• blah
  • blah

Something for the weekend, Sir...
No benchmark results this time just a little light reading... first a small addition to my words on bioses for Abit BP6's, BP6.com  is the place you need. In case you ever wondered how many times you can flash ROM, about 100,000 is the answer according to Ovonyx, somewhere near the bottom of the page. So now you know. While you are there the main article about OUM memory sounds interesting. Continuing on  the memory theme,  remember we had an RDRAM system time (5:12) from Charity's machine - disappointing was the verdict for this much touted 'faster memory'. There's a hardware RDRAM read at Anandtech which includes a potted memory history plus bandwidth & latency info. If you prefer the politics and power approach try Overclockers.com if you've got the time to give it a whirl.  Which is the best Coppermine for overclockers at Toms Hardware is one page of a PIII guide. A magnet of a title for most of you I would guess. It's two weeks old but I only just noticed (!) 'better late than now' I always say. Those of you who cherish your Coppermines will have some comments on this article.
Getting a little closer to Seti, honest. Diagnostic news WCPUID has gone to v2.7e-B4. Available at H.Oda's site. Want to know your bus and cpu speeds to 2 decimal places? This utility will tell you and far more. Small (90kb download) and neat it gives detailed info about your cache and cpu - a very useful little tool. Also while you are there have a look at the other tweak goodies - to alter L2 cache you need  WCPUL2 which is for Windows machines running Celerons, PII/PIII and Athlons. Bit tricky to get it to work for dual cpu's (end up hacking the registry) but straightforward (ish) for single cpu configurations. Some benchmarks suggest a 1 or 2% increase in speed but since for most of us Seti crunching times depend far more on Ram latency it's not going to help much. Changing L2 to a more conservative timing is probably most effective for increasing stability on OC'd systems. It's fun for control freaks but don't expect any real Seti benefit. Also there is softFSB (113Kb) which saves you going into bios to change the bus speed. Not for the faint hearted, prepare to lock - well I guess that's nothing new to most of us.

Great news on the recent surge past Intel, I think the stats' say it more eloquently than I ever could. As for zAmboni's overclocking and the space time continuum here's something almost as bizarre - 'server runs on potatoes'. It made it to Slashdot and the BBC amongst others but since removal only lies in memoriam at the Register. Goodnight, Max out.

Quick Addendum
Roelof contacted me today to point something out that I should have mentioned yesterday regarding upgrading to Xeons. While the current version does have a huge decrease in time with the larger cache, the 3.0 client (I dunno when it will be out. Nobody really does yet) will have a new FFT routine which is supposed to be more cache friendly. So perhaps with the 3.0 client, the only key will be raw FPU speed, instead of FSB speed or monstrous cache sizes. So if you're contemplating an upgrade to a dual Xeon simply because it will crunch Seti like nothing else in a reasonable price range, perhaps you should consider that it might not always be the fastest thing around. But it will be right now.. -Rat

Bits, Pieces, Bios & Benchmarks.

A few words on happenings in hardware and bios...
The Register has an article on Xeon pricing coming down (good) to corporate customers (ie not joe public yet - bad) that will eventually lead to those desirable beasties approaching our ability to pay for them. Plus Xeons at 700MHz (800 in August) with 2MB L2 sounds most yummy. The thread has picked up on this and chewed it over long before I could get my lazy carcass to scribe these words. So there's another reason for regularly keeping up with the postings there (up to page 30 as of last viewing). Though zAmboni receiving a nautical title caused hearts to flutter - fame being the oppressor of innocence.

Probably of interest to many of you, ABit have released new bioses for several of their boards. Info as to which ones is on the Ars Technica front page (23/5). Though for those of you with BP6's it's just another yawn, 'new bios, must be Thursday'.  To those not in the know the BP6 has had numerous versions, blink and you might have missed LP, NJ (beta & final), QQ (1 and 2 and beta), RU (beta and final) and that's not to mention the Highpoint 366 drivers/bios combos!! So for those of you who get a bios every six months or so spare a thought for dual cpu-junkies at the edge of madness. Just keeping up is hard enough let alone actually seeing any improvement.

Most of you will be aware of Xanrel's problems with officialdom, well commented on by zAmboni, Rat and a host of others. There is a whole thread devoted to 'crime and punishment'. Others have commented admirably (sorry for the pun) so I will take the pragmatic approach and suggest that what's needed is a reverse benchmark - how fast can you remove every client and clear all traces of Seti on the machines you run it on. Create that quick uninstall disk now. Keep one under your pillow, in your wallet and tucked in the hollowed out pages of the Dos 6.2 bible you have 2nd drawer down on the left. Hide one behind the newly-framed Interplanetary Society certificate that you gleefully obtained and personalised a few day ago. Just be prepared for trouble or better still be legit.

