|I'd wager that small discrepancy comes from the time between simulations, where there are more simulations occuring over the same amount of time. More simulations leads to a higher cumulative "non-production time" between simulations |
Another possibility is that I've noticed rapid particle time steps at the start of a simulation and a sharp drop in pts as the number of particles decrease substantially towards the end of a simulation (kind of the opposite of what I would expect).
|Hi All, Can anyone help me please? I downloaded muon1bench.exe,installed into my folder with all files, run muon1bench.exe.|
The DOS window shows _ Muon1Bench started. Interval = 300.0 sec
Call 'muno1bench 900' or similar from DOS to change the interval.
No change in file size.
I have been running muon1bench for the last 12 hours, though all I see is the above.
OS 98se. - 512megs ram - Asus P4PE - Intel Celaron 1.8.
|I assume you're running the bench in the same directory as the client itself. it'll change when the results.dat increases, at which point it'll show time = 0, and then it'll start incrimenting by 300's, until the next result added to the results.dat, when itll move to the next line and so on.|
[TA]z - yes, i've noticed as well that small numbers of particles are less efficient at producing timesteps/sec. I'm guessing its because therse less repeated calculations, anda slight time interval each time the next step is started, calulating the fieldstrengths or similar.
|Thanks K'Tetch, Yes in same directory, still no change. Now 27 hrs with Results.dat growing from 1.7 to 3.6 megs. Running PhaseRotC_bigS1 only with each run around six min to produce 29 to 35 Mpts/run.|
|that is odd. Wish my client wuold run that fast - I only manage 182mpts an hour, you're at least double that at 360.|
Failing muon1bench, there's always muon cockpit. I'm really liking this more and more as i use it. It includes a performance indiaction, and can monitor multiple machines on a network at once. It does, however, have the same drawbacks as muon1bench when it comes to sampleresults.
That's literally off the scale I'll redo the graph when we've got a few more new ones. Can't promise anything about the benchmarks otherwise...
|Dual Opteron 246 (SledgeHammer) @ 2.0 GHz, 200HTT, Windows Server x64:|
|Dual Intel Xeon 3.20GHz with HT enabled @ 387 kpts/sec|
|Now those have been added.|
|pretty, now I'll have to break 600 to make you redraw it again|
|try for a quad box, and break 1100|
|See last paragraph on here for X2 3800+ and 4800+ o/c reports.|
And what the heck is an "Althon"?
|I've definitely had the X2 3800+ over 2.7 without much of an increase in voltage, still breaking it in though. Hopefully I can show some 2.8+ benches soon|
|I can see a dual X2 3800+ CPU breaking all records. Maybe even try for a quad Opteron.|
|I didn't think you could put two X2 chips in MP configuration (unless you do a spldart-style hack somehow). You'd have to use two Opteron 270s to get the same effect, which unfortunately would be quite expensive.|
|I'd imagine there is no reason to slice the X2s even after the trouble of removing the IHS because there are no dual 939 boards and the pin structure of 940 being slightly different.|
I've seen a few quad dual-core Opteron workstations but haven't been lucky enough to bench muon1 on them What's really scary is the quad-core Opteron that AMD will be sampling soon O.o
|too bad we don't have a *ix version of muon anymore, I'd like to bench our new Sun Fire E6900 server |
With 8 Ultrasparc IV's and 16Gb ram i reckon it could crunch fairly effectively...
|100 individual pcs crunching will outdo even the fastest 4/8 CPU server. Maybe I should add more pcs to crunch but i too lazy to install muon in service mode and managing a whole horde of results.dat is a pain.|
|I have a pretty decent collection of programs for pushing out muon service installs/updates over a network but it's rather poorly designed and catered to my network right now. It would probably be pretty handy to some if I could find the time and energy to clean it up... Maybe I'll keep it to myself so kitsura doesn't out-crunch me with his 100+ pc farm|
|**looks at his lowly 5 1/2 year old dual p3-550**|
bahyou all make me god damned sick you know that!!!
Wish I had such machine farms nowadays.Then again, even with all those farms, i'm still outcyielding y'all
|Unfortunately or maybe fortunately for z, I don't have administrative rights to deploy anything over the network. But I do have full administrative rights to some 200+ P4 class PCs.|
|AMD Athlon64 X2 3800+ Manchester|
2000MHz @ 10x200 - 2-2-2-5-1T
475Kpts/Second (1710 Mpts/Hour)
All this according to Muon Cockpit (muonbench doesnt work for me )
I'll have another go at muonbench later, so the results can be included in the chart
|I've heard Muon Cockpit actually uses the modified timestamps on the files (rather than just sampling them periodically), so it's actually more accurate than my quick attempt at a benchmark program. Feel free to use that instead for these figures.|
|Nah, ignore those results - didn't let it run long enough.|
I've got muonbench working now, had it going for a few hours.
