VOGONS


Calling on all CPU's for the Ultimate 686 Benchmark Comparison

Topic actions

First post, by feipoa

User metadata
Rank l33t++
Rank
l33t++

In an attempt to make the Ultimate 686 Benchmark Comparison as complete as possible, I decided I'd like to include some obscure CPUs if I can find anyone willing to test them. For anyone who has these CPUs, a suitable motherboard, and is interested in collaborating, please come forth! Expect between 2-3 hours of commitment per CPU.

Unfortunately, the CPUs and motherboards for these processors have become too difficult and costly for me to acquire. Of the numerous members here, I figure someone must have these working systems. They are listed in order of preferance.

NexGen P80 – 75 MHz
NexGen P90 – 83 MHz
NexGen PF100 – 93 MHz
NexGen PF110 – 102 MHz
Or any NexGen system w/FPU

Pentium P5 – 60 MHz
Pentium P5 – 66 MHz

Intel Xeon PIII - 600 MHz
Intel PII-Dechutes - 450 MHz

For NexGen systems prior to the PF series, please have the mating NexGen FPU.

It is preferred to use a Matrox Millennium G200 graphics card and a Yamaha DS-XG sound card to be consistent with the other results, however use what you can.

I have attached a PDF template if anyone is interested in assisting with this project. Donations are also welcome.

Following is a list of CPU's benchmarked to-date. All the CPUs within a particular socket category have been tested on same motherboard, with the exception of the VIA Nehemiah.

SOCKET 3
Cyrix 5x86c-100 (50 x 2)
Cyrix 5x86c-120 (60 x 2)
Cyrix 5x86c-133 (66 x 2)
AMD X5-133 (33 x 4)
AMD X5-160 (40 x 4)
Intel DX4WB-133 (66 x 2)
Intel P24T-83 (33 x 2.5)
Intel P24T-100 (40 x 2.5)

SOCKET 7-GXM
Cyrix MediaGX-133 (33 x 4) - Simulated
Cyrix MediaGX-150 (30 x 5)
Cyrix MediaGX-166 (33 x 5)
Cyrix MediaGX-180 (30 x 6)
Cyrix MediaGX-200 (33 x 6)
Cyrix MediaGX-233 (33 x 7)
Cyrix MediaGX-266 (33 x 8)
Cyrix MediaGX-300 (33 x 9) - Simulated
Cyrix MediaGX-333 (33 x 10) - Simulated

