|
21)
Message boards :
Number crunching :
Plea for 'admin' help: merging computers
(Message 236633)
Posted 24 Jan 2006 by Mark A. Craig
Post: There's a small annoyance I'd like to fix: I first used BOINC for a while in June of 2004 with one of the two systems I'm using now (the XP 2500+), but then had to take a 'sabbatical'. I'd been forced to run Windows 2000 with Service Pack 2, due to an intolerable conflict that SP4 had with another critical app. During that sabbatical I managed to resolve the conflict and began using SP4. In November of last year, I then rejoined BOINC with the "same" system. Because of that small change, though, the earlier 2000 or so credits I earned with this system are treated as separate, though logically those results should be merged with the same system's current record. I've tried merging them myself, to no avail; apparently the mere switch from SP2 to SP4 was enough to con BOINC into thinking they're two distinct systems. Can someone on the BOINC/SETI staff investigate the feasibility of manually merging the two, when time permits? That might finally be worth a monetary donation, I think. :-) |
|
22)
Message boards :
Number crunching :
A64 optimizing: SSE2 vs. SSE3?
(Message 236625)
Posted 24 Jan 2006 by Mark A. Craig
Post: Ken said: Thanks Mark, you would think the 3700+ w/1MB cache would be a little faster than that over the 3000+. Should I go with the Athlon 64 SSE3 client? I tried running that one first and got some errors so I went with the SSE2. Maybe I should give the SSE3 versions another try. Ken: I'm jealous of that 1MB L2 cache (I bought cheap, figuring I'd save for an X2 later), but Saimek and his modest and mild-mannered but megacharged 3000+ is whacking the chips offa BOTH of our systems. Apparently we need to sign up for whatever overclocking class he's teaching? :-) Mark |
|
23)
Message boards :
Number crunching :
A64 optimizing: SSE2 vs. SSE3?
(Message 236621)
Posted 24 Jan 2006 by Mark A. Craig
Post: Saimek said: Vyper... no way... I've got my 3000+ clocked at 2.8 Ghz, memory 255(510 DDR) 1.5-2-2-5 timings and crunching 31 minutes average with Cruncher's client.. maybe his 3200+ have got 1 MB cache or what?! :D How'd you manage that? If the details aren't trade secrets, I'd like to learn how to duplicate that with my new 3000+. That is, if I can afford the required extras.... Mark |
|
24)
Message boards :
Number crunching :
A64 optimizing: SSE2 vs. SSE3?
(Message 236619)
Posted 24 Jan 2006 by Mark A. Craig
Post: Ken said: Hey everyone, I'm more of a hardware geek than a software or in this case client/performance geek and I was wondering if someone could look at my current stats and see if they're on par for my computer. I'm running 3 computers, but the one I'm mostly curious about is my Athlon 64 which I have running the SSE2. Here's the computer specs: Ken: I think there must be room for improvement. My 3000+ isn't overclocked and not water-cooled, but with the SSE3 app and client it's completing work in about 2900 seconds, versus about 2700 seconds for your 3700+. Mark |
|
25)
Message boards :
Number crunching :
A64 optimizing: SSE2 vs. SSE3?
(Message 235268)
Posted 21 Jan 2006 by Mark A. Craig
Post: Don wrote:Yeah, I have a 3200+ running at 3.72 Gig, and dual channel ddr running as ddr453, with 2-3-2-5 T1 timing. Just screams with the sse3 client. Check out: I think what Don was saying is that he locked his RAM at a fixed clock and then increased his FSB/HTT clock by 69%. I still find it pretty hard to believe he could get away with that without needing some alternative cooling, both for CPU and northbridge, to keep things from melting down. It seems incredible that I'm seeing a temperature of 46C with optimized code but stock speed of 1.8GHz, yet he can inject hydrazine into his and boost it by more than two thirds and see a temperature rise of only five degrees from that? He must be running it in a meat locker or a tank in a sperm bank.... Mark |
|
26)
Message boards :
Number crunching :
A64 optimizing: SSE2 vs. SSE3?