So on to the most recent batch of benchmarks. A number of people sent corrections and observations, thanks to you all for keeping me honest and humble. I hope I have finally woken up to the beauty of talaktalan's combinations and why the 686 client is not optimised for the 6x86 Cyrix. Also ploughing the Cyrix trail zack0 used an MII optimizer and knocked 23mins off to bring in a solid 22:17 with the NonIntel Client. A second bench with the 2.4 Win client produced 22:38 and his comments say it far better than I could. Litespeed ran 2 benchmarks on a K6-2 with a Win32 client (16:11) and then the NonIntel version (16: 32). The NonIntel seems to have a perversity all of its own, some get better results some don't and it has little to do with being a non-Intel cpu. Another time for Litespeed of 14:25 using CAS2 ram and Redhat puts it in the upper half  K6-2 submissions. Stepping up the processor tree Ian Downham and K6-3 managed a 10:41 which is probably 45mins shy of where it should be. Remember Kevin Carpenter who ran a dual benchmark on a 100FSB BP6, well he upped the bus to 106 and improved his time from 10:01 to 9:30 (5.2%). For a comparison he also repeated the single benchmark at 106 which improved from 6:50 to 6:37 (3.2%). Higher bus speeds help dual systems more than single ones it appears, untangling two clients on one bus is more useful than throwing data faster at a single client. A relative newcomer to these listings is Spook whose 7:31 is bang in the middle of the PIII 500 times. Another new name (to me) is kea whose time of  7:01 is pretty good considering other apps. were running. Sir Loin managed to run a 124FSB with a 3.5 multiplier PII giving a very commendable 6:42 (only Spiffy has a faster PII setup). Quick timeout.

When it comes to championing low multipliers Roelof is your man so this is a good point to let him say few words about the (under)performance of the CeleryII...
"Your comment about Sofa King's "poor" 7:03 (May 17th News) on his brand spanking new Celery II 800@100 got me thinking. If you translate 7:03 at 800MHz to SETI Spy's CpF number you get 7.05x800/555 = 10.16 CpF. If you look at my graph of CpF vs Multiplier, you will see that a Coppermine with an 8x multiplier is supposed to give about 11 CpF. But then values in the graph have been inflated by 10% to account for gaussians, so the 11 CpF in the graph corresponds to 10 CpF on the benchmark WU. Mmm... It seems like for the same multiplier the Celermine (as some like to call the Celeron II) performs just as well (actually, just as badly) as its bigger brother, the Coppermine! High multipliers are lethal for Seti performance! Just a little example for Coppermines at 100MHz FSB, based on the numbers in my graph.
500@100 is 7.00 CpF equates to time of  7:46,   600@100  is 8.33 CpF equates to time of  7:42
700@100 is 9.66 CpF equates to time of  7:39,   800@100  is 11.0 CpF equates to time of  7:38
Amazing, but true!  300MHz increase in speed yields an 8 minute reduction in time! This shows that for Coppermines (and Celermines), WU time depends almost exclusively on FSB speed! So the only hope for decent performance is to crank up that FSB... For example, if you can get hold of some 150MHz SDRAM you reported on, overclock your FSB to180MHz and get that 600EB (4.5 multiplier) up to 810 MHz (provided that the rest of your system is happy) you have a shot at 5.66 CpF and 3.88 hours (3:53) at only 10 MHz more than the 800@100." (minor editing).
Only a couple of times left to mention, Shirty and 504MHz PIII are in at 6:30. Why? - you guessed it low multiplier (4.5) high bus (112), lots of  nominally faster Cpu's in this bracket (6-7) hours but a faster bus will often be the quicker machine. Finally I will leave you with another Ultra Sparc II time from Hamlet with a 6 minute reduction to 2:27. Most pleasant, I will sleep happy tonight. Right that's a wrap. Bench those aliens, Max out.