SOCKET 7
Cyrix 6x86-75 (75 x 1)
Cyrix 6x86-83 (83 x 1)
Cyrix 6x86-100 (50 x 2)
Cyrix 6x86-110 (55 x 2)
Cyrix 6x86-120 (60 x 2)
Cyrix 6x86-133 (66 x 2)
Cyrix 6x86L-150 (75 x 2)
Cyrix 6x86MX-133 (66 x 2)
Cyrix 6x86MX-150 (60 x 2.5)
Cyrix 6x86MX-166 (66 x 2.5)
Cyrix 6x86MX-200 (66 x 3)
Cyrix 6x86MII-233 (66 x 3.5)
Cyrix 6x86MII-250 (83 x 3)
Cyrix 6x86MII-262 (75 x 3.5)
Cyrix 6x86MII-266 (66 x 4)
Cyrix 6x86MII-292 (83 x 3.5)
Cyrix 6x86MII-300 (75 x 4)
Cyrix 6x86MII-333 (83 x 4)
AMD K5-75 (50 x 1.5)
AMD K5-90 (60 x 1.5)
AMD K5-100 (66 x 1.5)
AMD K5-105 (60 x 1.75)
AMD K5-117 (66 x 1.5)
AMD K5-120 (60 x 2)
AMD K5-125 (83 x 1.5)
AMD K5-133 (66 x 2)
AMD K6-133 (66 x 2)
AMD K6-166 (66 x 2.5)
AMD K6-200 (66 x 3)
AMD K6-233 (66 x 3.5)
AMD K6-262 (75 x 3.5)
AMD K6-266 (66 x 4)
AMD K6-292 (83 x 3.5)
AMD K6-300 (66 x 4.5)
AMD K6-300 (75 x 4)
AMD K6-333 (83 x 4)
AMD K6-2-133 (66 x 2)
AMD K6-2-233 (66 x 3.5)
AMD K6-2-266 (66 x 4)
AMD K6-2-300 (75 x 4)
AMD K6-2-333 (83 x 4)
AMD K6-2+-133 (66 x 2)
AMD K6-2+-300 (75 x 4)
AMD K6-3+-133 (66 x 2)
AMD K6-3+-300 (75 x 4)
AMD K6-3+-333 (83 x 4)
AMD K6-3+-500 (83 x 6)
Intel P5-60 (60 x 1) - Simulated
Intel P5-66 (66 x 1) - Simulated
Intel P54C-75 (50 x 1.5)
Intel P54C-90 (60 x 1.5)
Intel P54C-100 (66 x 1.5)
Intel P54C-120 (60 x 2)
Intel P54C-133 (66 x 2)
Intel P54C-150 (60 x 2.5)
Intel P54C-166 (66 x 2.5)
Intel P54C-200 (66 x 3)
Intel P55C-133 (66 x 2)
Intel P55C-166 (66 x 2.5)
Intel P55C-200 (66 x 3)
Intel P55C-233 (66 x 3.5)
Intel P55C-262 (75 x 3.5)
Intel P55C-300 (75 x 4) - Simulated
Winchip C6-133 (66 x 2)
Winchip C6-200 (66 x 3)
Winchip2A-133 (50 x 2.66)
Winchip2A-133 (66 x 2) - Simulated
Winchip2A-166 (66 x 3.5)
Winchip2A-200 (66 x 3)
Winchip2A-225 (75 x 3)
Winchip2A-233 (66 x 3.5)
Winchip2A-240 (60 x 4)
Winchip2A-250 (83 x 3)
Winchip2A-262 (75 x 3.5)
Rise mP6-133 (66 x 2)
Rise mP6-166 (66 x 2.5)
Rise mP6-166 (83 x 2.5)
Rise mP6-200 (66 x 3)
Rise mP6-208 (83 x 2.5)
Rise mP6-233 (66 x 3.5) - Simulated

SUPER SOCKET 7
Cyrix 6x86MII-250 (100 x 2.5)
Cyrix 6x86MII-285 (85 x 3)
Cyrix 6x86MII-300 (75 x 4)
Cyrix 6x86MII-300 (100 x 3)
Cyrix 6x86MII-333 (83 x 4)
Cyrix 6x86MII-350 (100 x 3.5)
Cyrix 6x86MII-400 (100 x 4) - Simulated
AMD K6-300 (75 x 4)
AMD K6-300 (100 x 3)
AMD K6-2-300 (100 x 3)
AMD K6-2-350 (100 x 3.5)
AMD K6-2-400 (66 x 6)
AMD K6-2-400 (100 x 4)
AMD K6-2-450 (100 x 4.5)
AMD K6-2-500 (100 x 5)
AMD K6-2-550 (100 x 5.5)
AMD K6-2-600 (100 x 6)
AMD K6-2+-350 (100 x 3.5)
AMD K6-2+-400 (100 x 4)
AMD K6-2+-450 (100 x 4.5)
AMD K6-2+-500 (100 x 5)
AMD K6-2+-550 (100 x 5.5)
AMD K6-2+-600 (100 x 6)
AMD K6-3+-350 (100 x 3.5)
AMD K6-3+-400 (66 x 6)
AMD K6-3+-400 (100 x 4)
AMD K6-3+-450 (100 x 4.5)
AMD K6-3+-500 (83 x 6)
AMD K6-3+-500 (100 x 5)
AMD K6-3+-550 (100 x 5.5)
AMD K6-3+-600 (100 x 6)
Intel P55C-250 (100 x 2.5)
Intel P55C-262 (75 x 3.5)
Intel P55C-300 (100 x 3)
Winchip2A-233 (100 x 2.33)
Winchip2A-250 (100 x 2.5)
Rise mP6-200 (100 x 2)

