L2 Cache: 512kb vs 1024kb

[Audioaficionado]

Has anyone documented an advantage in having 1Mb cache over just 512kb when folding? Especially the Big Gromacs, Double Gromacs and QMDs.

I'm trying to see if the extra expense for the larger cached chips is justified.

 

[ChasR]

Cache size above 512 KB doesn't make much difference on any WUs other than p147x/81.

On QMDs a P4C (512KB cache) with dual channel, decent ram will do 145ppd/GHz while a 640 (2MB) will do 158ppd/GHz, the 5xxs are in between.

On regular Gromacs A64s with 512 KB caches produce the same ppd/GHz as those with 1MB caches ~60ppd/GHz.

On 600 pointers an A64 Winnie (512KB cache)puts out 125ppd/GHz while a FX-53 (1MB cache )puts out 133ppd/GHz.

On p147x/81, cache size makes a difference. THe A64 Winnie makes about 185ppd/GHz while the FX-53 makes 208ppd/GHz. A p4C makes about 100ppd/GHz while the 640 makes 170ppd/GHz.

On DGromacs the winnie makes 165ppd/GHz while the FX does 168ppd/GHz.

On GBGromacs they both do about 83 ppd/GHz. In fact, sig rig #3 also produces 83ppd/GHz on GBGromacs and its 85ppd/GHz for all three on Tinkers.

I'd say the chip that OCs the best will be the best producer unless another batch of WUs similar to p147x/81 comes out.

If you're going to use the rig for purposes other than folding (ie gaming), I'd get the big cache.

 

[pik4chu]

For a bit of "real world"
my 530Js do about 30-50ppd less than my new 630. all on more or les the same generic ram (pqi or kingston)

this is on QMDs with 1gig of ram each bigpackets=yes and -advmethods

 

[AlabamaCajun]

I can't believe that all the Intels only have 12K and 16L-L1 cache unlike the Amds have 64K and 64K. Not an Intel bash but it made me stop and think before buying an 820D.

 

[ChasR]

For a bit of "real world"
my 530Js do about 30-50ppd less than my new 630. all on more or les the same generic ram (pqi or kingston)

Not quite sure what you mean by "real world". I stand by my numbers and am sorry you're having trouble with your 530.

 

[Audioaficionado]

Thanx ChasR.
I knew I could count on you

 

[pik4chu]

Not quite sure what you mean by "real world". I stand by my numbers and am sorry you're having trouble with your 530.

not sure what you mean.. I said almsot exactly what you did. and by real world I simply meant 2 examples of specific processors. Again Im not sure why you think im 'having trouble' with something. My results match what you said just fine....

On QMDs a P4C (512KB cache) with dual channel, decent ram will do 145ppd/GHz while a 640 (2MB) will do 158ppd/GHz, the 5xxs are in between.

my 530 is currently ~155ppd/ghz and the 640 is ~168ppd/ghz sooooo @~3ghz the difference is 30-50, so what sthe problem here?

 

[David]

I can't believe that all the Intels only have 12K and 16L-L1 cache unlike the Amds have 64K and 64K. Not an Intel bash but it made me stop and think before buying an 820D.

AFAIK its not really fair to compare the two directly as they work in different ways in each case.

 

[ChasR]

I took the "real world" quote to mean you were questioning my numbers validity. Obviously you weren't. The problem with your 530 is that your 6xx is doing better than mine since it's getting 168 ppd/GHz. EMIII will report that high on frames with 20 steps but most have 21 and it then reports less. 158ppd is on an average lenght QMD, I might get just over 160 on a 2040 step (short) QMD. I'd put difference in production per day at closer to 15ppd @ 3 GHz based on the numbers I have at hand. I'm going to have to go looking for another 10 ppd on the 640 now.

 

[Dan Derrig]

Intel probably could benefit from larger L1 to keep it's pipelines fed better but it's like 12kb and 16kb, which is odd numbers so I'm guessing it's like that for a good reason. At least there is a tremendous 1 or 2mb L2

 

[pik4chu]

I took the "real world" quote to mean you were questioning my numbers validity. Obviously you weren't. The problem with your 530 is that your 6xx is doing better than mine since it's getting 168 ppd/GHz. EMIII will report that high on frames with 20 steps but most have 21 and it then reports less. 158ppd is on an average lenght QMD, I might get just over 160 on a 2040 step (short) QMD. I'd put difference in production per day at closer to 15ppd @ 3 GHz based on the numbers I have at hand. I'm going to have to go looking for another 10 ppd on the 640 now.

