vetz wrote:

For Europeans, check this auction out: http://www.ebay.de/itm/Matrox-G200-G2-MSDP-8B … #ht_1432wt_1397

Hey this is the card that i won, good find. ! 😀

Mine is assembled in thailand.
Still looking for a memory upgrade.
Found in the usa for 8 dollar or so, but shipping is to expensive sadly.

JaNoZ wrote:

Mine is assembled in thailand.
Still looking for a memory upgrade.
Found in the usa for 8 dollar or so, but shipping is to expensive sadly.

Great! You are going to run the P60/66 right?

My testing is finished as of today. I've tested all I can for this.

Would anyone be willing to run these two extra quick tests:

AMD X5-160 and POD100
64 MB RAM
UMC motherboard with 2-1-1-1 L2 timing, 0ws/0ws RAM timing, HOST-to-PCI at 1:1
WinNT 4.0 (install Service Pack 6, IE6, and the DirectX6 patch)
Matrox driver 5.06 or 5.07
Quake II Timedemo1 - OpenGL
WinTune98 OpenGL test

While I ran these tests, I had to run the PCI at 1:2/3 and L2 had to be at 3-2-2-2, making the Quake II score only 0.1 fps faster than it did at 33 MHz.

I know I've done this before at 2-1-1-1 and 1:1 in NT4.0. I think the trick was to drop the cache down from 512 KB to 256 KB. For consistent results, you'd need to put L2 into Write-through mode.

Yes i will do the benchmarking, would be nice to do.
But i do not have the upgrade mem yet, and no YS yamaha sound card.

JaNoZ wrote:

Yes i will do the benchmarking, would be nice to do.
But i do not have the upgrade mem yet, and no YS yamaha sound card.

That's fine, just use any ole sound card. The soundcard probably has little affect on the results.

Neon_WA wrote:

Also I do have P version Nexgens & both VESA & PCI boards.. but no PF Nexgens

Has there been any progress on this?

finishing up on the p2 xeons tonight, jsut one more to run before I start on the few p3 xeons I have. Sorry for teh delay, but waht with thanksgiving and having my furnace go out its been one hell of a week. If all goes well, I should have p3's done through 550mhz this weekend.

luckybob, thanks, I have added your results to the massive spreadsheet file.

I have now included a list on the 1st page of this thread detailing all of which CPU configurations have been tested. 177 to date. Are there any post-486 CPU's less than 601 MHz which are missing (and of benching interest)? Well, I just thought of a few configurations which would have been of interest - Cyrix 6x86-60/66. This would at least give the original P5 someone to run clock-for-clock comparisons with. Cyrix 6x86's support 1x. It might also have been of interest to run the Cyrix 6x86 at 1x50 MHz to be the slowest CPU supported on the i430TX.

I am really hoping Janoz will get to the P60/66 and that Neon_WA will be able to test the NexGen's. There has been no testing to date.

Cool. On page 1, I might have a p2 450. Also, I listed all possible slot 2 xeons. Except for the 133 fsb ones. They were nothing more than regular p3's in a double size package.

Well, the Cascades PIII Xeon did have 100 MHz FSB as well as 1 MB and 2 MB cache options, perhaps not on 600 MHz unit though. So if there existed a 1 MB or 2 MB PIII Xeon-800 at 6x133, we could run it at 100x6 just to top up the 600 MHz benchmarks.

hi guys i checked my p5 socket4 boards they are both fic pm900's and 256kb cache, i can move 64 mb to it edo.
i have also an original p60 but the boards lacks any 60 mhz adjustment through jumpers.
so i will have to look for an 60mhz osc, i do have an 80mhz one.
but i dont think you would like me to test that.

also there are only 8 pcs of umc 257 15ns cache chips and the board should be set to wb policy as i can see in my manual.
how is it possible to use wb without any tag ram socket?

If it is direct-mapped cache, you'll want to set your L2 cache to write-through mode, otherwise you'll need to upgrade to 512 KB cache to leave the system in L2 write-back mode. If you use 64 MB of RAM and only have 256 kb cache in write-back mode, some Windows benchmarks scores will drop.

Is there an empty DIP socket for the TAG? Is there a TSOP TAG soldered-in on the board somewhere? Can you provide photos of the board and the manual? FIC has now password protected their ftp download site.