Xeons, and why they make sense
I find it amusing that Max mentioned the pricing on Xeon systems this evening, as I was just going to share some comments on this. I've reached my max limit of computers at home, as I can't justify having more than 5. Feeling the need for more power, the only viable solution is to upgrade. I have a Cyrix P200 that's currently chugging away one unit every 36 hours or so, and is also doubling as a web server. Since this is by far and away the slowest system I have, it would make sense to start the upgrades with this one. Since the case is an old AT case, I planned to upgrade the case anyways, but the rest of the components can be re-used. So the options where fairly simple ones, I could use an existing Celeron processor (I have a couple spares), upgrade it to another PIII coppermine, build a SMP PII/III system, or go all out and get a Xeon. Upgrading to a Celeron system would obviously be the cheapest (roughly 170 for MB and case), and I would probably be able to crank out about 3 units per day or so. Not bad. 
Going up a step to an overclocked PIII will make the work unit load increase to around four, but the price would jump up to around $400, since I don't have a spare PIII. Nor will the memory that I have run above 117 and still stay CAS2, so that limits my performance. The third option is a good jump in price, being a dual PII/III system. Assuming I purchased a dual 500Mhz PIII and managed to overclock it slightly (like 560) , I should be able to get roughly 6 units per day, perhaps a touch less. The  price for the two PIII 500's alone would be $400, plus another 225 or so for a motherboard and case. 
The last option, and what I think is the best, is a dual Xeon. I've been spending some time looking at Xeon's on EBay, and I've learned a few things. First of all, don't buy the motherboard from an auction. I've found that most of the boards are ALR (Gateway) overstock, will only run at 400Mhz, and will cost you around $250. If you find a SuperMicro, you can run anything up to a 550, and expect to pay more like $350. So I did a quick check on Pricewatch- and discovered that Epox has an SMP Xeon board that will handle 550 Xeons, and includes the terminator needed if you only run one CPU (most on EBay don't, so you'll have to buy one for another $30 or so). Furthermore, being a new board, it has this really handy thing known as a warranty. And the best part is the price. How's 230 sound? You could pay that much for a PII SMP board from a company like SuperMicro! Next is the Xeons themselves. Make sure that you buy 1M Xeons, unless you happen to get a good deal somewhere on a pair of 2M Xeons. They won't run any faster with more cache, but if it's a good deal...  Anyways, I checked pricewatch again and found that you can get Xeon PII 400's with 1M cache for 125 each. That's less than the price of a standard PII 400!  So for a total price of less than $600, you can get 8 units per day, easy. And if you can find a good deal on a 450, 500 or 550 Xeon on Ebay (there are good deals to be had if you're patient), your output will only go up.

So to sum it all up, I think that instead of building that new PIII system, perhaps you should build that new PII Xeon. If you don't have the cash for the whole system, that's no big deal, you can always start with the board and 1 CPU, and buy the second later on when you have the cash. And with the Epox board, you don't even have to worry about running ECC SDRAM, FP SDRAM, or any other 'interesting' memory that you would with some other systems, it will run just fine on PC100. There aren't any overclocking options, but it's still a better deal than an overclocked PIII.  -Rat

Ask a simple question...
I read all the mailings that come my way and a couple of things appearing recently are worth reprinting...
"...if anybody can explain to my why, when I upgraded from a PII-450 to a PIII-866, my WU time stayed the same (about 12 hours), I'd love to hear it". Innocently wrote Nick Owens in a Seti mail list. One of the answers was from Guy 'Memory' Olinger and I thought since upgrade-itus often results in such apparent contradictions the reply might illuminate. You might have to read twice to follow the analogies or maybe I'm just not very imaginative. 
"Suppose you had a team race where your time was the sum of when the first and the LAST member of your team crossed the finish line. Then you gave the faster, stronger guy new lighter track shoes and Rocket Sledd vitamins, and then made the slower, weaker guy carry a bookbag full of rocks. What happens? Silly? No...
Good news: You went from 1/2 speed cache at 225 to full speed cache at 866, and over double the FPU speed. (Rocket Sledd Vitamins).
Bad news: SETI has a 1 meg or so working set, which means SETI is a memory starved application, and the increased FPU will be buried in memory access. (Why it's a TEAM race, cache AND FSB accessed memory).
Worse news: You went from 512k cache to 256k cache which means that the slow regular memory access across the FSB went from 512k up to 768k and pretty well buried any other increases. (Bookbag full of rocks).
If you care: Games and other small memory working set applications probably took off like a rocket ship on the new setup. If you are not (or can't) run CAS 2 (or 2-2-2) for your FSB speed, then set it or get some CAS 2 memory. It will make a difference on SETI. If you can increase your FSB some, that too, but then you are into overclocking and all the stuff that goes with it. Your times indicate that you maybe still are running some useless stuff from OS startup. I'd expect something more like 8-10 hrs. See the tips on the benchmarking web page. Guy."  Like your style Guy but go easy on the 'normals'.

Guy is obviously on a roll as another recent posting on the LambChop Mail List gave a more numerical analysis of FSB/L2cache cost/benefit amplifying his comments above. The numbers used all hinge on the fact that the Seti clients largest FFT (fast fourier transform, an amazing bit of maths but a little out of our depth) data set is  ~1MB. This is borne out by the fact that once you reach Xeon's with 1MB L2 cache, adding more cache makes almost no difference. What Guy has done is to remove the L2 cache size from 1MB to leave a remainder, in kb, that the FSB has to shift and ram has to store. There are some big assumptions here about relative speeds and efficiencies of  ram v. cache and CAS settings are ignored, but the overall methodology hangs fairly well. Also note that 'relative time' is just a number that allows a comparison between one systems cache & fsb against another, though as it turns out the 'reltimes' are in the right area to be taken as hours as well. Finally the contant factors of  MHz and Kb have been omitted as necessary. Over to you Guy.
"Let's put the two concepts together, cache speed and FSB and see how the combo's work out. There are some real surprises...
Method of calculation to get a relative time based on the access to 1 MB of memory. For Katmai chips, that's 512k cache and 512k across the FSB and for Coppermine PIII 550E that's 256k cache and 768k across the FSB. C=cache, M=size of Seti FFT in ram. First 550 Coppermine @ 100
timeC = (1/speedC) x memC for the Cache  we get timeC = (1/550) x 256 = 0.465
timeM = (1/FSB) x memM for the SDRAM  we get timeM = (1/100) x 768 = 7.680
reltime = timeC + timeM   total = 8.145
for a 400 Katmai @ 100 we get ...
timeC = (1/200) x 512 = 2.560 (cache at 1/2 speed)
timeM = (1/100) x 512 = 5.120 for a total = 7.62

