking is warranted.

It's also possible that you've been running mostly bonus WUs for the past year or so. The Amber projects were specifically announced as temporary high priority bonus projects and if that's what you're comparing them to, you are not understanding the nature of the temporary bonus.
See viewtopic.php?f=24&t=2241
Some time later, the WUs were deemed stable so the requirement for the -advmethods flag was removed thereby increasing participation.

How does your hardware compare to the benchmark hardware which earns 110 PPD?

[DrBB1]

bruce, sorry it took me so long to get back to you. I've been busy and I haven't had much time to figure out how my main machine compares to benchmark machine re PPD. As best as I can tell, based on a sample of recent runs and total points accumulated to date (I used to have three CPUs, so I needed to try to account for recently running just two), I am probably close to 110 PPD on my "speed" machine (about 40/day on my turn-of-the-century model). I checked the benchmarks on my Amber projects and they have been 30-50% higher, but make up a small portion of my main processor's WUs.

With 10 projects numbers for WUs between 4434 and 4458 (I don't know how many WUs; probably 15-20), eight of the ten come in at about 45-60 PPD. The two exceptions have been 4440 (110 PPD) and 4444 (128 PPD). So pending further analysis than I can do, I am convinced that the 4430-4499 projects--at least in the range I have seen so far--are either undervalued or the characteristics of my PC are uniquely unsuited to running them. However, since I otherwise average close to the 110 PPD benchmark on my main CPU, I think further investigation may be warranted, particularly if others may be having the same issue.

Anyway, I'm not losing sleep over the points. I am more concerned that the benchmarks are accurate and that the drop in performance on some machines does not reflect a broader problem with the results. After all, there is a greater purpose than points in the Stanford work, isn't there?

[MtM]

Sorry to bump an old thread, but getting 90ppd on these is abit low for a q6600 don't you all agree Idk what

It's outweighed on average with the 3798's which give me 2.5k ppd per core if I get them, but on their own they should clearly be rebenched as they are not really representative of a kentsfield if a p4 would get 110 ppd