So far, same CPU @ 2400MHz (10x240) 2-3-3-6-1T is ~490kpts/sec.
I'd be interested in some more info on [TA]z's rig & settings, to compare against mine. I'm not sure what else I can tweak to make it faster.
Is Muon better with lower latencies or higher FSB or clockspeed?
CPU: Athlon64 X2 3800+
Clock Speed: 2500MHz
Core: E4 Manchester
FSB / RAM: 250/250
AVG Kpts/sec: 581.53
The average was calculated from the above list, with lowest & highest result removed.
Unfortunately, I think the memory isn't quite up to the task, and was actually causing it to go slower than it should due to errors. Muon crashed after the last entry in the list.
Look out Stephen, I'm gonna make you redraw the graph to suit 600+ scores
|at 2500/1T it looks like you are about the same as my X2 at 2600/2T... ~4% is about what I would guess for the 1T boost too. I miss my fast memory|
|Yeah, I wanna get faster RAM, maybe some good BH-5. The TCCD just doesn't cut it. I think my Corsair 3200XL (TCCD) isn't very good. I'm running it at 2.9V, and just cant get it stable. |
Dont really know what the CPU is capable of yet, but I dont think it's got much more.
I really wanna crack the 600 kpts/sec!!!!!
|same X2 @ 2800 (196 FS: 628 kpts/sec|
pretty sure the ram divider is having a considerable hit on this one though.
|An observation I've made on Corsair TCCD: increases in voltage tends to produce stability problems. I'd stay between 2.65/2.75|
Your CPU will probably push quite a bit higher clkfreq wise, and pushing high FSB @ 1T isn't always easy (especially on the 3200 rated stuff). I'd say drop the voltage .15 and go to 2T to squeeze the last mile of FSB out of your sticks.
Curious, what motherboard are you using?
|Thanks for the tips. Nice score BTW - I've got a way to go to catch ya|
I'm running a DFI Lanparty NF4 Ultra-D
I've just taken the TCCD out, and put in 2x 256MB Corsair 3500C2 (BH5) in, and I'm stability testing ATM using the overclocking guide on i4memory.com (Thx EVA2000)
I've been able to overclock the CPU better now, on lower voltage, so I'd say the TCCD was holding me way back.
Currently running @ 2560-8x320 with mem @ 232-2-2-2-5-1T using divider.
Now that I know what my RAM limit is, I'm gonna push the LDT, then see how much further the CPU can go.
It's starting to look quite promising so far - SuperPi to 8M no probs
EDIT: Found the CPU maxes out at around 2.65GHz. I'm back on the TCCD to see what I can make it do. Doesn't look too good tho. TCCD only seems to work upto 216Mhz @ 2-2-2-5, cant get it to work stable at all with looser timings
|Same motherboard here. Which bios and what slots are you running your memory in? I have to use the yellow slots exclusively to get any sort of stable overclock out of this thing. Have you tried disabling CPC (thus enabling 2T) to see if that produces any different results?|
I use 704-2bt for this board and to be honest I've come rather close to putting my fist through the thing more than once. Probably the most finicky board I've ever dealt with. What vcore are you running on the X2? (1.55 here, but I should probably back off a little)
Are you sure you mean 490 not 590? There's a big difference between that and your next post, which doesn't seem to be explained by a +10FSB increase.
 ...and here's the graph.
[edit2] You know, I've just noticed two distinctive things about this graph. One is how the top bunch are *exclusively* dual machines and how the single core ones form a sort of plateau below that in the 200s§. Also note how the Xeon is the dual machine with the worst scaling .
The second one is that I got a peculiar surprise when I took the ratio between [TA]z's 2x2.8GHz Athlon and my old P-II 400MHz. There's a factor of 7 in clock speed and 2 in number of processors, but the performance increase is 37.5 and not around 14 as you might expect, so the architecture (IPC) has come on a long way as well.