Surprising? Look how the 550E times are completely dominated by the SDRAM time, .465 in the cache and 7.68 out on the FSB. No wonder that times for the Coppermines seem to be related to FSB, CAS setting, SDRAM quality, and seemingly nothing else. The greatest benefit of the cache is not so much it's speed, but how much it keeps you off the FSB. Let's make a little table of different CPU/FSB choices based on the formula above. You're gonna love this...The math is easy to verify.
CpuS/FSB  cache  c.speed   reltime    Comments
550E/100     256kb       fs       8.145     Stock PIII Coppermine
400K/100     512kb       hs      7.620     Stock PIII Katmai
616E/112     256kb       fs       7.273     OC'd 550E
500K/100     512kb       hs      7.168     Stock PIII Katmai
733E/133     256kb       fs       6.472     OC'd 550E 
550K/112     512kb       hs      6.433     OC'd 500K (on Soyo SY6BB baby AT mobo)
866EB/133   256kb       fs       6.365     Stock 133 Coppermine <added for Nick Owens>
533B/133     512kb       hs      5.770     Stock PIII Katmai. Remember these?
600B/133     512kb       hs      5.557     Stock PIII Katmai
560B/140     512kb       hs      5.486     OC'd 533B
600B/150     512kb       hs      5.120     OC'd 533B
450X/100     1meg        fs       2.275     Stock Xeon 1mb cache zero use of FSB
550X/100     1meg        fs       1.861     Stock Xeon 1mb cache zero use of FSB

Oh, and did you notice that the numbers above bear a striking resemblance to best times for various combos on the benchmark page. (Heh, Heh, Heh...) That's best times. Regards,  Guy." (some minor editing - Max)
Since the only way you can change your on-chip cache is to buy another processor those in thrall to consumerism, sorry faster Seti times, would do well to bear these numbers in mind before coughing up. Until later, Max out.

The Department of Snappy Titles is resting at present...
Well I'm back from France after a  week supporting some ten and eleven year olds at a watersports centre. Excellent. But you have come here for numbers and pertinent comments not my post-vacation drivel. I try to leave the human interest stuff and tyre-talk (see mid April news) to RB! First a little general news...

The thread has been going overtime about the ramifications and morality of running Seti on kit you don't own or have admin privileges for. A whole heap of views and thoughts, though thankfully no rabid nutters installing clients on all and sundry and claiming that it's a basic freedom NRA style. Even those who have done questionable things are at least thinking about the situation. Good stuff if you like a long read as the thread is up to page 22 in it's present incarnation. What's the limit before it goes off somewhere else without telling anyone and you (i.e. me) wait 2 days wondering why it isn't being updated? New version of SetiSpy is v2.4.4.2 for those who use it, apparently quite a few if you look at roelof's counter at the bottom of his page. normally I get an email update notification from roelof and he mentions it on a few newsletters which also come my way, but not this time apparently. So I will give it a plug it as it's so useful. Also his creation gets a mention because it gave me a pleasant surprise this morning. Those who have been 'congatulated' know what I'm talking about and those who don't have it yet to come. It would spoil the fun by telling.
As most of you Seti crunchers overclock you might find this interesting - "Kingmax will soon launch the PC-150 SDRAM using Tiny BGA DRAM Modules. The new PC-150 SDRAM use the normal 168pin DIMM and can support 150Mhz FSB. This is indeed a great news for all overclockers who plan to get their system running stabily over 150Mhz FSB when their PC-133 SDRAM can't achieve. If a PC-133 can achieve 150Mhz+, that means this new PC-150 SDRAM may achieve 170Mhz+." Thanks to Rizen (a graphics info site) for that snippet by way of AGN hardware. Before getting into the numerous times that are backed up due to sloth and absence on my part there are a couple of points to make to people who I don't have emails for: Gary resubmit but use the benchmark wu,  thanks. Team Delios - a tremendous time (3:36) but I need more info on your hardware before I can accept it. mail me or Rat. As always errors and omissions can be rectified by a reasonably pleasant email to Max. Failing pleasant it must be funny. Now on with times and crimes.