I had a feeling thats what you thought

As for mine, it seems my 640 is having some issues. just got an EE on it so will have to poke around with it and lower its OC a bit. Ill then recalc everything. the 530 was from this machine based on what EMIII was reporting. its clocked to 3.15Ghz I think (oddly I get less OC than I did with generic ram) So Ill come back in a day or two when I have more accurate numbers with my 640 and we can compare again. But all in all my 30-50ppd difference has held true for the most part. I just wish dual procs were more efficient. My 3ghz dual Xeon is producing about the same as these 530s

Anywho, will be updating this later with some better results and some specifics on RAM... which strangely CPU-Z wont report RAM timings on my 640. And not sure why, doesnt have to do with shared video cause one 530 and the 640 both hav eit. Just on different boards.

 

[ChasR]

Sig rig #2 averages 680 ppd on one QMD, one big WU and one Tinker. 660ppd without the Tinker. Two QMDs are only good for about 590ppd. If you happen to get a little Gromac on the big instance and aren't running the Tinker, you'll make less ppd than two QMDs. Maximum observed ppd is 778 on one QMD, one DGromac, and 600 pointer.

 

[pscout]

Anywho, will be updating this later with some better results and some specifics on RAM... which strangely CPU-Z wont report RAM timings on my 640. And not sure why, doesnt have to do with shared video cause one 530 and the 640 both hav eit. Just on different boards.

When I got my first 830, i had to update cpuz --- was running 1.26, now 1.30 ... not sure if that is ralated to your RAM timings not showing.

 

[pik4chu]

When I got my first 830, i had to update cpuz --- was running 1.26, now 1.30 ... not sure if that is ralated to your RAM timings not showing.

Im pretty sure I downloaded a fresh copy fromt eh site for that comp, maybe I got the wrong version, checking now. as well as messing with the OC to keep FAH stable. Seems to be going well so far.

 

[pik4chu]

the 640 was having 'corrupted' resuslts, for lack of a better term. it is actually a bit less than first calculated its closer to 160ppd/ghz atm. Might putz around later to see if I can get he OC back up, had to drop the fsb from 225->220 maybe it needs a little more juice. Oh well, too much on my schedule to mess with it now, but it should be producin its strong ~500+ppd and bring my stats up to where they should be. plus the extra ram in another HT. hopefully will reach top 20 in production with these fixes. If not, might have to add another rig soon since another 640 rig would only cost ~300$ from the egg

Ill try to get some ram timings soon to give a more accurate show of what I have running ppd wise. Later

-Chu!

 

[pik4chu]

Update

530J @3.0Ghz (OC is mobo limited I think, as my generic ram could goto ~215fsb, this OCZ wont go at all, anywho :Shrug RAM timings is 2.5-2-2-5 and current PPD/Ghz is 165.2 The estimate PPD by EM III is ~490-495 This has dropped to 470 in the past.
Current Protiens are P1911_ALA-depeptide_umbrel and P1912_ALA-depeptide_umb2

Calculations were done after rig sat idle overnight with all other applications closed. This is my office machine so the numbers get off during the day (as I use outlook, lots of IE windows, remote desktop, VPNs etc.)

To me this number sounds high, if its not thats great but Ill poke through the logs and try to get a more accurate number.

Will try and get info from other rigs tonight.

