• blah
  • blah

I'm still here
Yep, I'm alive. And things seem to be a bit happier with the site, although I still do not have FP extensions. I'm working on a work around for that. Other than that, I've been keeping busy with the little one, and casting a longing eye towards my airplanes, although I'd just as soon skip the chopper- I don't have the money to fix crashes at the moment. 

Got some bad news at work- I didn't get the help desk position that I applied for. Someone else was better qualified than me, apparently. Okay, I have no problem with that. Then I found out who. It's the guy who doesn't speak English. The same one who was asking me all sorts of stupid questions last week- "Can I use Outlook on AOL?" F*ck. I'm depressed. And the XML group decided to hire a couple of contract people, so I'm stuck in my boring old position of peon. Life is harsh.

But I digress. I do have new times for you tonight- but there's one little catch- you have to find them. That's right, I don't have time to go over them. Instead, I'm going to spend the evening going through my mail backlog. I was planning on doing that last night, but Devin (aka 'The little one') woke up and was hungry, and then wished to play. Until 4 AM. So I didn't get anything accomplished after all. And to make matters worse, my new motherboard didn't arrive today like it was supposed to. This is the one I ordered two weeks ago. Going to have to call them tomorrow. And I got a 10G SCSI drive on Ebay, but the seller wants to charge me twenty dollars to ship it! I mailed him about it, and his reply was 'I'm using Mail Boxes Etc, and they charge a premium. I'd go somewhere else, but it's too inconvient. Sorry'. Perhaps I'm in a complaining mood, but this guy is not making me happy.  Grrr..

Okay, other news related items. I'd like to apologize for the results table- I'm currently having all sorts of fun with it. When I cut and paste it in from Excel, it sets it's own width. HUGE. And to make matters worse, the HTML is bloated with table information. I'm not sure why this is so bad, it wasn't a problem before. Anyways, Guru is currently working on a solution for the issue, I hope you can bear it until then. Which brings me up to my last point. I'm currently being pulled in several directions at once. I'm spending more and more time at work, including weekends. I've got a limited amount of time outside of work when the baby is awake- which means that all that time is his. Whatever I have left is what the webpage gets. And since the baby is being more fussy in his sleeping as of late (He's teething), that leaves me no time at all. Therefore I'd like to post an ad again. 
Currently hiring
One halfway html proficient individual, must be able to do simple menial work such as update tables, discuss results, and attempt to be interesting (trust me- it's hard!). Must be familiar with computer hardware principles such as FSB, Latency, and why Cyrix sucks. Knowledge of diverse operating systems (NT, Win2K, Linux, MacOS) is a plus, but not a requirement.  This person will be responsible for the co-managed upkeep of a website with established reader base, and strict demands of reader base (see interesting comment). Benefits include- Work experience that may be helpful in a future interview. Then again, probably not. They didn't help me. You'll also be able to earn the recognition of a huge group of Seti users, most of whom will be happy to listen to you. Some will also send you naked pictures of small furry animals too- but that's what email filters are for. 
All interested parties should send their resume, scribbled on the back of a napkin, to me. Or else email me, and tell me about yourself. What do you know? What do you like to do? What's the most insane overclocking trick you ever tried? 

Okay- you're free to go now. Thanks for stopping by. -Rat

HUGE Weekend update
First of all, I'd like to apologize for those who have been having trouble accessing the site as of late. We're working with Nimbus Networks to improve the quality of our service, and I'm confident that everything will get straightened out, hopefully without too many teething problems. Secondly, I'm going to be canceling my account with Mediaone very quickly here, so there might be a time delay between when my submission page (which is still hosted there) switches over to this server. If there is, I've got a couple other options eyed, so hopefully the transition will be a seamless one. No promises though :)

I'd like to break a bit from the update status quo, and talk about some new things, rather than rehash how fast a 400 Celeron can finish the benchmark. I don't often get the time to do this, so I'm making the time today. The first thing I'd like  to discuss is the matter of memory latency. For the regular readers of this site, and of it's former incarnation, you're probably pretty aware of the effect of memory latency. This is because the Seti client's 'working set' is too large to fit in the L2 cache (high speed memory, versus main system memory) of anything that's not a Xeon or G3/4 (not including high end hardware like Sun). Let's take a look at this, using a typical PIII 600 for an example. 

The standard PIII 600 has a 512K L2 cache, which runs at half of the CPU speed, or 300 MHz.  When you start the Seti program, it has approx 1M of code that it uses over and over- this is the working set. You'll notice that the working set doesn't fit into the fast L2 cache. This means that the computer will have to swap the information in and out of the main system memory. So, your memory speed has just dropped from 300MHz down to 100MHz. And if the Seti client needs info that is in the main system memory, it will sit and wait for it, and not do any more work. And since a 600Mhz CPU being fed with 100Mhz memory will only recieve information from the memory once every 6 clock cycles, you can see why his is a huge slowdown when there is a cache miss (Needed code isn't in the cache).

