Project 1231 p2357_Ubiquitin and the huge DGromac WU's

[Kurnous]

Some time ago I received ;

Protein 1573 p2363_mAcP_mutant totalling 722 points.

With my X2 3800+ @ 2.5 and memory @250 2.5-3-2-6 it would take roughly 50hrs. It also used only 48mb of RAM!!!!

However when a 364 pointer hit my other core it added an extra hr and abit to the time. My 364 pointer usually takes alittle over 20hrs unless there is another "bandwidth sucker" usually another 364. That usually means that both cores take an extra 2 hours (22hrs)

In this case the newer core only upped the 364 time by alittle over an hour.

Just some interesting analysis for you guys. I can get more accurate readings if you'd like.

The new large WU's are nice but I still prefer to get 364 pointers as they give me higher PPW.

BTW how do we know if SSE2 is being used with these new large WU's? How are P4's handling them?

Do P-D's suffer from "bandwidth suck" when processing 2 bandwidth greedy WU's?

M2 looks promising due to the fact that there is more bandwidth and therefore there will be less of that Phenomena I have just noted in this post.

BTW shouldn't someone attempt to compile a rough Database on how different WU's perform with different processors/speeds/RAM/timings etc? I'd gladly donate any and all my results to it!

EDIT: Fark! SSE2 is enabled in the new WU I mentioned. How did Stanford manage that? Also I have in my EMIII an blue circle with 1 in it meaning that there is a WU still unsent. It's been there for months. Can someone explain how I can get rid of it?

[Davidvfr]

I finally picked up one of these dgromacs this afternoon.



P2355 for 683 points.

Taking just over 60 hours. I'm running a stock AMD A64 3200 with 1gb ram.

[Bollocks]

Originally Posted by Kurnous

Also I have in my EMIII an blue circle with 1 in it meaning that there is a WU still unsent. It's been there for months. Can someone explain how I can get rid of it?

You can delete the wuresult file in the work folder if it has been there that long.

[Jarwedy]

Ive been getting a steady p14xx somthings, worth 364 points lol. Hopefully fininsih sometime tommrow night, and hope to get one off these new WU's .... mmm dunno if big_wu's is on, *checks *

Anyone willing to pose up frame times for whatever one off the new WU's they've got, And probably CPU.

eg.

Originally Posted by Kurnous

Protein 1573 p2363_mAcP_mutant totalling 722 points.

With my X2 3800+ @ 2.5 and memory @250 2.5-3-2-6 it would take roughly 50hrs

[phathag]

just finished my first 1280 pointer!
Protein: p2360_Coiled_Coil

seemed to take at least 4.5 days i beleive (most times taking 1 hour and 4 minutes per step (100 steps)

Nice to see my new dual core optie getting some big pointers... just doesnt look as if the ppd is amazing though. maybe later after the heatwave dies down i might OC this bad boy.

now if i can only grab 2 concurrently!

[pengy]

just started folding again after a long break - got one of these first up. Nearly fell over when I saw the estimated time to completion... ~7days! on a 3500+ !!! ahh well. 500 frames at approximately 25 mins/frame.

[sam_allen]

Heres a nice comparison between CPU's for you

A screenshot of my EMIII shows my 4 active cpu's...

'Zeus' is my X2 4400+ Boxen (Zeus-1 being core 1, Zeus-2 being core 2)
Artemis is my A64 3200+ Boxen

Hercules is my P3 667 Server Boxen.


Looking at the frame times... it pretty much shows that the 4400+ is more than twice as quick as the 3200+ doing the same WU



Click to view full-sized image!
Hosted by UGBox Image Store

Im happy Still.

[Kurnous]

Dude, sorry to shoot you down but if you average Zeus 1 and 2

Time per Frame...13:05 and 13:08 / 2 = 13:065.

Then Artemis which has a TPF of 15:45.....

Thats actually only roughly an 18.25% increase from a single core of a 4400+ compared to a 3200+ (Which core BTW?)

The 4400+ core also runs at 200mhz higher then the 3200+ and thats an extra 10% increase in clock speed.

Assuming you have the same m/b and same ram running same timings (also same bandwidth available), the extra 512k of cache only adds 8.25%.

However I will qualify that statement by saying that when a dual core has 2 364 pointers running, the total time does go up around 1 hr or so.

I am assuming is because they both demand alot of bandwidth and hence memory bandwidth bottlenecks them alittle. Not sure how true that is for dual cores with 2 x 1mb cache but it is the case for the proc. in my sig.

Peace

[sam_allen]

Umm... you're silly? Correct?

My 4400+ is completing TWO 364 point units in less time than the 3200+ is, take another look.

The 4400+ is completing 728 points in around 23hrs, whilst the 3200+ is completing 364 points in 25hrs....

Whilst I understand you've said gain per core, im talking in a sense of overall points output...

[Kurnous]

Okay I was talking per Core.

Yep you are getting twice the production which is always sweet.

BTW Are you overclocking at all cause the score seems alittle low? with a 4400+ you should at least get 2.4 probably 2.5-2.6 with a 1/2 decent heatsink.

Oh and how long does a 364 pointer take if your other WU is one of those pissy little ones?

[sam_allen]

Using all stock gear atm... The standard heatsink seems half decent actually, 4 heat pipes

Not overclocked at the moment, not running overly hot at the moment, wouldnt be suprised If I could get it to 2500Mhz with no problems...


Having a smaller/different WU on the other core seemingly makes no difference at all.

[Kurnous]

hmmm thanks for the info Sam. It seems that the 1mb Cache lessens the dependency on bandwidth.

[SteveMenheere]

I just realised why my P3@600 hasn't got any points for the past 10 days - its 49% through a 686 point p2352 and taking 6+hours per frame. At least he deadline is long enough to let it finish.

[ssmack]

I'm hoping my P3@535 doesn't get one. It gets 100ppw at the moment. I've had a steady diet of 600's on my other 3 boxen. Makes for a rollercoaster on the stats line.