I like people with slow kit as their times seem to possess a certain dignity due to numerical size alone. So wehringer  at 23:55 fits the bill nicely. Unless you are able to slightly OC your PPro 200, try NT4 or lower the memory timings in Bios otherwise there's not a lot to be done. Next up is zackO who according to Rat would be classed as a 'suckee' (Quick Note- I never stated that zackO sucks. Just Cyrix sucks. -Rat), a 22:40 from your Cyrix is par for this cpu and at least gives a WU a day (assuming no nasty time-consuming Gaussians) as long as you never use your machine for anything else, ever. Continuing on a theme talaktalan sent in another 4 times which, forgive me if I have missed something, seem to show that slow Cyrix (Cyrii?) and ramdisks were made for each other. Using MuLinux gave 20:29 and then a surprising 18:52 using a ramdisk - a worthwhile lump off. Haven't seen this kind of improvement from a ramdisk anywhere else in a well set up system. As if to agree, talaktalan's K6-3 @400 machine only improved times by 10mins from 9:36 to 9:26 by using a ramdisk and even that is more than might be expected. Keep investigating. Now on to the Celeron appreciation section...
The rodent has been busy as well. Rat Bastard has popped up with several times packed with meaning (wish he'd tell me what the meaning was). For starters a reasonable 10:11 on 400@66 Celeron looks pretty damn good compared to 13:23 on a 450@100, perhaps the translation goes along the lines of  'Via Apollo133 chipsets suck...' if indeed that was the only difference. (Indeed, that was the only real difference. -Rat)  Next offering is 11:44 for a dual 533 Celery and W2K spitting out 2 wu's. Production-wise useful, but nowhere to go as already using CAS2 ram and increasing the fsb results in the dreaded NT Blue Screen of Death.  Bus contention is Overlord in the kingdom of 128Kb cache. Pithy or what? Finally a spanking 5:23 from an overclocked Coppermine which with a little luck might get you a sub 5 with some work fsb wise. So easy to be smug from the bunker where I live.

Sofa King gives us a first look at the CeleryII but even OC'd to 800MHz returns a poor 7:03, there's something seriously amiss there and very little is down to Hardware Dr in the background. Should be looking to nudge under 6 at least... keep trying. Orange Kid and PIII come in at 6:45 which again seems a little on the slow side, screaming to be overclocked. Forty minutes quicker at 6:02 is Lysator with similar system but running W98 with minimum graphics. I don't think anyone has shown ' 32bit 1024x768 = slow' and '256 640x480 = quick'. Unless you play games when benching (!) or run a UT gamescript as screensaver, your cherished PCI/AGP graphics card is doing zilch and takes next to nothing in cpu resources. Beyond's Athlon produced a solid 5:43 but it's noticeable how Athlons just don't oveclock very well being heavily optimised already.

Now we are getting to the sharp end and Chad E. Cowan serves up a 4:58 on a 750 PIII, any way you can push the bus on that will give big returns. A recent regular, Steve Common cracked the 5 with a 4:53 and then blitzed that with a 4:24. Comments reveal that he dropped ram latency back to 3 and hit a monster 160 fsb to get the quicker time. One of the rare occasions where higher fsb more than compensates for worse CAS settings. Steve A in similar vein posted a 4:31 on a PIII 880 (getting closer to 4 figures) but was forced to run at CAS3 to get stability and also used a 160 fsb. At present they are tied for second place for top fsb title, Tomslik and 166 still rule. That'll change though with the roll out of PC150 ram in the next few months. Both Steve's use Abit boards (with 1MHz increment softmenu II bios) which with PIII's have a lot to commend them to Seti crunchers if a Xeon is out of reach. Speaking of Xeon's jelmore ran a bench wu on each cpu of a Quad Gateway to post 4:05. Even though cranking 4 at a time is very satisfying, you must be a little miffed not to get into the 3 hour bracket. Still a superb time. Finally another rare beast puts in an appearance. A very swift 2:33 from Hamlet using an UltraSPARC II running Solaris 8. At this level of price and performance time is rarely available for non-business pursuits. A parting shot to all you all  speed worshippers, your fast times are only given meaning by far slower times. Respect.
I think that wraps it up for this evening, keep benching, stay legal, make LambChop clean and lean. Max out.

Boring Commentary
I just thought I would take a moment to mention a couple of items. While Sofa King's times seem very very slow (and they are) for the speed of the processor, the Celeron II has looked to be somewhat disappointing in certain areas, such as the L2 cache implentation. For those not familiar with the Celeron II, Intel has basically taken a Coppermine CPU, set the multiplier incredibly high (it's still intended for a 66MHz FSB), and castrated the L2 cache by disabling half of the 256K onboard, and adjusting the latency to 2, while the Coppermine has a latency of 0. Add to this the lower FSB setting, and it's no surprise that the Celeron II doesn't perform as well as a Coppermine in this instance. It'll still work great in Quake III though- and it's still alot cheaper! One last thing to take into effect is this graph that I yanked from Firing Squad, which shows the FPU memory bandwidth in relation to a 300A. You'll notice that while the speed has increased to 195% of the 300a, the memory transfer rate is only up 24%. 