But it's worse than that! Even if your CPU requests information, it won't get there in 6 clock cycles! That's because of latency, which is where the CAS 2 and CAS 3 stuff comes in. CAS stands for Column Access Strobe, but you don't need to know that, or even care. You can pretend it stands for Can't Access Seti. So, if your Can't Access Seti value is 2, your CPU will have to wait 2 clock cycles before the memory will even start to send information to the CPU. Therefore, you'll have 8 cycles pass between when the CPU asks for information, and when it shows up. That's alot of wasted time. Luckily for us, Seti doesn't have a cache miss each time it goes for information, but it does happen enough that we really do need to be concerned with latency of our memory.  This goes for the L2 also, as it also has latency, although it does operate at higher speeds than the main system memory, and therefore isn't able to give such dramatic boosts. 

So, where am I headed on this? Way, way back when, Max sent me a link for a cool program called AGPinfo. I myself haven't been able to test it, doesn't seem to run under Win2000. I don't know about NT- it might, it might not. But it does work under Win95 and Win98. Although it looks to be an AGP util from the name, it is actually very versatile, and has all sorts of cool memory tweakeage. It goes far beyond the level of adjustability that Bios gives, especially if you have a lame Bios. Max found that by changing the enabling the host bus fast data, setting all of the Dram signals to 2 (CAS 2), setting the EDO ras/cas to 1, and setting lead off command to 3, he was able to take a huge chunk out of his time. His 550MHz Celeron dropped about 30 minutes off his previous benchmark time- a very significant improvement. So for those looking to improve their memory performance, that is one possible place to start. Be forewarned though..since you have the ability to change alot of low level system variables, you may experience lockups! Then again, we've all had them trying to see how far we can push our machines in over clocking.

The next thing is very, very, very long. But it's well worth the read! I received a letter from Max, which included a copy of this. The next day, the original author contacted me and sent me a copy, which I'm going to simply post here. It'll save me alot of typing that way :)

Extreme short summary>>> What predicts SETI performance? Discovery, hurrah. No one runs this benchmark. No one wants to, not suppliers, not reviewers on their web pages. Want to do serious upgrade to 800 MHz board. Can't afford shot in dark. FINALLY one shop runs various combos for me, and comes up with THE ANSWER. The story, the shop, the person, and the answers follow...

Trying to keep this as concise as possible... If you want to get a motherboard that is fast on games AND maxed on SETI, you need to read all this. I am currently running 6 boxes 24/7 on SETI, up to 1208 WU last time I looked. Have a pair of PII 400's using BX based motherboards, a pair of K6III-400's running on VIA MVP3 chipset. I also have some earlier Pentium MMX chips overclocked and chugging along taking just under a day for a WU, all together doing 11 or so WU per day on average. In addition to SETI, I do some nasty number crunching with antenna design programs which are FP intensive, VERY large working sets in memory, and similar to SETI in response to the PC's resources. I kludged the K6's together as an inexpensive alternative to the PII boxes to based on some general benchmarks I had seen. The PII 400's do a WU in roughly 9 hours. I was disappointed when the K6's took 16 hours. Some applications I have (MS Word for one)actually run a little faster on the K6 boxes than the PII 400's, so it was hard to explain. None of the usual benchmarks downloadable would explain this LARGE disparity, indicating instead that the K6's should be doing something like the PII-400's. UNTIL I chanced to download SISoft Sandra, which has among it's various checkers and benchmarks something called Floating Point Memory Bandwidth. (No idea how they compute it.) Those of you who use SETIWATCH will know the %WU per hour figure. My PII-400's do about 11% and my K6III's do around 6%. A ratio of about 1.8, same as the 9 hours/WU and 16 hrs/WU. What were the SISoft Sandra Memory Bandwidth numbers? 245 for the PII-400's and 137 for K6III-400's, or a ratio of about 1.7. FINALLY something that would predict SETI performance. All of these were running at 400 with 100 FSB. Benchmarks on the slower machines indicated that there is a multiplier from the FSB and CPU speed that one needs to factor in to get the numbers to track on faster or slower machines. But I had a formula! Now, go shopping for a 700-800 mhz class MB and processor that gives the high numbers on SETI, great press on overclocked coppermine PIII 550's. I got the benchmark I need, everything. Right? Wrong. NOBODY that sells MB's & processors has the faintest idea of what I am talking about. Go direct to some of the overclocker's boards and similar ilk where everybody lives and breathes benchmarks, right? Wrong, the gamer's indexes and some of the general business figures are what gets run and published. NO SiSoft Sandra FP memory bandwith. What do I do? Beg, plead, threaten, throw tantrums over the phone, cannot get the ONE benchmark that will guarantee that what I get will ALSO max out SETI as well as Doom/Quake/etc. The only information around is that the venerable BX chipset on various MB's seems to be the fastest. Problems, though. At 133 FSB speeds, flakiness starts to emerge, and overclocker's problems start to emerge. Although many report success and stability -- and I believe them -- the shops won't GUARANTEE them in these modes. The top of the line commercial PIII's running on BX boards are using 100 FSB. Further, at 133 FSB, the AGP bus on a BX is in LALA land, since the BX was not designed for that and doesn't have the requisite dividers. The PCI does, but I would have to chuck my high end AGP video card for a PCI to use it.
Too much money for shots in the dark, too much money to be disappointed again like the K6III/MVP3 business. Some questionable motherboard implementations of otherwise decent chips lurk to foil me. How do I know if the PARTICULAR chip and MB I happen to get will run at the speeds reported by some other luckier soul? I NEED SOME FACTS !!!
LADY LUCK. Following yet another web link to yet another supplier, gets me Scott at Advanced Designs of Ky, http://www.advanceddesignky.com , who although skeptical of my story at first, sticks with me, ACTUALLY puts together some various boards and ACTUALLY RAN the SISoft Sandra memory bandwidth benches on newer stuff designed for 133 FSB, and on the venerable BX. Now we get to see who smoked who running Floating point intensive programs with huge memory working sets, like SETI, like antenna engineering programs, etc, NOT ONLY JUST the games. Some results, thanks to Scott..., all boards running around 700-800 and FSB around or slightly above 133, with 128 meg premium memory. All are overclocked 500 or 550E running at a speed burned in by ADK on that individual board, cpu & memory, which they will guarantee (meaning I have excellent expectation of stable 24/7 SETI gen at that speed, and if not I can talk to their tech support about it).