I have some pictures of one machine.
I will try to get the manual scanned tomorrow.

I will Put some 8x 512Kbits 15ns sISSI cache, that would be 512KB so can it cache 64MB? then.
Is there a big drawback with Write trough compared to WB mode? as benchmark performance wise, i think i sould go for 64MB WB mode, or do you like benches like they were originally intended, as the Mainboard was sold with only 256KB to begin with.

Btw why do modern Coppermine for example cache more with only 256K

Ideally, you'd use 512 KB of cache in write-back mode and 64 MB of EDO or FPM memory. Having 256 KB of cache in write-through mode and 64 MB of memory will lower a few benchmark scores by only a few percent. Having 256 KB of cache in write-back mode and 64 MB of memory will lower a few Windows benchmark scores by 30%.

After looking at your motherboard photos, I only see that you have DIP-28 sockets for the L2 cache, so 256 KB will be the max. I recommend running only 32 MB of RAM and write-back L2 cache for the DOS tests, with the exception of Quake1, which you should use 64 MB of RAM in write-through mode. For all Windows tests, use 64 MB of RAM in write-through mode.

Socket 3, 4, and 5 motherboards use direct-mapped SRAM cache. The Coppermine's and Tualatins used 8-way set associative cache. There are pros and cons to each cacheing strategy, which you can read up on here,

http://en.wikipedia.org/wiki/CPU_cache
http://www.ecs.umass.edu/ece/koren/architectu … e/tutorial.html
http://www.dauniv.ac.in/downloads/CArch_PPTs/ … vity%281%29.pdf

any updates?

I have taken over 15,000 data points by hand for this project. My goal was to finish data collection in 2012, which happened on schedule. A vast amount of disinterest seemed to set in sometime after completing the massive Excel sheet. The excitement will kick in again at some point this year, I'm sure.

For what I have in mind, it will be another massive undertaking to analyse all the data. This is complicated further because many cell values are dashes, have paranthesis, subscripts, etc. I will need to replicate the table and manually add a numerical value in all these unfriendly cells.

My excuse for now is that I am waiting for Janoz to tabulate the P60/66 data to compare with my synthesised P60/66 data. What is a 686 comparison without the original Pentium included?

I have resumed work on this project. Data is being normalised to that of a Pentium 233 MMX.

Can't wait 😀

Aside from the usual averaged charts for ALU, FPU, and Overall performance (refer to the Ultimate 486 Benchmark Comparison), which benchmark programs do you think deserve their own chart depecting the raw scores for each CPU? The CPUs tested are from 60-600 MHz, with the exception of NexGen CPUs.

I was thinking these might be the most interesting for stand-alone charts: Quake 1, Quake 2 (software mode), Quake 2 (OpenGL mode), Speedsys Score, Cachechk L1, 3DMark99Max, Passmark MMX, Winbench99 ALU, Winbench99 FPU, and SuperPi. I did not include 3Dbench, Doom, and Pcpbench because their scores saturate asymptotically with frequency, but I am open to suggestions.

The benchmarks programs utilised are as follows,

Symantec Sysinfo
PiDOS
Landmark
Bytemark
RLB Dhrystone
RLB Whetstone
Speedsys Score
Speedsys L1
Cachechk L1
3Dbench
Doom 1
Pcpbench
Quake 1

SuperPi (128K digits only)
Winbench96 CPUMark32
Winbench99 CPUMark99
Winbench99 FPU WinMark
3D Winbench97
WinTune98 (Integer, FP)
Sandra99 (ALU, FPU, ALU MMX, FPU MMX)
PassMark (Math Mark, Integer, FP, MMX)
3DMark99Max
Final Reality
MDK
Quake 2

my 2 cents? 3dmark and Quake 2. I think most people that stumble around are looking for a cpu for gaming, not for calculating Pi.

I'd stick to ID software engines.
They always used the most of systems, and show how a style of coding can follow or not CPU evolution.
Personnaly i'd really like Doom, Quake, Quake 2 and Quake 3 😁

Synthetic benchs on the other hand, mean IMHO nothing. It's good to know how fast your cache is, but a CPU is a whole.