(Message 234892)
Posted 21 Jan 2006 by Mark A. Craig
Post: Don wrote: Yeah, I have a 3200+ running at 3.72 Gig, and dual channel ddr running as ddr453, with 2-3-2-5 T1 timing. Just screams with the sse3 client. Check out: You're using a multiplier of 16 instead of the default 11? How did you manage to unlock the CPU to accomplish that, considering that (I thought) all the Athlon64s are locked and the die and traces are bunkered under a metal shroud? I assume you had to employ equally extraordinary means to keep temperatures within safe limits? I've only tinkered idly with overclocking this new 3000+ Venice core, using the supplied ASUS/Nvidia utilities, but my attempts to push it beyond about 5% above normal had nasty consequences. Either this one isn't very tolerant of being pushed beyond rated specs, or my overclocking technique is flawed. Even without overclocking it, running optimized BOINC/SETI on it pushes the temperature up to 46-47C; a few degrees hotter and all sorts of bad things start to happen. Mark |
|
27)
Message boards :
Number crunching :
A64 optimizing: SSE2 vs. SSE3?
(Message 234791)
Posted 21 Jan 2006 by Mark A. Craig
Post: Jack said: You have a proper pair. That's what the British bloke said... to the waitress at Hooter's. :-) Mark |
|
28)
Message boards :
Number crunching :
A64 optimizing: SSE2 vs. SSE3?
(Message 234125)
Posted 20 Jan 2006 by Mark A. Craig
Post: Jack said: Tests on standard and real WU's show that the Crunch3r SSE3 on an Athlon 64 runs at least 2 minutes to maybe three minutes faster than the Crunch3r SSE2 does on the same system.... You might need a fast and optimized memory configuration to get the bigger performance improvement. Jack: Thanks for the concise and focused reply! How exactly would you define an optimized memory configuration? I have two Mushkin PC3200 256MB CL2 DDR400 dual channel (128-bit) SDRAM modules, a total of 512MB; eventually I'll swap that with the two Mushkin 512MB modules in my other system (or buy 1GB ones). The RAM timing right now is 2.0-2-2-6-11-16 (per CPU-Z). What would you suggest, if anything, to further optimize them? Mark |
|
29)
Message boards :
Number crunching :
A64 optimizing: SSE2 vs. SSE3?
(Message 233641)
Posted 19 Jan 2006 by Mark A. Craig
Post: My 3000+ is a Venice core, which thus allegedly supports SSE3. Mark I did exactly that, silly wabbit. That's how I refreshed my RAM that it was in fact a Venice core (the invoice also stated it but I'd forgotten). Of course CPU-Z also reported the revision as DH-E6 and claimed it supports SSE3, but something someone else said hinted that such a proclamation can't always be trusted, so I phrased my comment less authoritatively. Mark |
|
30)
Message boards :
Number crunching :
A64 optimizing: SSE2 vs. SSE3?
(Message 233429)
Posted 19 Jan 2006 by Mark A. Craig
Post: My 3000+ is a Venice core, which thus allegedly supports SSE3. Is it always the case, then, that a SETI app optimized for SSE3 will perform better/faster than one optimized for SSE2? I'm using CrunchR's A64 SSE3-optimized app and client now, so is there no reason at all to consider or experiment with the SSE2 versions? I ask out of complete ignorance, because I know absolutely nothing about the purposes or details of the SSE2 and SSE3 instruction sets, nor how they differ. I'm afraid to Google it for fear I might be up all night on yet another Google-abetted tangent... anyone have a Cliff's Notes synopsis? Mark |
|
31)
Message boards :
Number crunching :
Posters beware...if you're from the USA.
(Message 233250)
Posted 18 Jan 2006 by Mark A. Craig
Post: LINK What was it that Ben Franklin once said about security and liberty...and then there's George Orwell. So I guess people with Asperger's Syndrome are off the hook, then? ;-) Mark |
|
32)
Message boards :
Number crunching :
A64 optimizing: SSE2 vs. SSE3?
