feipoa wrote:

The AMD X5 at 180 MHz makes for a pretty nice system, even without cache. Of the 7 or so X5 chips I own, I cannot seem to get one operating stably at 180 MHz. What voltage and cooling solution are you using?

I just did the tests and then turned it off and while it ran all the tests without problems, I do not know how long term stable it is. The board is placed on my "testing rig station" and I just use a small 486 heatsink that I place on top the CPU and then place a 80mm fan which blows on it. I could go much more heavier on the cooling if it were required (which I would do for more long term solutions). I couldn't figure out how to get 3.6V on the motherboard (voltage jumpers are different from the SYL board) so I had to run the CPU at 4V. Even if the CPU is stable at this speed I do not want to build a system around it. The reason is because I just have one X5 CPU and I don't want to risk it dying on me. The speed area given from the X5 running at 180mhz I can cover on my socket 7 Compaq machine. I use my 486 VLB system for "486 DOS gaming" and it is more than quick enough for that usage 😀

feipoa wrote:

For the U6BC, IBM 5x86, AMD X5, and Intel DX4 chip results had sound enabled at 640x480. fps went up only 0.2 or 0.3 fps. Likewise, for the U4BC, only the IBM 5x86C-133 previously had sound enabled. At 320x200, I now get 19.3 fps. I'll update these docs at some ponit. For comparison, a non-overclocked Cyrix 5x86-133/4x gets 18.2 fps at 320x200.

Nice that you found some corrections to be made. Always nice to be as accurate as possible 😀

Does the Matrox accelerated mode code require Pentium instructions to run properly?

I'm starting to think so as well. It does make sense though as it was marketed towards the Pentium. When the Millennium came out in Summer of 1995 were the PCI 486 boards available at all? I always thought those boards first came out later in 95?

Anyway, the video won't be out until start of June. I want to be as time accurate as possible for these tests + video (November 1995) and for that I can't use a Socket 7 430VX board. So I have a Aladdin Socket 5 board on the way 😀

Do you have a photo of your ADW chip which seems to run well at only 4 V and 180 MHz? Most examples I've seen of this were at 5 V.

Does it run Win9x well?

I'm not exactly sure when the first PCI 486 board was made, but I have seen some boards with 1994 chipset datecodes. Of course, the maker could have used older chipsets in a 1995 board as well.

People run this CPU at 5V? Isn't that a bit too much and could fry it? I thought 4V to be pushing it. If you say it is safe to test 5V I could do that and test 200mhz again. I've also attached a picture of it.

Didn't do much testing in Win98SE. Is there anything specifically you want me to test?

Feb. of 1996 - that is a pretty old chip. I'm surprised it can do 180 MHz so easily.

If it were me, I wouldn't run it at 5 V for more than 5 minutes. It would be interesting to see if your chip could push 200 MHz though.
Since your chip seems to run well at 4 V and 180 MHz, perhaps somewhere in the 3.7 V - 3.9 V range would allow the chip to run cool enough for long term stability. With adaquate cooling, even 4 V and 180 MHz might be good enough for a fully setup system. A large factor in cooling when running near the operating temperature/frequency limits seems to come from putting the motherboard and CPU into a case. The rise in ambient case temperature end up being too much. Ideally, a large case with plenty of airflow is the way to go for such conditions.

For Windows 98SE, I'd be interested if your ADW chip could complete Ziff-Davis 3D Winbench97, Final Reality, and Quake II (software mode) at 640x480. These particular benchmarks seem to really tax a system and are pretty good for an initial assessment of stability.

vetz wrote:

I also have a Socket 4 system with a Pentium 66. […]
Show full quote

I also have a Socket 4 system with a Pentium 66.