Secondly, Guy brought to my attention that the Soyo SY-6VCA has a new bios rev out which allows for a factor 4 intleaving on the mem- and shows some very impressive numbers on the FPU Bandwidth test. All the good on that can be found here.

And on a happy side note, the telco came by today and fixed the problem with my DSL, so I don't have to visit the terrible scary world of old crusty ISA 28.8 modems anymore. And with that, I'll wrap up this evening's update, play with my son some more (Boy the get big fast, I can't believe he's 6 months old already), and then start working on that Linux server I've been meaning to play with.  -Rat

New times!
That's right, the results table has been updated with some new times. Max went through the times that had been submitted, and cleaned everything up. He'll come up with all sorts of witty comments to dazzle you with as soon as he figures out where he placed his wits. :)  As for myself, I spent the day playing with my son, since I was 'sick' today. Again. Did I mention I love my job? However, don't worry, I'm going to work on a few other site related items right now, so don't despair. I've a few more times to add to the mix, so don't worry if your time isn't in the table right now- I'm working on it. Also, congratulations to the site, which managed a record breaking 1005 hits yesterday! Not too bad for a bunch of people who are just wasting valuable bandwidth. 
Speaking of which, the telco is supposed to be by tomorrow to get my DSL fixed. No more 28.8 for me. Yea! -Rat

Got a pulse?
Well, we do. But we had to check. Not to worry though, everything is fine and dandy here. I've been away and sorta out of touch, and Max has just been away. However, we do have more times, and we will be updating those. I just didn't have time, and Max hasn't either. I'll try and do them tonight, if I have time to do it between filling out applications. Did I mention I got pushed off of the top of the list for the help desk position I've been waiting for? Time to go to another company, I don't see myself going anywhere with the one I'm at now. Even if I have been able to get a few work units out of them.  :)   -Rat

Cyrix, Linux, and why they suck
I realize that statement will upset a few people. So, now that I have your attention, please put your axes down and listen to everything I have to say. 

First up is njpoz, who's run the 2.4 CLI on his PIII 540 (120 FSB). His 5:57 time is 2 minutes slower than his previous time with the 2.0 CLI. To be honest, I don't think 2 minutes is enough of an improvement to count for much- but it is faster. And faster is better! Lucien tested his PIII, which is running at 500MHz. Also running Win98, but with the 2.0 CLI, he finished in 7:33. The last of the 'normal' tests come from Joker, who is running a Celeron 300a at 450Mhz under Win98, finishing the test run in 8:18. 

Now onto the weird stuff. Talaktalan not only has the most difficult name to pronounce, but he's got some funny hardware too. He's got a Cyrix 6X86 CPU which he's run in quite a few different configs. He started with a 250MHz run under Win98 with the 2.4 CLI client, using a 83MHz FSB. 19 hours and 4 minutes later, it was done. After 19 hours and 11 minutes, the NonIntel CLI was done. Then he clocked the CPU up to 263, and dropped the FSB down to 75. the times went up slightly to 19:31, less than I would have expected. Then in an attempt to evoke pity, he clocked the system to 120 Mhz with a 60Mhz FSB- which took 43:20. Ouch. Even slower was the run at 100MHz, with a 100MHz FSB. With the FSB and the CPU speed matching, it took the cache out of the way, but you can only do so much at 100MHz. That, and you can only do so much with a Cyrix CPU. I believe Cyrix actually holds the rights to the slowest FPU in the world short of a 386.  That's why the run took 47 hours and 22 minutes! Lastly, Talaktalan ran the system under BeOS with the 250/83 setting used in Win98, and finished about two hours behind the Win98 times with a 21:51. And that is why Cyrix sucks. 

And that leaves Linux, and why it sucks. Too be honest, Linux doesn't suck, it's awesome. However, Guru proved that the Linux client sucks. He ran two tests on his BP6, which is running a pair of 366 Celerons running at 550. With the i686 dynamic client, the test finished in 7:24. Then Guru got creative and ran the same test again, but this time he ran the Win32 CLI in wine, which is a Windows Emulator. This allows you to run Windows programs in Linux, but you do take a performance hit. And Guru's performance hit was- a decrease in time to 6:35! For those long time readers, you might recall that the Win32 client is optimized for Intel hardware, and it's the only client that's been optimized. That's why Windows has replaced Linux as the fastest OS for Seti, and also why the Windows client is faster running through an emulator than the Linux client is running natively. 