Excerpt from log of current protien (1911_ALA)
[16:02:41]
[16:02:41] *------------------------------*
[16:02:41] Folding@Home QMD Core
[16:02:41] Version 1.04 (Apr 7, 2005)
[16:02:41]
[16:02:41] Preparing to commence simulation
[16:02:41] - Assembly optimizations manually forced on.
[16:02:41] - Not checking prior termination.
[16:02:42] - Expanded 260836 -> 826209 (decompressed 316.7 percent)
[16:02:42]
[16:02:42] Project: 1911 (Run 68, Clone 4, Gen 12)
[16:02:42]
[16:02:42] Writing local files
[16:02:42] Extra SSE2/SSE3 boost OK.
[16:02:42] Entering QMD...
[16:03:34] System: p1911_ALA-depeptide_umbrel
[16:03:34]
[16:03:34] Performing initial WF calculations
[16:03:34] - Number of total steps will change until convergence
[16:04:50] Completed 0 out of 2000 steps (0)
[16:18:02] Timered checkpoint triggered.
[16:23:26] Completed 21 out of 2021 steps (1)
[16:23:26] Writing local files
[16:39:17] Timered checkpoint triggered.
[16:41:06] Completed 41 out of 2041 steps (2)
[16:41:06] Writing local files
[16:56:10] Timered checkpoint triggered.
[16:59:44] Completed 62 out of 2062 steps (3)
[16:59:44] Writing local files
[17:13:55] Timered checkpoint triggered.
[17:13:59] WF converged, jumping to MD
[17:14:00] Verifying checksum
[17:14:00] Finished
[17:15:14] Completed 78 out of 2078 steps (3)
[17:22:27] Completed 84 out of 2078 steps (4)
[17:22:27] Writing local files
[17:38:26] Timered checkpoint triggered.
[17:47:09] Completed 104 out of 2078 steps (5)
[17:47:09] Writing local files
[18:02:44] Timered checkpoint triggered.
[18:07:02] - Autosending finished units...
[18:07:02] Trying to send all finished work units
[18:07:02] + No unsent completed units remaining.
[18:07:02] - Autosend completed
[18:12:28] Completed 125 out of 2078 steps (6)
[18:12:28] Writing local files
[18:28:18] Timered checkpoint triggered.
[18:38:06] Completed 146 out of 2078 steps (7)
[18:38:06] Writing local files
[18:53:07] Timered checkpoint triggered.


Excerpt From log for current protien (1912_ALA)