Floating point memory bandwidth: 

• Intel 820e chipset, 803 mhz 212 Mediocre press in general reviews, deserved
  
• Intel BX chipset, 733 mhz 260 The old master, great if no AGP & run ~ 700
  
• Via Apollo Pro, 750 mhz 158 Bad press deserved.
  
• Via Apollo Pro A, 781 mhz 282 Finally, the 133 fsb successor to BX!

The Apollo Pro A was on an Asus P3V4X which has full support for 133mhz, which means I can use my AGP card, which is good SLIGHTLY overcooked.
WHAT A STRUGGLE. It's ordered. Oh, the formula for SETI performance... this is still rough, and I don't know why it works. Need lots more figures to explain. SiSS FP memory bandwidth devided by 22. Then multiply by the square root of (FSB x CPU speed / 40,000) to get %WU per hour (V2.03 with the usual stuff straightened up.)

The formula says that particular P3V4X as served up by ADK should run WU in about 5 hours, possibly just a tiny bit less. ~ 5 per day. We shall see.

Hope all this is of some use to everyone. I sense a lot of frustration on these issues.

In further communication with Guy, I found out a couple of funny things. First of all, Advanced Design's webite has received about 1000 hits since Guy's letter was circulated. Furthermore, when he tried to call back, he found that all of the phone lines were busy. Hehe...  I'm pretty happy for the owner of the shop- it's not everyone who will take the time to build  4 different systems and test them for a potential customer. I think that says ALOT about how far they will go to make you happy. (I also noticed that their front page now has a Seti Crucher Special, consisting of a Asus P3V4X MB, a PIII 550, guaranteed to run at 781, a GW HS, and 128M of PC 133 for 600 dollars)

I'll be looking into Sissoft's Sandra test later today, time allowing. I think that we finally have a nice simple test that will allow us to make small tweaks and check them out in less than 5 hours. Cool! Perhaps this will allow us to further fine tune our systems, and open up a new world of tweakage. Then again, perhaps not. Only time will tell.

Okay, I'm sick of typing. I'll talk to you later!  -Rat

Misc ramblings
Hello, RB here again. Just to let you know, zAmboni is a terrible person. He went out and bought a PIII600E with thoughts of mad overclocking. Not to be outdone, I was forced to purchase a PIII550E, which is stable up to a FSB of 150, for a total speed of 825MHz. (That's what the previous owner ran it at, but I don't think I'll make it quite that high, need better memory first). This is, of course, all zAmboni's fault, and he should be ashamed of himself.  

