OK. And the 5 digit Intel Sspec of the klamath is?
I give it zero chance of actually messing up the FSB speed like that. So make that 2x66MHz. I have seen all kinds of silly FSB speed readouts, but when benchmarking it just turns out exactly like 66MHz. If one looks up the mainboard PLL chip datasheet one can see what FSB speeds are possible with that chip.

I wrote it's SL28L.

Yeah the memory performance scaling looked to me like it would be right where 2x66 should be. Feipoa thinks so too.

Oh.. yes you wrote it already. Thanks!

I've plotted out the PII Klamath at 133, 166, 200, 233, 266, and 300 MHz. 133 - 266 MHz are with a 66 MHz FSB, whereas 300 MHz was with a 100 MHz FSB. If the 133/5x CPU is running with a 66 MHz FSB, the memory read speed would, presumably, be linear with other CPU frequencies at 66 MHz FSB. And it is. Note the deviation from linearity with a 100 MHz FSB as theorised. I was going to measure the FSB with an osciloscope, but I think I am satasfied enough with the benchmark charts.

The Quake2 benchmark results do not appear to show this non-linearity with a 100 MHz FSB, but it does show a linear trend for 133 MHz on up.

One other observation is that the Quake 1 score at 640x480 was 9.9 fps, which was the score I observed for an Intel P54C-90 on a 430TX board. The Quake 1 score-to-frequency ratio doesn't come together until about 300 MHz. P55C-300 was 18.5 fps, PII Klamath 300 was 18.2 fps. The Klamath-133 scored 8.6 fps in Quake 2, which was between a P150 and a P166. Is this lack of Quake 1 performance chipset or Klamath related? Anyone else with a PCI Matrox G200 and a non-VIA slot1 board want to check this?

For additional reference, the Klamath PII-266 I used at 133 MHz with L2 enabled had s-spec SL269.

Download Fastvid and run the vspeed benchmark that comes with it. This could be a bus bottleneck affecting the video card in DOS. I don't know what to do about a VIA PCI bottleneck in DOS though if it is the case.

I have Fastvid bundled with 3dbench. I ran it, but I don't know how to determine if the results are due to a bottleneck. Using a Celeron 400A,
DRAM to banked VGA: 36.86 million bytes/sec
DRAM to linear framebuffer: 38.34 million bytes/sec.

I changed some settings in the BIOS to see if they had any affect, but they did not. I tried enabling,

PCI to DRAM Prefetch
Byte Merge
Delayed Transaction,

and setting,
DRAM Read Latch Delay to 0.0 ns.

I tried Quake 1 with and without these enabled, but no change.

The results given sound like those of a VSpeed run. I don't believe that 40MB/s PCI bandwidth can be a bottleneck for Quake 640x480 on a Pentium 2 133MHz.

Back when I ran my 133 MHz tests, the Pentium MMX and Pentium II were neck and neck in Quake at the default 320x200 resolution. Apparently that VIA Pentium 2 platform has some performance degradation somewhere. Have you tried using a different video card?

133 MHz? Those vspeed test results that I posted were with a Celeron 400A at 400 MHz, which is where I am in the testing right now. I didn't bother putting the PII-133 back in.

For comparison, a Geforce 6200 on a 440BX board w/PIII-850 gets,

37.6 million bytes/sec
49.77 million bytes/sec

for those two tests, respectively.

I threw together a testbed to try Quake.

Abit BF6 440BX
PII 5x=133 MHz
G200 8MB AGP
256MB SDRAM @ 2-2-2

Vspeed results without FastVid - banked=22.1 MB/s & lfb=24.0 MB/s
Vspeed results after FastVid - banked=64.5 MB/s & lfb=89.2 MB/s

tested in pure DOS 7 (Ctrl-F5 boot)
Quake 1.08 -nosound 640x480 timedemo demo1
w/o FastVid = 10.5 fps
w/ FastVid = 15.3 fps

So, video card / PCI bandwidth does play a role at 640x480.

5 FPS in Quake 1 is a big jump!

I decided to try Fastvid on my VIA slot 1 board with the Celeron 400. I had originally thought it was only for Intel 450/440 boards, but the VIA FastVid results jumped 3-fold.

Before FastVid:
DRAM to banked VGA: 36.86 million bytes/sec
DRAM to linear framebuffer: 38.34 million bytes/sec.

After FastVid:
DRAM to banked VGA: 105 million bytes/sec
DRAM to linear framebuffer: 101 million bytes/sec.

I did not enable FastVid on any of the benchmarks. At the time, I figured that was the fairer approach. Perhaps the 430TX board has some of these features enabled by default, whereas the VIA board does not?

Your PII-133 Quake 1 results are not too far off of mine without FastVid though. I observed 9.9 FPS, whereas you noted 10.5 FPS, albiet on an AGP bus compared to my PCI bus.

I checked vspeed with the Klamath 133 on the VIA board, here is what it gets.

