Wow.. Look what this thread has turned into! And to think... I just wanted to build a few socket 3 systems to play some old games. I love VOGONS!

Since there is so much hype for the ET4000/W32P I want to present my benchmark results. I finally got my hands on a Et4000/W32p card without giving a kidney. They back up the ones done by Artex.

Hardware:
Pentium Overdrive 83@100mhz
ASUS VL/I-486SV2GX4 Rev. 2.0
32MB RAM (FPM)
1024kb L2 cache
Samsung IDE harddrive

S3 Vision 864 2MB (#9 GXE64)
3DBench2: 96,6
PCPBench: 27,8 FPS
Doom: 1325 realticks, 56.37 FPS
Quake: 23,6 FPS

ET4000/W32P 1MB (not sure about vendor)
3DBench2: 95,8
PCPBench: 28 FPS
Doom: 1327 realticks
Quake: 23,6 FPS

For comparison sake I reposts Artex's results:

Hardware:
AMD Am486 DX4-100 (A80486DX4-100SV88)
ASUS VL/I-486SV2GX4 Rev. 2.0
16MB RAM (FPM)
256KB L2 Cache
CF Card on VLB Promise IDE Controller

SuperScape
S3 Vision 864 - #9 (9GXE64) Revolution Ticket to Ride (1MB): 635
Tseng ET4000 W32P - Diamond Stealth 32 (2MB): 635
Tseng ET4000 W32P - STB Lightspeed 50/50 (2MB): 635

PC Player Bench
S3 Vision 864 - #9 (9GXE64) Revolution Ticket to Ride (1MB): 13.4
Tseng ET4000 W32P - Diamond Stealth 32 (2MB): 13.4
Tseng ET4000 W32P - STB Lightspeed 50/50 (2MB): 13.4

DOOM
S3 Vision 864 - #9 (9GXE64) Revolution Ticket to Ride (1MB): 2108 realtics
Tseng ET4000 W32P - Diamond Stealth 32 (2MB): 2091 realtics
Tseng ET4000 W32P - STB Lightspeed 50/50 (2MB): 2090 realtics

Quake
S3 Vision 864 - #9 (9GXE64) Revolution Ticket to Ride (1MB): 8.6 fps
Tseng ET4000 W32P - Diamond Stealth 32 (2MB): 8.7 fps
Tseng ET4000 W32P - STB Lightspeed 50/50 (2MB): 8.6 fps

The Vision 864 is in the same series as the Vision 868, 964 and 968. So they should perform the same in DOS. All those were available as VLB cards.

Too long didn't read: There are alternatives to the ET4000/W32p.

vetz wrote:

Since there is so much hype for the ET4000/W32P I want to present my benchmark results. I finally got my hands on a Et4000/W32p card without giving a kidney. They back up the ones done by Artex.

Too long didn't read: There are alternatives to the ET4000/W32p.

I never understood why people like the ET4000 at all. Now granted the ISA version was quite fast,
but if you have a VLB of PCI system, there are much better cards. Ones that are equally fast or faster that have decent acceleration

And 1MB video RAM? eww! It will limit your resolution or color depth to sad levels

smeezekitty wrote:

I never understood why people like the ET4000 at all. Now granted the ISA version was quite fast,
but if you have a VLB of PCI system, there are much better cards. Ones that are equally fast or faster that have decent acceleration

And 1MB video RAM? eww! It will limit your resolution or color depth to sad levels

It's upgradeable to 2MB, and I hope to do that soon. It only matters for Windows usage though, which isnt that important on a DOS machine like a 486 will be used for nowadays.

If you have PCI, yes there are better cards, but when it comes to speed and compatibility there aren't any known better VLB cards than the S3 Visions and ET4000/W32p.

^ ah good to see these ET4000 cards getting de-hyped 🤣 , thanks for the numbers
tbh I don't even like the ISA Tseng cards 🤣 🤣 🤣 S3/CL ftw 😊 😊 😊

vetz wrote:

It's upgradeable to 2MB, and I hope to do that soon. It only matters for Windows usage though, which isnt that important on a DOS machine like a 486 will be used for nowadays.

hey hey I am posting from a 486 right now. Running Windows 2000 and Opera 9.6
Works fine with windows. With CF cards swapping OSes is easy. I have one of each for DOS,95,98 and 2000

If you have PCI, yes there are better cards, but when it comes to speed and compatibility there aren't any known better VLB cards than the S3 Visions and ET4000/W32p.