SOCKET 8
P6-133 / 256k (66 x 2)
P6-150 / 256k (60 x 2.5)
P6-166 / 256k (66 x 2.5)
P6-180 / 256k (60 x 3)
P6-200 / 256k (66 x 3)
P6-210 / 256k (60 x 3.5)
P6-233 / 256k (66 x 3.5)
P6-133 / 1mb (66 x 2)
P6-166 / 1mb (66 x 2.5)
P6-200 / 1mb (66 x 3)
P6-233 / 1mb (66 x 3.5)
PIIOD-300 (60 x 5)
PIIOD-333 (66 x 5)

SLOT 1
PII-Klamath-133 (66 x 2)
PII-Klamath-166 (66 x 2.5)
PII-Klamath-200 (66 x 3)
PII-Klamath-233 (66 x 3.5)
PII-Klamath-266 (66 x 4)
PII-Klamath-300 (100 x 3)
PII-Dechutes-350 (100 x 3.5)
PII-Dechutes-400 (100 x 4)
PIII-Katmai-500 (100 x 5)
PIII-Katmai-550 (100 x 5.5)
PIII-Katmai-600 (100 x 6)
PIII-Coppermine-600 (100 x 6)
Celeron-Mendocino-300 (66 x 4.5)
Celeron-Mendocino-400 (66 x 6)
Celeron-Mendocino-450 (100 x 4.5)

SLOT 2

PII-Xeon-400 / 512k (100 x 4)
PII-Xeon-400 / 2mb (100 x 4)
PII-Xeon-450 / 512k (100 x 4.5)
PII-Xeon-450 / 2mb (100 x 4.5)
PIII-Xeon-500 / 512k (100 x 5)
PIII-Xeon-550 / 512k (100 x 5.5)

SOCKET 370
VIA C3-Samuel-550 (100 x 5.5)
VIA C3-Samuel-600 (100 x 6)
VIA C3-Samuel2-600 (100 x 6)
VIA C3-Ezra-600 (100 x 6)
VIA C3-Nehemiah-400 (66 x 6)
VIA C3-Nehemiah-600 (66 x 6)
VIA C3-Nehemiah-600(100 x 6)

SLOT A
AMD Athlon-500 (100 x 5)
AMD Athlon-600 (100 x 6)

Last edited by feipoa on 2012-11-28, 13:29. Edited 11 times in total.

Plan your life wisely, you'll be dead before you know it.

Reply 1 of 144, by luckybob

User metadata
Rank l33t
Rank
l33t

Dibbs on the pentium pro testing. I have a pci matrox g200 as well. ^.^ In addition, because i'm awesome, I will also do the 1mb cache version AND the overdrive. ^.^

It is a mistake to think you can solve any major problems just with potatoes.

Reply 2 of 144, by feipoa

User metadata
Rank l33t++
Rank
l33t++

@luckybob

Much appreciated! I listed the most important PPro frequencies, but if you feel like you haven't had enough benchmark torture, feel free to do more PPro frequencies.

To be consistent with my results, only use 64 MB of RAM. If you have multiple options, a motherboard with 512 KB of cache is preferred but not required. Also, please use Windows 98SE with the unofficial service pack. Ensure that no network cards are installed. EDIT: PPro's might not have L2 motherboard cache...

When doing the Windows benchmarks, I've noticed that two benchmark programs that are very cache/RAM sensitive should be run first, otherwise your results might change once Windows has had its fun with the available memory. Please run Winbench CPUmark32 and CPUmark99 first.

If you have the 8 MB memory upgrade card for the Matrox Millennium G200, that's even better. I used my G200 w/16 MB RAM.

You have the 333 MHz S8 overdrive? Even better! Can it also be downclocked to 133 MHz?

If you also have the 256 kb PPro, can you run the benchmarks at 133 MHz as well? I'm also doing a massive 133 MHz challenge and it would be nice to see the performance difference between 1024 and 256 KB cache.

My default Windows monitor resolution is 1280x1024, 16-bit colour, 60 Hz.

Only the Windows98SE-default DirectX 6.1 is installed.

Please use Matrox driver version 6.28. This is not the latest driver, but is the one I used and is the latest driver which would work with Cyrix 6x86MX CPUs.

Last edited by feipoa on 2012-07-24, 11:56. Edited 3 times in total.

Plan your life wisely, you'll be dead before you know it.