(Message 233248)
Posted 18 Jan 2006 by Mark A. Craig
Post: I now have an Athlon64-based system, and I intend to run SETI/BOINC on it. I figure to run an optimized app as well. Poking around a bit, I see that there are optimized A64 apps and clients for both SSE2 and SSE3 instruction sets. If one was clearly better than the other, why indeed would both exist? What criteria should I use to decide which one(s) will be most efficient on my Athlon64 3000+ system? I hope I don't have to resort to trial and error to figure it out? Mark |
|
33)
Message boards :
Number crunching :
Optimized BOINC and SETI software, specifically CrunchR's
(Message 230967)
Posted 14 Jan 2006 by Mark A. Craig
Post: BTW, to bring this whole thing full circle... Since I've fine-tuned my use of CrunchR's app and client, my Recent Average Credit has been steadily climbing; right now it's at 256 and still i'm waiting for it to plateau! Contrast that with my steady RAC values in the low 90's when I first re-joined SETI/BOINC in December. That's quite a remarkable increase. I also seem to be seeing less results returned where granted credit is less than claimed, and more often where it's more; if I understand the granting mechanism correctly, that might be because I'm now no longer consistently returning CPU times longer than the average, and so I'm benefitting from average now instead of suffering from it. My actual CPU times to complete units have also been cut by more than half, down to about 90 minutes now from over three hours in December. So, I've seen more than 200% increases in credits granted and work units completed per day. I guess this has now become just another "bubbly glowing effusive" testimonial for these optimized apps. Mark |
|
34)
Message boards :
Number crunching :
Optimized BOINC and SETI software, specifically CrunchR's
(Message 230964)
Posted 14 Jan 2006 by Mark A. Craig
Post: BTW SX or DX for those 486? Steve: Okay, so I see that Cyrix did in fact make 5x86 CPUs that didn't have an integrated FPU. However, the one I have is mounted on a generic 486 board that indeed does have a socket for a dedicated FPU, but it's redundant if the CPU already contains one; the the CPU in my instance is in fact a "DX2" that had an onboard FPU. (I'd dig it out again and get more details, but it went into storage almost two weeks ago.) However, Cyrix was also never known for producing efficient FPUs; would it have been possible to "upgrade" the onboard FPU with an external Intel or AMD FPU? If so, it should have resulted in better floating-point performance. Thinking back, I'm wondering if we're confusing an "Overdrive" socket on a '486 board with an FPU socket? Further, I've just done a bit of digging and found hints that there might have been more to the "SX" and "DX" designations than just the presence or absence of an onboard FPU; in at least some cases they seemed to also describe the data and address bus widths and maximum memory address space and cache sizes as well (or instead). How lucky the young newbs are that they've never had to worry about separate chips for floating point! They take its integration for granted now (hell, now we have CPUs with more than one FPU) but it weren't always so. Same goes for Level 2 cache... how lucky we are now to always have it inside the CPU rather than having 32-pin(?) DIP sockets taking up motherboard space! You do recall the definition of the term "DIP", right? Motherboards haven't been DIPpy in such a very long time. Hey, Right Said Fred had a tune "Deeply Dippy"... that musta been their geeky celebration of old tech, eh? :-) Mark |
|
35)
Message boards :
Number crunching :
Optimized BOINC and SETI software, specifically CrunchR's
(Message 229955)
Posted 12 Jan 2006 by Mark A. Craig
Post: Sad sad sad. Intel 386SX 33 MHZ here. Anyone still remember the days of the turbo button and the MHz display? I thought it was super cool when they got the display to read 100 finally. Gerald, you're just too young! You missed out on all the really fun stuff, like the debuts of the Apple and Tandy and TI-99a. And overclocking NEC V20/25/30 chips, back before CPUs even had heatsinks and fans. "Case fans? What are those? There's a fan inside the power supply, what else do ya need?" :-) Not only do I remember the "days of the turbo button", I still have three AT tower cases here that I'm now using to reassemble some of my dusty legacy parts into working stuff. All three have turbo buttons, and two have the two-digit LED speed displays... which, BTW, have to be programmed to show a specific speed with a tangle of pins and jumpers on the back of them (sure hope I can find the instruction sheets)! Ah, those were the days, when every card sported IRQ jumpers, Seagate 20MB hard drives were all the rage, having a "green" PC meant having one that could blank yer 15" amber CRT, and the revolution of the 9-pin RS-232 connector meant that everybody had at least one funky 25-to-9 adapter having off the back of the box that forced them to inch everything forward on the desk.... *sniffle* Mark |