Vspeed results without FastVid - banked=26.54 MB/s & lfb=29.49 MB/s
Vspeed results after FastVid - banked=72.62 MB/s & lfb=94.74 MB/s

a few questions & points

there was no production board made for P version Nexgen that included a FPU socket.
P versions came in 75, 80, 90, 100 & 110 ratings

600MHz Xeons are 133FSB.. only 600/100 are Engineering samples

what OS do you intend to use for Xeon benchmarks?

do you have full list of what CPUs you intend to eventually benchmark?
are CPUs like this on the list
VIA C3 Samuel & Samuel2
Intel Pentium ODs 63, 83, 120, 125, 133......

Hello Neon_WA and welcome to the forum! Your avatar is certainly recognisable from other PC-related forums.

I was keeping a list at one point, but it kept growing and changing, so I stopped updating it. I'll re-update a list near completion. A very outdated list can be found here,
Cyrix MII-433GP Build
There is not much point in commenting on this list as there has been many corrections, additions, and deletions to it since.

Thank you for the information on the Nexgen's, however it is unlikely those will be on the comparison. Nobody has stepped forward. There is also a possibility that the Xeons won't make it either. Engineering samples for Xeon 600/100 are perfectly acceptable for test. The Xeon OS would be Win98SE, to be fair with the other test platforms in this comparison.

Yes, I have sourced what I beleive to be every generation of VIA chip which will run at 600/6x MHz. A Samuel, Samuel-2, Ezra (maybe an Ezra-T), and Nehemiah. For the most part, I'm using S-spec and CPUID to identify each chip. The final chart will contain the S-spec, CPUID, operating frequency, and multiplier. The goal is to test as many Pentium-class CPUs <=600 as possible with as few motherboards as possible.

The tests are nearing completion. On my side of this work, I have only the Cyrix Media GX, Via C3, and a Slot A Athlon left to test.

On a side note, I'm still waiting for your test results of your Cyrix 5x86-133/4x, if it is stable, and what the CPUID is.

not that I have much time at the moment for testing...
I do have complete set of P2 & P3 Xeons and both 100 & 133FSB boards
Also I do have P version Nexgens & both VESA & PCI boards.. but no PF Nexgens
As for socket 4 I have 60 & 66 and both 120 & 133 OD

As for CPUid programs.. most with Samuel2 C3s & about half with Ezra give incorrect L2 cache info

will try and make time over the weekend to look at my 5x86 again

Thank you for interest in this project; please send me a PM when you have time and if you are still interested in testing. It takes about 3 hours per CPU and requires about 100 numbers to tally per CPU. You will need a Matrox G200 PCI w/16 MB RAM.

in the mean while I will check thru my 300+ video cards and check if I have one, otherwise I will source one so I have on hand

think I noticed that you didn't want a network card installed?
can I have one installed but disabled in Windows, so I can carry out some other testing after the bench-marking of each CPU?

Neon_WA wrote:

in the mean while I will check thru my 300+ video cards and check if I have one, otherwise I will source one so I have on hand

think I noticed that you didn't want a network card installed?
can I have one installed but disabled in Windows, so I can carry out some other testing after the bench-marking of each CPU?

What CPUs will you be covering?

I'll be able to cover:

P2 Xeon 400/2MB
P2 Xeon 450/2MB
P2 Xeon 450/512KB
P3 Xeon 550/1MB

Just need to get a decent deal on a PCI Matrox G200 with the 8MB upgrade and I'll be set.

Usually it is OK to simply disable USB and network cards in Windows Device Manager. For slow CPU's, like 486's, I noticed that when USB (and network?) cards were installed, they slowed down the benchmark scores, but the scores bumped back up when they were disabled in Windows. I believe the main scores which were affected were from Winbench CPUmark.

m1919, I personally haven't come across the Matrox G200 being sold with the upgrade module pre-installed. In the past, I've always had to source them seperately. They were plentiful on eBay earlier this year and sold for $5 shipped.

You should also use a Yamaha XG PCI sound card if possible. Sound is enabled for Windows benchmarks. Further instructions can be given before getting started. Due to the desire for consistency between testers, the testing can be somewhat particular. I didn't think so at first, but ask Luckybob!

feipoa wrote:

Usually it is OK to simply disable USB and network cards in Windows Device Manager. For slow CPU's, like 486's, I noticed that when USB and network cards were installed, they slowed down the benchmark scores, but the scores bumped back up when they were disabled in Windows. I believe the main scores which were affected were from Winbench CPUmark.

m1919, I personally haven't come across the Matrox G200 being sold with the upgrade module pre-installed. In the past, I've always had to source them seperately. They were plentiful on eBay earlier this year and sold for $5 shipped.

I thought I saw one 4-5 months ago, before I got my A-Trend board. It seems it's actually fairly challenging to find a G200 PCI for a decent price right now. At least I can try getting the upgrade module and the card will come later.

For the cost of a g200 you could lend me those chips... I'm looking for an excuse to spend a weekend in the garage. ^.^