Reply 3 of 144, by feipoa

User metadata
Rank l33t++
Rank
l33t++

PDF template was updated to reflect added PII OD and 256 KB PPro at 133.
I'll send active participants the Excel spreadsheet to input data to.

Plan your life wisely, you'll be dead before you know it.

Reply 4 of 144, by nforce4max

User metadata
Rank l33t
Rank
l33t

You left out one Media GX. Before I had moved out of Dallas I had a Compaq system that had a Media GX 150BP as it was a gold capped bga model. I remember it being SLOW in win 9x. Sadly I lost the machine in storage.

On a far away planet reading your posts in the year 10,191.

Reply 5 of 144, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I purposely omitted the MediaGX 150. That is such an uncommon frequency that there wouldn't be a whole lot of other CPU's to frequency compare it with. 133/166/200/233/266/300 are most ideal for comparison, but if anyone has a 150 MHz GX, we can add that.

Does nobody have a P60/66 and is ready to step up to the challenge? Your acts of generocity will help preserve retro hardware performance history for decades to come.

Alternately, known good socket 4 motherboards can be donated to the cause and I'll run the numbers.

Plan your life wisely, you'll be dead before you know it.

Reply 6 of 144, by luckybob

User metadata
Rank l33t
Rank
l33t

If I had a nice socket 4 board, I would sit up and watch it all night... I havent even seen on on ePay for a while now. At least not one that is 10x overpriced from that a-hole in canada. you know the one.

ANYWHO... I jsut god done writing a term paper for my sister so I should start benchmarking tonight. I have the 1mb and 256kb pentium pro @ 200. they should run fine at all frequencies. I also have the 333 OD's however their multi's are locket to 5x. so 250, 300,333 are all that can be done. I'm not sure 250 is worth doing. but I will see.

It is a mistake to think you can solve any major problems just with potatoes.

Reply 7 of 144, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Fantastic!

I've attached an image of my Win98SE benching desktop. Try to run the first 4 programs in the order shown (top to bottom). Note that you are only using WinBench99 for the FPU Winmark.

250 MHz PII POD not terribly important. Let me know if you find any Klamath 233/266's. I cannot afford these eBay prices anymore. Anything I do buy I need to haggle over it for days with back and forth messages.

Plan your life wisely, you'll be dead before you know it.

Reply 8 of 144, by swaaye

User metadata
Rank l33t++
Rank
l33t++

If nobody else would like to, I can jump in with my PII 266 SL28L (unlocked Klamath).

-would be on a 440BX board.
-I have a Millennium G200 AGP but it only has the stock 8MB.
-my smallest stick of SDRAM is 128MB but it's possible to use a system.ini setting to restrict Windows 9x to a set amount. link

Reply 9 of 144, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Swaaye, much appreciated. If I cannot source an unlocked Klamath, you can run the ensuite at 133 MHz. Ideally, I'd like to run the Klamath myself because I would like all PII/PIII CPUs run on the same motherboard. I'm also not sure how much performance variation the G200 AGP has over the PCI version for the selected set of tests. One user here mentioned that he probably has an unlocked Klamath somewhere that he'd donate to me. Otherwise, my eBay spending limit for this CPU is $9 shipped. Activity on CPU-world has been surprisingly slow in the past few months, probably because everyone is on holiday.

My hopefully realistic goal is to have all raw data collection finished some time in September. I gather data on an empty chart in pencil, then put them into Excel every few days. I've noticed that in every 300 data points, I have 1 transcription error, so I may need to double check all values. These errors become known to me when synthesising data. The pencil data was correct though.

Since there were no takers on the P60/66 (yet!), I've started asking around. I know a non-vogoner guy who may have a board to unload, but he thinks it is in a heaping pile of computer hardware boxes...somewhere. I'm particularly interested if a P60/66's FPU woud outperform the X5-160 in synthetic benchmarks and in Quake 1, particularly because S4 boards tend to be overly sluggist haredware-wise.

Thanks for noting the S-spec. Is there a list of Klamath S-spec's for the 266 variant which are known to be unlocked?

Plan your life wisely, you'll be dead before you know it.