Also, I won't be able to update the benchmark site tonight, even though I do have a few more times. I simply don't have the time to do it, I have some PC hardware to work on instead, and bills to pay as well. Yech. However, I should be able to update the times tomorrow night without fail. In the mean time, make sure you take a look at the team standings, it's good to see what the competition is up to! -Rat

I'm done playing games
At least until I finish the update. Then I'll play some more. I'll go ahead and start right in with the times that I have. Starting at the top, there's Steve, who whipped out the benchmark unit in a screaming 4:24 with his PII 550, which he overclocked all the way to 825 MHz courtesy of a 150MHz FSB. Damn that's fast! With his memory set at 2-3-2, Win98 and the Non Intel CLI, Steve is pretty well set. Action is doing well too, with his new toy, a Dual PII 400 Xeon. Xeon boards are not known for their overclocking abilities, mainly because they don't. However, Action has a 'turbo' option on his, which allows him to select at 103 FSB, giving him a 412MHz clock speed. Running a single client (he'll run two later) under NT4, and with his memory set to 2-2-2, he finished one unit in 4:51. While he didn't state it, I seem to recall that Action's Xeons have 1M L2 cache, which makes sense with his time - he's competing with 700MHz PIII's. 

Speaking of 733MHz PIII's, Turbo sent in another time today. He reran the benchmark to compare the effects of disabling the L2 ECC, and setting the memory's CAS latency to 2-2-2. His first time was at 773MHz with a 140FSB, which resulted in a time of 4:43. Clocking down to 733 allowed his memory to run stably at CAS 2, but his time of 4:59 isn't as good. While improving CAS latency is helpful, it looks like the 40MHz drop in speed negated any benefits. Tsai Ren-Zheng sent in another bench for his 558 Celeron, which is sporting a 124Mhz FSB, Win98, and the CLI client, along with a low 5:49 time, very good for a Celeron. Gotta love what a fast FSB will do for you.

Hydrobo! wanted to show that the Bios tweaks didn't help his 600 Athlon at all. So he turned on all the Bios caching on his Win98 powered machine, and ran the CLI through the benchmark again. The result? Umm.it was 2 minutes faster. Go figure..looks like the Athlon doesn't benefit from the Bios tweaks. 

Jetpants ran his PIII 525 through another benchmark, this time with the GUI and Win98. His time was teh same as the Win2000 time, 6:59. To be honest, this surprised me, Win2000 and NT have been slower for the 2.0 clients. Perhaps it was because Jetpants is running the GUI.  

Moving down the list a few hours, Wyvince sent in the time for his K6-2 500, which is running the Non Intel CLI client, along with Win98. His previous time of 12:28 wasn't much to his liking, so he disabled the on board sound, and dropped his time down to 11:44. And frankly, when you have a K6 CPU, you'll take anything you can get!

Last on the list is Chris Marble, who dug around in the bottom of a closet and managed to find an old HP 9000/705. With all sorts of weird things in it, like an OS that I'm not familar with, as well as the fact that it has 16.48 MB of ram (.48??), it's a Frankensteinien beast. And with a time of 105:39 for the benchmark, it's a slow monster too!

I've recieved a few times lately that had the wrong info in them. specifically, the time indicated was the wrong one. The results.sah file has multiple times in it, the one we are interesting in is towards the beginning. The string you're looking for is called cpu_time=xxxxxx.xxxxx, where the xxxxx.xxxxxx is the number you'll need to submit. Some people have sent in times that are actually their total CPU time for the project. Either that, or it took them 3000 hours to finish the benchmark! So, if you've submitted a benchmark, and you don't see it listed, that might be why. If you included your email address, I've already contacted you. If not, you'll have to resubmit the information. 

As for the site in the next couple of days, I'm going to work on getting all the proper FrontPage extensions loaded, as well as try and find out some more info about CGI scripting on this server. And I'll be playing more games...  -Rat

I didn't forget about you,
I'm just ignoring you. :) I'll update again tomorrow, but tonight is not a good time for me. I'm busy playing games. -Rat

Site Launch!
Well, it's finally here, the new redesigned look for the Seti Benchmarking site. And the new redesigned URL too! It's still rough around the edges, the reason for that being that I've been forced to move off of Mediaone's servers, and to do it as quickly as possible. This is due mainly to the fact that I'll be cancelling my service with them very shortly, but also due to reliability problems I've had with them, such as being unable to access the site, as well as general FTP issues. So, what can you expect to be different in the future? Well, not alot, and alot at the same time. 

First of all, I'll be providing you with the same information that I always have. You'll be able to check the latest news on speed tips for the Seti client, be aware of new clients, and be able to bench your system and compare it to others. However, the new server will allow some things I can't do now. The CGI capabilities will mean that a sortable results page will finally be a reality (no time frame on that yet), and I also hope to add other interesting features, such as user polls and free downloadable goodies. What sort of goodies? Umm...more baby pictures! For those who are sick of seeing the same darn picture over and over, I'm still trying to find Win2000 drivers for my SCSI card, which means my scanner is caput at the moment. I'll probably end up A) getting a new SCSI card (woohoo!) B) getting a new scanner C)throwing it on one of my Win98 boxes. Since I need (want) a new SMP workstation, I'll probably just throw it on a Win98 machine. Oh well...  -Rat

Site almost ready for launch
Yeah!! -Rat