|
36)
Message boards :
Number crunching :
Optimized BOINC and SETI software, specifically CrunchR's
(Message 229453)
Posted 11 Jan 2006 by Mark A. Craig
Post: BTW SX or DX for those 486? The one I seem to have, that Cyrix still stuck on one of the boards, is of course DX. I don't recall that Cyrix ever made an SX, though perhaps they should have, considering their reputation for sloooow FPUs. The boards themselves are 3rd-party generic and can handle either. |
|
37)
Message boards :
Number crunching :
Optimized BOINC and SETI software, specifically CrunchR's
(Message 229248)
Posted 11 Jan 2006 by Mark A. Craig
Post: You want my ancient Gigabyte board with the Athlon 900 I used with it? That should make up the difference. :-) I guess the boat is slow to get the new tech to you folks out there off the Mainland? ;-) Speaking of ancient tech, I have not one but two old 486 boards with slots for that cutting-edge VLB (VESA Local Bus, for you newbs) bus! One of them still has a Cyrix CPU on it. And VLB SCSI cards, Promise and Adaptec, to go with each. I also got me three Iwill motherboards, P54TS, P55TU, and P55XUB, with awesome embedded 20MBps Fast SCSI! Each one cost me about $450 back in the day. Now I realize why I'm still poor. 8-/ Oh, and truly ancient tech: an old/original Microsoft mouse, with a solid metal ball and BEARINGS instead of teflon slider pads! Mark |
|
38)
Message boards :
Number crunching :
Optimized BOINC and SETI software, specifically CrunchR's
(Message 229015)
Posted 10 Jan 2006 by Mark A. Craig
Post: Mark, just checked to see how your results are doing and you are not that much behind now, just a few hundred seconds. But then my times are not quite as good as they could be either since I'm running an FTP daemon as well as streaming music 24/7. Wish I could afford to have a second system running the other processes. You want my ancient Gigabyte board with the Athlon 900 I used with it? That should make up the difference. :-) |
|
39)
Message boards :
Number crunching :
Optimized BOINC and SETI software, specifically CrunchR's
(Message 228528)
Posted 9 Jan 2006 by Mark A. Craig
Post: Steve said: Yes my AGP is locked and at standard voltage Steve: Did some RAM refreshing :-) at MSI's Web site. The one big reason why I might never achieve your result is because of the one major difference between our two boards: yours uses the nForce2 Ultra 400, and mine just a plain ole nForce2 IGP. The original nForce2 was never rated for more than a 167Mhz clock/333MHz FSB, which is why even at just 358MHz mine is developing a serious fever (39C). MAYBE when I watercool it I can push it to 400MHz, but I recall that doing so has almost guaranteed side-effects with that chipset. Your Delta board has the Ultra, which was rated for 400MHz FSB, so you only had to fret over the health of your CPU whereas I get to fret over both.... Mark |
|
40)
Message boards :
Number crunching :
Optimized BOINC and SETI software, specifically CrunchR's
(Message 227528)
Posted 7 Jan 2006 by Mark A. Craig
Post: Now that Mark's problem is solved, why is my client claiming only around 6 credits per WU? That's exactly what has happened to me. The time spent on each workunit is now less than half what it used to be, but so also is the credit claimed (and mostly now exactly what's granted). If one is being fair-minded about it, this is exactly what one would want, assuming that credit is granted based on the "effort" a system expended. Isn't that right, that credit is claimed and granted based on the effort rather than the result? Since our systems have become more efficient, they're expending less effort to achieve the same result, so thus the credit granted is also less. Of course they're also now completing more of those results every day, so the net effect is at least equal. In the case of my system, though, I'm seeing a noticable rise in the "recent average" credit, from the 130s a week ago to 166 right now. Mark |
©2026 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.