[03:03:32] 
[03:03:32] *------------------------------*
[03:03:32] Folding@Home QMD Core
[03:03:32] Version 1.04 (Apr 7, 2005)
[03:03:32] 
[03:03:32] Preparing to commence simulation
[03:03:32] - Assembly optimizations manually forced on.
[03:03:32] - Not checking prior termination.
[03:03:32] - Expanded 260960 -> 826210 (decompressed 316.6 percent)
[03:03:32] 
[03:03:32] Project: 1912 (Run 235, Clone 2, Gen 79)
[03:03:32] 
[03:03:32] Writing local files 
[03:03:32] Extra SSE2/SSE3 boost OK.
[03:03:32] Entering QMD...
[03:04:25] System: p1912_ALA-dipeptide_umb2
[03:04:25] 
[03:04:25] Performing initial WF calculations
[03:04:25] - Number of total steps will change until convergence
[03:05:39] Completed 0 out of 2000 steps  (0)
[03:18:48] Timered checkpoint triggered.
[03:24:07] Completed 21 out of 2021 steps  (1)
[03:24:07] Writing local files
[03:39:11] Timered checkpoint triggered.
[03:41:54] Completed 41 out of 2041 steps  (2)
[03:41:54] Writing local files
[03:56:55] Timered checkpoint triggered.
[04:00:33] Completed 62 out of 2062 steps  (3)
[04:00:33] Writing local files
[04:15:46] Timered checkpoint triggered.
[04:20:11] Completed 84 out of 2084 steps  (4)
[04:20:11] Writing local files
[04:36:00] Timered checkpoint triggered.
[04:39:34] Completed 106 out of 2106 steps  (5)
[04:39:34] Writing local files
[04:55:14] Timered checkpoint triggered.
[04:58:48] Completed 128 out of 2128 steps  (6)
[04:58:48] Writing local files
[05:14:36] Timered checkpoint triggered.
[05:19:06] Completed 151 out of 2151 steps  (7)
[05:19:06] Writing local files
[05:25:14] Timered checkpoint triggered.
[05:25:19] WF converged, jumping to MD
[05:25:19] Verifying checksum
[05:25:20] Finished
[05:26:35] Completed 158 out of 2158 steps  (7)
[05:40:35] Timered checkpoint triggered.
[05:45:39] Completed 173 out of 2158 steps  (8)
[05:45:39] Writing local files 
[06:00:42] Timered checkpoint triggered.
[06:07:02] - Autosending finished units...
[06:07:02] Trying to send all finished work units
[06:07:02] + No unsent completed units remaining.
[06:07:02] - Autosend completed
[06:13:16] Completed 195 out of 2158 steps  (9)
[06:13:16] Writing local files 
[06:29:22] Timered checkpoint triggered.
[06:39:14] Completed 216 out of 2158 steps  (10)
[06:39:14] Writing local files 
[06:54:23] Timered checkpoint triggered.
[07:06:34] Completed 238 out of 2158 steps  (11)
[07:06:34] Writing local files 
[07:22:22] Timered checkpoint triggered.
[07:32:06] Completed 259 out of 2158 steps  (12)
[07:32:06] Writing local files 
[07:47:45] Timered checkpoint triggered.
[07:58:42] Completed 281 out of 2158 steps  (13)
[07:58:42] Writing local files 
[08:14:21] Timered checkpoint triggered.
[08:25:18] Completed 303 out of 2158 steps  (14)
[08:25:18] Writing local files 
[08:40:54] Timered checkpoint triggered.
[08:50:37] Completed 324 out of 2158 steps  (15)
[08:50:37] Writing local files 
[09:06:20] Timered checkpoint triggered.
[09:17:16] Completed 346 out of 2158 steps  (16)
[09:17:16] Writing local files 
[09:32:59] Timered checkpoint triggered.
[09:42:43] Completed 367 out of 2158 steps  (17)
[09:42:43] Writing local files 
[09:58:23] Timered checkpoint triggered.
[10:09:18] Completed 389 out of 2158 steps  (18)
[10:09:18] Writing local files 
[10:25:01] Timered checkpoint triggered.
[10:35:53] Completed 411 out of 2158 steps  (19)
[10:35:53] Writing local files 
[10:51:35] Timered checkpoint triggered.
[11:01:15] Completed 432 out of 2158 steps  (20)
[11:01:15] Writing local files 
[11:16:55] Timered checkpoint triggered.
[11:27:46] Completed 454 out of 2158 steps  (21)
[11:27:46] Writing local files 
[11:43:21] Timered checkpoint triggered.
[11:52:56] Completed 475 out of 2158 steps  (22)
[11:52:56] Writing local files 
[12:07:02] - Autosending finished units...
[12:07:02] Trying to send all finished work units
[12:07:02] + No unsent completed units remaining.
[12:07:02] - Autosend completed
[12:08:32] Timered checkpoint triggered.
[12:19:17] Completed 497 out of 2158 steps  (23)
[12:19:17] Writing local files 
[12:35:00] Timered checkpoint triggered.
[12:44:40] Completed 518 out of 2158 steps  (24)
[12:44:40] Writing local files 
[13:00:28] Timered checkpoint triggered.
[13:11:22] Completed 540 out of 2158 steps  (25)
[13:11:22] Writing local files 
[13:27:09] Timered checkpoint triggered.
[13:38:03] Completed 562 out of 2158 steps  (26)
[13:38:03] Writing local files 
[13:53:53] Timered checkpoint triggered.
[14:03:34] Completed 583 out of 2158 steps  (27)
[14:03:34] Writing local files 
[14:19:18] Timered checkpoint triggered.
[14:30:09] Completed 605 out of 2158 steps  (28)
[14:30:09] Writing local files 
[14:45:59] Timered checkpoint triggered.
[14:55:36] Completed 626 out of 2158 steps  (29)
[14:55:36] Writing local files 
[15:11:25] Timered checkpoint triggered.
[15:22:13] Completed 648 out of 2158 steps  (30)
[15:22:13] Writing local files 
[15:37:55] Timered checkpoint triggered.
[15:47:40] Completed 669 out of 2158 steps  (31)
[15:47:40] Writing local files 
[16:03:16] Timered checkpoint triggered.
[16:13:08] Completed 691 out of 2158 steps  (32)
[16:13:08] Writing local files 
[16:29:04] Timered checkpoint triggered.
[16:40:05] Completed 713 out of 2158 steps  (33)
[16:40:05] Writing local files 
[16:56:08] Timered checkpoint triggered.
[17:05:50] Completed 734 out of 2158 steps  (34)
[17:05:50] Writing local files 
[17:21:37] Timered checkpoint triggered.
[17:32:35] Completed 756 out of 2158 steps  (35)
[17:32:35] Writing local files 
[17:48:30] Timered checkpoint triggered.
[17:58:09] Completed 777 out of 2158 steps  (36)
[17:58:09] Writing local files 
[18:07:02] - Autosending finished units...
[18:07:02] Trying to send all finished work units
[18:07:02] + No unsent completed units remaining.
[18:07:02] - Autosend completed
[18:13:50] Timered checkpoint triggered.
[18:24:50] Completed 799 out of 2158 steps  (37)
[18:24:50] Writing local files 
[18:40:45] Timered checkpoint triggered.
[18:52:03] Completed 821 out of 2158 steps  (38)
[18:52:03] Writing local files

for more accurate numbers of what this proc/rig does go by the earliest steps after convergence as Ive been using this computer for the last hour or so.