In that case I would pick the S3 still.

AFAIR the Tseng cards were legendary because they could hold a candle to the pro cards, but were much cheaper. And for DOS gaming, you didn't need high resolution or colour depth, just pure framebufer speed. A S3 Vision was kinda wasted on gaming.

Artex wrote:

vetz wrote:

Will you test the Asus VL-i486SV2GX4 @ 160mhz?

Could you post the gamebenches for the Biostar board @ 160mhz?

Certainly can do both!

No luck yet getting this processor to run at 160Mhz stable on this board, but here are the gamebenches:

AMD X5-133ADZ @ 133Mhz
SuperScape: 78.1
PC Player Bench: 20.4
Doom: 2134 gametics in 1659 realtics
Quake: 969 frames in 76.8 seconds, 12.6 fps

AMD X5-133ADZ @ 160Mhz
SuperScape: 94.1
PC Player Bench: 24.5
Doom: 2134 gametics in 1378 realtics
Quake: 969 frames in 64.0 seconds, 15.1 fps

5u3 wrote:

AFAIR the Tseng cards were legendary because they could hold a candle to the pro cards, but were much cheaper. And for DOS gaming, you didn't need high resolution or colour depth, just pure framebufer speed. A S3 Vision was kinda wasted on gaming.

Funny how the times have changed. Now the Tseng VLB cards have a much higher premium than the S3 Vision cards.

No luck yet getting this processor to run at 160Mhz stable on this board, but here are the gamebenches:

Thanks for the benches 😀 Did you use the same CPU you clocked to 160mhz in your Biostar board? I have no problems running my X5-133 at 160mhz in my Asus VL-i486SV2GX4 board. I even ran some benches at 180mhz.

vetz wrote:

Thanks for the benches 😀 Did you use the same CPU you clocked to 160mhz in your Biostar board? I have no problems running my X5-133 at 160mhz in my Asus VL-i486SV2GX4 board. I even ran some benches at 180mhz.

No problem! Can you shoot me a list of all your jumper/BIOS settings? I'm using the same processor as I used in the Biostar board. If I can get X5-133 to run at 160Mhz stable in this board, then I'm going to drop the Cyrix processor into the Biostar board.

I even ran some benches at 180mhz.

That's a funny frequency. Its not a multiple of any common FSB
How did you achieve that even?

smeezekitty wrote:

I even ran some benches at 180mhz.

That's a funny frequency. Its not a multiple of any common FSB
How did you achieve that even?

My memory was a bit wrong here. I never ran it at 180mhz in my ASUS board if anyone got the impression of that, but in my UMC based PCI board using 3x multiplier and 60mhz FSB.

I still would like to know your jumper and BIOS settings for running it in your SV2GX4 @ 160Mhz. 😀

Artex wrote:

I still would like to know your jumper and BIOS settings for running it in your SV2GX4 @ 160Mhz. 😀

That was over a year ago. I don't remember all the details, but AFAIK I used the jumper settings given on the internet (see Asus link below). BIOS settings about the same as on Pentium Overdrive, just L1 set to WB.

https://web.archive.org/web/19970616064538/ht … g/sv2g-cpu.html

Edit: Btw, you should use your Cyrix CPU in the rev 2.1 board:
"Rev 2.0 and 2.1 of the VL/I-486SV2GX4 only differ in one detail. 2.1 provides voltages between 3.3V and 5V (3.45V, 3.6V, 4.0V) intended for Cyrix processors."

Btw, maybe stupid question, you are aware the turbo setting is reversed on the SV2GX4? Meaning you need to have a jumper on the pins to activate it.

Edit2: Try these jumper settings from Bastlermike:
VLB 486 recommendation required please

Per the following archived Asus page, the recommendation for 5x86 is to set the hardware trap to 2-3. I think perhaps I was thinking this was for Am5x86, but I think it's actually for Cyrix 5x86 upon closer inspection.

Hardware trap setting:
VLB 486 recommendation required please

The jumper settings for running the AMD X5-133 are quite a bit different on the above Asus page compared to what BastlerMike is using...the differences are bolded below.

From Asus Page
JP16 1-2, 5-6 short for WT
JP17 1-2, 5-6 short
JP18 1-2 short
JP19 1-2, 3-4, 5-6 short
JP20 All Open for 3x clk, (1-2 short for 4x clk)
JP21 2-3 short for WT
JP22 1-2, 4-5 short

According to BastlerMike, it looks like the trap needs to be jumpered for Pin 1 & 2. I'll give this a shot when I get home:

From BastlerMike
JP16 1-2, 4-5
JP17 1-2, 5-6
JP18 1-2, 5-6
JP19 1-2, 3-4, 5-6
JP20 1-2
JP21 3-4
JP22 2-3, 4-5

&

JP5 1-2 (hardware trap)

For today's installment...