Reply 10 of 144, by swaaye

User metadata
Rank l33t++
Rank
l33t++

Regarding unlocked Klamath chips, I have not seen a list of known unlocked S codes. From what I've read, the 233 and 266MHz models are always unlocked.

But I have some bad news. I remembered that I did the 133 MHz testing on a Supermicro 440FX-based board that I had at the time. It had PCI bus problems so I have long since discarded it. My 440BX board will only run down to a 3x multiplier...

Reply 11 of 144, by kool kitty89

User metadata
Rank Member
Rank
Member
swaaye wrote:

Regarding unlocked Klamath chips, I have not seen a list of known unlocked S codes. From what I've read, the 233 and 266MHz models are always unlocked.

But I have some bad news. I remembered that I did the 133 MHz testing on a Supermicro 440FX-based board that I had at the time. It had PCI bus problems so I have long since discarded it. My 440BX board will only run down to a 3x multiplier...

I wonder if LX boards fare any better here.

Reply 12 of 144, by luckybob

User metadata
Rank l33t
Rank
l33t

are you modifying windows 98 registry for different L2 cache sizes? It defaults to 256kb, but with p2's having 512 and my p-pro having 1mb...

http://support.microsoft.com/kb/183063

When i rean 98, i ALWAYS made sure the value matched what I had installed.

It is a mistake to think you can solve any major problems just with potatoes.

Reply 13 of 144, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I didn't do anything do the Win98 cacheing scheme. I always thought this registry fix was only for WinNT. I have this hack done on my WinNT 4.0 computer with 512 KB cache, however I did not notice any performance improvement. As noted in the article, the improvement would only be 0.2%.

Plan your life wisely, you'll be dead before you know it.

Reply 15 of 144, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Haha, Microsoft seems to have contradicted itself here. I think the second link which says not to alter the L2 cache amout is refering to W2K and above, whereas the first link was refering to WinNT 4.0.

Plan your life wisely, you'll be dead before you know it.

Reply 16 of 144, by luckybob

User metadata
Rank l33t
Rank
l33t

Ok, while running my first benchmark, I got a little distracted and made a nice little batch file for all of Feipoa's benchmarks. If you want to have a look!

Its pretty useless without the benchmarks from Feiopa, if he gives me the ok, then I will re-release everything as a easy to use package!

Keep in mind, this was written at 4AM. >.<

It is a mistake to think you can solve any major problems just with potatoes.

Reply 17 of 144, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I'm not sure which of these benchmarks have issues with copyrights, but I would assume that Norton Symantec Sysinfo may. I don't have Landmark's copyright/readme information either. That said, I'm leaving the burdon of decision with you on this one.

The way I run the dos benchmarks is such that I put all files into one folder called "comp", which is short for comparison ("bench" is already used on my HD). This way, I do not need to keep changing folders. I already have all the DOS commands commited to memory. I run 3dbench2 last because this probram will send you back to C:\ when its finished.

I don't think this batch file will save much time because you still need to tally the numbers. What would be ideal would be a script which can parse the command console for the values of interest and insert them into a CSV file. For the pictoral results, like quake, pcpbench, etc, you'd need to do some kind of analogue image to digital number to extract the value. I'm not sure how pheasable this would be in DOS alone, perpaps with some kind of DOS capture card whereby the images are relayed into MATLAB or LABVIEW would be easier. I think speedsys also saves the raw text data to a txt file.

I can run the DOS bench suite in 15 minutes, the real time kill is the Windows benchmarks.

Plan your life wisely, you'll be dead before you know it.

Reply 18 of 144, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Updated to reflect the need for Slot 2 Xeon testers at 400 - 600 MHz (100 MHz FSB only).

Plan your life wisely, you'll be dead before you know it.

Reply 19 of 144, by luckybob

User metadata
Rank l33t
Rank
l33t

p2 & p3 xeons are ALL 100fsb. Except for the "overdrives" @ 933 & 1ghz. these were nothing more than a coppermine on a slot 2 board.

I'd be more than happy to run benchmarks on the xeons, but I dont have any processors save for my 900/2m/100's If someone here were to ship me some... (i'd return them if you wanted) If I could get a hold of a p2 and p3 engineering samples... I could change the multi's and test everything.

It is a mistake to think you can solve any major problems just with potatoes.