On a side note, Max is on vacation this week, so I'll be taking back over on the page duties while he's gone. I've a few more benchmarks to add to the mix myself, as well as a couple side articles I'm working on it the back burner, so keep checking back for new info. -Rat

RDRAM sucks official!! (does Bill Clinton know?)
First off a time from Charity which is interesting on two counts. One it's the fastest CPU clock, an 866 PIII, so far submitted, which Intel and AMD would have you believe is the only number that counts. Good job the truth isn't out there. Secondly and more meaningful the memory is PC800 RDRAM. I know that a certain rodent was squeaking in anticipation of a time for this new memory spec. and lo it appears. Rat assures me that it's very expensive with monster bandwidth (like the old Meatloaf) and high CAS latency (like 4 or 5, not good as we know). But unfortunately this tempting machine only coughed up a disappointing 5:12. Far slower than normally aspirated BX kit with slower CPU's. So boys and girls, pricey rare RDRAM is not the way to get just yet. In the words of the Bastard "looks like I'll have to take out my 'jury is still out on RDRAM' comment on the speed tips and simply state that it sucks." Can't get clearer than that. 
Another person who always provides the goods is Redbeard who's posted a PIII 700 time of 5:09. Might sound okay but I'll let him explain... "For a few months I have been itching to test out one of these VIA 133A motherboards since I heard they didn't stink quite as bad as the 133 when it came to memory bandwidth. I recently picked up the Asus incarnation and slapped it into my box with otherwise the same original hardware. I attempted to make all memory timings and BIOS settings match but as you probably know some of the options are quite different. To sum up the results the SiSoft Sandra 2000 Pro memory benchmarks were around 410 CPU/450 FPU (MB/sec) for the BX setup with my overclocked P3 500E at 700 MHz while the VIA 133A came in a disappointing second at 310/350... I'll continue to try a few tweaks but I don't think VIA will be challenging BX as the best mainstream Pentium chipset for SETI crunching anytime soon." So there you succinctly have it.
zAmboni jumps up the results table with a 5:06 by using the financial tweak - new PIII to replace the Celery. I was seriously pleased to stuff your Celeron time some months back but your way out of reach now, damn. A sub five, middle 4 awaits if you can push that fsb up and/or make all 2's on your memory (or buy a Xeon). God speed John Glenn. What am I on tonight?  Onwards and beyond who's 748MHz OC Athlon time of 5:53 is just more proof that Athlons with big big numbers just don't Seti that well. But consolation in that it's always nice to crack the hour. You say you're using a standard h/s & fan setup, brave man,  OCing is well dodgy unless you live in a fridge, so more cooling is advised. Quattro3's (= mono12?)  PII bench of 7:55 is fair to middling but I hope it's a comparison job as the using the CLI will make a big dent in that GUI (!) time. Look forward to the next one.To round off tonights offerings Haji slips in a 6:57 on an NT4/PII box. Cutting it a bit close there with 64MB, only one system above you runs in less than 128MB, which being Linux can get away with it.

Considering the extended Berkeley server outages recently I imagine some of you are catching up on your benchmarks. I guess times will be poring in over the next few days. A little unfortunate as I'm off to France for a week so it could be a little slow on the updates for a while. Rat will do his best but he's only one person amongst all this madness so go easy on him. Keep the faith and tweak till you bleed. Max out.

The Bunny has left the building...
Fancy a little whimsy! Remember the Registers Seti competition, of course you do - perhaps you even sent in a time or three to get a crack at that mouth-watering prize list. Well the excitement has bubbled over the edge of apathy and into results. If you want a little amusement check it out. I don't think there is anything there I could quote without appearing horribly smug, so I will let you scan it minus my gurgles of delight. Now you know and can sleep easy until the next jovial test of silicon appears. Mind you I could do with a laptop with a sub 5:30 output and the Spectrum time (~298 yrs) would make a great addition to the bench results! As for the guy using Seti to determine server speed, well there's a hundred or so more imaginative excuses peppering the Ars Seti thread. Perhaps I will dredge the archives and compile a Sys. Admin. top ten list of "reasons to run Seti on every box I can find". It's easy to scoff and good to remember that there really is a purpose to crunching wu's beyond locating "the signal". Thanks to the Register for that minor interlude.

So how about a few new bench times after all that excitement...gagging for them I bet. There are some quick times by Celery's in the table so for Thunder to post a 6:30 on a 595 Celeron might not seem anything special until you look at the fsb. If a bus speed of 86MHz (i.e. multiplier of 7, so an OC'd 466) is to be believed this is a storming benchmark. I was going to use the word "staggering" but you just never know when someone is going to come up with something even more impressive.  Running NT4 goes a small way to explaining the time but I'd love to know what other tweaks that piece of kit has. A very good PII Xeon time (4:52) from Geordie needs to be compared  with Actions PII Xeon time of a minute faster but running 40MHz slower. NT4 again showing it's merits compared to W2K as a Seti OS. Onto Wildcat and a 5:52 for a PIII 600, thanks for that but you have a long way to go, it's a low 5hr machine, so see what you can do. Coming in amongst the pack is Xanrel's W2K Advanced Server with a 7:54. Finally for this update is Vspace who's stomped all over the K6 fraternity with an 8:25. Way to go dude.