I was able to get the AMD X5-133 running at 160Mhz stable on the Asus VL/I-486SV2GX4 Rev. 2.1 board using the above jumper settings.

Build 2:
Asus VL/I-486SV2GX4 Rev. 2.1 running latest available BIOS (0402.001 BETA BIOS from 1999)
AMD-X5-133ADZ (Am5x86-P75) - 133Mhz running at 160Mhz (40mhz fsb x 4)
32MB RAM (2 x 16MB FPM 60ns SIMMS)
16KB L1 Cache (In Write-back mode)
256KB L2 Cache
TSENG ET4000/W32P - Diamond Stealth 32 2MB VLB (1993)
4GB CF Card (CF->IDE Adapter) on VLB Promise Controller
GOTEK USB Floppy Emulator

Build 2 Benchmark Results:
SpeedSys Overall Score: 60.00
DOOM: 2134 gametics in 1262 realtics
Superscape: 92.1 fps
PC Player Bench: 24.2
Quake: 969 Frames 63.9 seconds 15.2 FPS

Cache Performance:

CacheChk -d -t6
L1 (16KB) - 164.9 MB/s
L2 (256KB) - 74.5 MB/s
Main Memory Speed - 51.0 MB/s
Effective RAM Access Time (read) - 82ns
Effective RAM Access Time (write) - 50ns

CacheChk -d -w -t6
Main Memory Speed - 82.8 MB/s (12.7ns)
Effective RAM Access Time (write) - 50ns

For today's installment... (Part Deux) 😀

I fired up my Biostar MB-8433UUD Rev. 3.1 board this afternoon with my Cyrix 5x86-120GP (Code: G5K8548B -> 48th week of 1995) "M1sc"

Build 1:
BIOSTAR MB-8433UUD-A Rev. 3.1 with "2012" BIOS
32MB EDO RAM (1 x 32MB 60ns SIMM)
16KB L1 Cache (In Write-back mode)
256KB L2 Cache
STB S3 Trio64V+ (2MB)
4GB CF Card (CF->IDE Adapter)
GOTEK USB Floppy Emulator

Build 1 Benchmark Results:
SpeedSys Overall Score: 54.94
DOOM: 2134 gametics in 1639 realtics
Superscape: 87.8 fps
PC Player Bench: 21.4
Quake: 969 Frames 77 seconds 12.6 FPS

Cache Performance:

CacheChk -d -t6
L1 (16KB) - 246.0 MB/s
L2 (256KB) - 95.8 MB/s
Main Memory Speed - 55.7 MB/s
Effective RAM Access Time (read) - 75ns
Effective RAM Access Time (write) - 50ns

CacheChk -d -w -t6
Main Memory Speed - 83.8 MB/s (12.6ns)
Effective RAM Access Time (write) - 50ns

Artex wrote:

Build 1 Benchmark Results:
SpeedSys Overall Score: 54.94

This seems low to me for the Cyrix 5x86-120... thoughts? Someone else with the same CPU got a higher score.

EDIT: I missed the part where he set some Cyrix CPU registers for DOS:

LH 5x86 /loop_en=off /rstk_en=on /lsser=off /fp_fast=on /btb_en=on

Build 1 Benchmark Results (UPDATED):
SpeedSys Overall Score: 68.01
DOOM: 2134 gametics in 1525 realtics
Superscape: 94.8 fps
PC Player Bench: 22.5
Quake: 969 Frames 65.4 seconds 14.8 FPS

That X5-160 in your VLB board has really nice memory throughput results - slightly better than in PCI-based 486 boards.

Your Cyrix 5x86-120 seems to be configured nicely now, however I do not understand your slow L1 cache speed before you set the Cyrix register bits. It is as if you had L1 set to write-through mode. What all did you alter?

feipoa wrote:

That X5-160 in your VLB board has really nice memory throughput results - slightly better than in PCI-based 486 boards.

Your Cyrix 5x86-120 seems to be configured nicely now, however I do not understand your slow L1 cache speed before you set the Cyrix register bits. It is as if you had L1 set to write-through mode. What all did you alter?

Not sure - I'd have to go back and check.