[§ Though I think an FX-57 could go above 300]
Ooops, didn't realise you had replied. Anyway, here goes:
1. Didn't bother trying 2T - too much of a performance hit....
2. I'm using 704-2bta, and was using Orange slots as recommended for TCCD. I've just now recieved the G.Skill BH5's and am testing CPU@2650 (10x265) with 183 divider giving me MEM speed of 240-2-2-2-6-1T. (cant get to 250 on 3.2v). Using Yellow slots now as recommended for low latency RAM (BH5 and such). I'll keep you posted once I have stable results.
3. Running vcore of 1.5v - not really willing to push more while using Zalman 7700-ALCU to cool it.
Burning in takes far too long for an impatient person LOL
STEPHEN: To be honest, I'm not really sure now. Anyways, I'll hopefully have a new result for you shortly
|OK, new results after a good few hours benchtesting:|
CPU: Athlon64 X2 3800+ @ 2650Mhz (10x265)
MEM: G.Skill F1-3200BWU2-1GBGH 1GB PC3200 Winbond BH5 (2x512M @ 240.9Mhz 2-2-2-6 (CPU/11 divider)
I've nearly got ya [TA]z - more burning in required.....
|very nice, now I'm gonna have to borrow those fast memory sticks I sold off recently|
|Opteron 146 @ 2764MHz|
did you rerun the P-II again with the latest version of muon? (do you still even have it?) I remember significant speed differences between the code used back then and the current code. I remember one version change reduced output by about 25% on my machines. this would mean the newer machines are even faster than it would appear.
edit: looking back, the change from v4.34 to 4.41 in may 2004 dropped mpts output by about 30%.
|Well, er, no I don't have that PC any more - actually I'm not quite sure what happened to it. I left it at my parents' house for disposal, though it might still be in their garage(!)|
So I suppose all we can conclude from that is that the IPC CPU factor improvement multiplied by the Muon1 efficiency improvement is about 2.7.
|From my main rig ,an XPM 2500 o/ced 2.5GHz ,178MHz FSB|
I estimated my average as 261.5 Kpts/s from this.(run for over 2 days,forgot about it ,lol)
Next to test will be my Sempron 3100 @2.39GHz
|Would be funny if the Sempron was faster, but it's entirely possible if I'm right in thinking the "3100" model is indeed an A64 derivative rather than an AXP.|
|That's just what I was thinking! ,& ironically it seems it is.|
Yes its derived from the Ath64 venice core ,single channel RAM & 256KB L2 cache instead.
Looking at the charts above it seems FSB counts alot for DPAD.
My Semp 3100 is running 265FSB & RAM at 217MHz ,so a lot higher than the XPM rig
Early results say its doing about 285 kpts/s ,that's faster than the P4s!
I'll post back proper results tomorrow ,when its been running for a while.
Oh ,inccidently ,the only reason I can't o/c it higher atm is because I'm being held back by a cheapo PSU ,it'll be interesting to see how it does if I can get 2.7Ghz out of it (once I eventually get a working 'decent' PSU!)
How did you get MC to give you that figure?
|Assimilator1 - Which Figure?|
|This one |
475Kpts/Second (1710 Mpts/Hour)
All this according to Muon Cockpit (muonbench doesnt work for me )
Final average for my Semp 3100 @2.39GHz is 286 Kpts/s ,taken from this.
|Sweet ,thanks for adding me|
Though I've just upped the clock speed on my 3100 to 2.51GHz ,279FSB/228RAM ,wanted 2.7GHz but I'll live with it .
Going to start benchmarking it now
|Latest benchmark for my Semp64 3100 @2.51GHz|
Btw this CPU has 256KB L2 cache & is the 'E' revision.
Uptime (secs),Mpts in file,Estimate kpts/sec
Looks like the average is 301 Kpts/s
|I'd like to report my benchmark on an Athlon Thuderbird 1400MHz, FSB 140, RAM 420.|
I am getting 148 Kpt/s.
|Anymore single core Opty or 1MB cache Ath64 owners scores?|
|Results from me:|
Pentium 3E 1GHz (Coppermine-T) - 65 Kpts/sec.
Pentium M 725 1,6 GHz (Dothan) - 131 Kpts/sec.
Intel Celeron 330 2,66@3,30 GHz - 195 Kpts/sec.
AMD Athlon 64 (Venice E6) 3200+@2,4 GHz (FSB 240 2,5-3-3-8-1T) - 285 Kpts/sec.
|Odd that your A64 Venice is basically level with the Sempron at the same speed on the chart. Maybe Muon1 really doesn't care about cache size?|