Thanks to Steve for pointing out he is the un-Common one and I look forward to that 600 OC to 900 time! That'll be one to tell the grandchildren, "I was there, you could read mobo markings by the red glow of roasting processor". I digress, that Register article has obviously removed me from the perch. I would also like to mention a slight adjustment to some of the times in the table. I've slipped this little bombshell in at the end hoping that most of you suffer terminal word fatigue after a paragraph. I have calculated them in a slightly different fashion with the outcome that some have changed by +/- 1 minute. Not great but I'm sure some of you will be howling in anguish especially as your precious 4:59 becomes a 5. This is because I noticed that the Excel function was rounding times incorrectly in some cases. So I've put a stop to that with a bigger, better formula. As a bonus the new setup can handle 10^10secs, just in case that ZX Spectrum shows up! If you think your time has altered more than a minute let me know and I'll check it again. Otherwise nothing major to worry your sweaty bench-furrowed brows. Keep crunching. Max out

Sunday Tuesday supplement with Mail & Times...
Thanks to Kevin Carpenter  for running  a one client bench on his dual celeron system and then a two client run. Many of you know what's coming but a little repetition can help. Looking simplistically you might expect two processors with a client each to run just as fast as a single processor with one client. Snag is of course that they have to share system resources. Here the bottleneck is once again the memory bus to ram as the Celerons small onboard cache means that ram is used extensively to store the Seti FFT working set. One processor alone crunches faster than the ability of the bus deliver data to it, (hence importance of high fsb), so with two procs it becomes even more problematic. His time rose from 6:49 for one to 10:01 for two. This is due almost entirely to the conflict between the processors to use the fsb. On higher spec systems with more L2 cache (ie less use of ram) and higher fsb (faster ram access  when used) more cpu's just mean more wu's processed in the same time (nearly). At the top of the results table multi-processor machines lose little in completion times whether running one client alone or one on each available processor, look at 4:07 v 4:11 from Action. Of course if what you want is maximum units completed then of course run as many client instances as you have processors but if the vanity of a fast average completion time is your bag stick to one client on anything less than a 1MB Xeon Dual setup. I think you get my drift. 

Kevin is also the guy who setup a mail list  for TeamLambChop which is rapidly turning into a 'Ask Rat about Processors' forum. So if you want to posit a view or leave a query about your system chances are that you'll get a quick comment  from  the illegitimate rodent post haste. Also some hardcore nitty gritty about mobo's going on recently which is valuable to those who need it and just general background for those who don't. Even when you can't follow what they are on about it's useful to impress people at all those Seti parties (that's a sad thought) you attend. Slightly off track, so what's new - where I live Kevin (in popular mythology) is a gerbil.

Other people other times...Spiffy, I must concur with your comment especially as your 5:59 client time is pretty damn good for a PII. That 4x 129 bus speed sure helps. Remember Roelof's musings on low multipliers recently. There's a lot of PIII and Athlon boxes in your wake. Nice to see themartins3 workhorse, a 233MMX, posting a respectable 18:57. Remember it's a 66 bus before you cast withering glances. Another Mac cracks the 6hr mark with 'trstick' sending in a 5:57. Good time but there's a little more to get out of that box yet. Steve Common deserves some praise for knocking a chunk off to get to 5:01 with his Coppermine, must be gnawing away at his soul not to make a sub-five. Perhaps you should ask Faust for some tips on bartering with the Devil. As I have mentioned before, the 4 hour bracket is highly competitive so saving best till last,  Steve (another?) has produced a seriously OCed time of 4:18. This is the quickest non-Xeon PIII so far, but sub 4 looks just out of reach. Only a handful of processors ahead of this one, congrats. Max out. 

Rambling from a Bastard
Hello, you'll notice that we're back up and running again- which is good. The Seti servers are also back up again- which is good. I have the day off of work tomorrow- which is good. So I'm having a good night (morning?) overall. Some of you may have noticed a slightly increased level of activity on the benchmarking side of things in the last couple of weeks. I'd just like to take this time to state that it has nothing to do with the fact that I removed CounterStrike from my PC a couple weeks ago. Just like zAmboni's recent flurry of inactivity is totally unrelated to his rececent CounterStrike addiction.
Anyway, I seem to have forgotten what the point of my ramblings were. Damn. Oh..somebody wanted Max's e mail address, but I don't remember who it was. In anycase, his mail address has now been listed. Umm...think that's all for now. Enjoy. -Rat