Hardware:

• Pentium 66mhz (1x66)
  4x8MB FPM RAM
  Matrox G200 PCI
  Soundblaster 16 CT2230
  Intel OEM motherboard /w 256kb L2 cache

Here are the benchmark results of it:
Quake 320x200 no sound: 19.2 FPS
Quake 640x480 no sound: 7.3 FPS
Doom no sound: 42.4 FPS
3DBench 1.0: 71.4 1.0c: 70.6
PCPBench: 10.5
Speedsys:
CPU: 49.49
RAM Bandwidth: 100.7 MB/sec
VID Bandwidth: 46.6 MB/sec
L1: 124.8 MB/sec
L2: 60.46 MB/sec
Mem throughput: 39.1 MB/sec

Upon further examination, the extrapolated "socket 7 P66" scores are 21-46% higher than the scores you noted above. The socket 4 platform was more of a dog than I had originally anticipated. The only value which extrapolated correctly was the Speedsys CPU score.

I believe that to testing NASCAR Racing on Matrox Millennium and nVidia NV1 on a "period correct" fast system, we need to seek CPUs and chipsets that available in time when NASCAR Racing Matrox and NV1 versions was come out (and both are available).
This date is November 1995.
CPUs: Intel Pentium CPU-s up to 133MHz (for the big NASCAR Racing comparison Pentium 100 would be the most fear). AMD K5 and Cyrix 6x86 come out in 1996. Although Pentium Pro has introduced in 1995-Nov-01 but Pentium Pro CPSs have weak DOS gaming performance to my knowledge.
Chipsets: Socket 5 and first Socket 7 chipsets like Intel 430FX, SiS 551x.
I think Intel 430FX (the original "Triton") is the best but all Socket 5 and early Socket 7 boards are welcome here.
Intel 430VX and 430HX are later chipsets from 1996 so these are not "period correct". Some users say that there is a small speed difference between 430FX and 430VX. I have not tested. I have found my period correct FX board and I also have the later Intel VX, HX and TX boards too but I have not compared them yet so let we see their performance with Pentium 100 under DOS games:

Test setup:
Intel Pentium 100 (or a Intel P54* CPU on 1.5 x 66MHz, but P55C/MMX is not fear because that is faster)
PCI bus on 33MHz (this is the standard PCI speed)
Matrox Millennium II (MGA-2164WP-C) PCI (sorry I have neither Millennium I nor G200 PCI)
64MB EDO ram (Hynix chips) 60ns for FX, VX, HX;
64MB SDRAM (Hynix chips) 8ns cycle time (CK3) for TX.
Very small different between EDO and SD in TX: Quake 1.08 320x200:
25.5 fps with EDO vs. 25.6 fps with SDRAM.

OS:
MS-DOS 6.22 with "clean boot" (for example: "F5" at boot)
Of course the games has started without file manager/shell from command promt (by batch file).

Games:
Quake v1.06
Quake v1.08
Duke Nukem 3D v1.3D

Parameters for Quake 1.06 (my Q6.BAT batch file):

cd \quake106
quake -nomouse -nojoy -nocdaudio -nosound -nolan -nonet
cd \

Parameters for Quake 1.08 (my Q8.BAT batch file):

cd \quake108
quake -nomouse -nojoy -nocdaudio -nosound -nolan -nonet
cd \

Parameters for Duke Nukem 3D (my D.BAT batch file):

cd \duke3d duke3d /l6 /v1 /m /ns /nm duke3d /l8 /v1 /m /ns /nm duke3d /l3 /v2 /m /ns /nm duke3d /l7 /v2 /m /ns /nm duke3d /l11 / […]
Show full quote

cd \duke3d
duke3d /l6 /v1 /m /ns /nm
duke3d /l8 /v1 /m /ns /nm
duke3d /l3 /v2 /m /ns /nm
duke3d /l7 /v2 /m /ns /nm
duke3d /l11 /v2 /m /ns /nm
duke3d /l6 /v3 /m /ns /nm
duke3d /l8 /v3 /m /ns /nm
duke3d /l10 /v3 /m /ns /nm
cd \

means: Level 10 ; Volume 3 ; without Monsters ; No Sound ; No Music

I have made my tests in "fullscreen" mode so without statusbar (where you can see the ammo, health etc.).
In Quake, I have selected DEMO3 for the test because this is the longest.
Quake testing step-by-step:
press "Esc" for main menu
select "SINGLE PLAYER"
than "NEW GAME"
to open the (half size) console, press the grave key (the key left of the 1 key)
write to the console "timedemo demo3"
press "enter" key
and immediately press the grave key again to close the half size console (needs only for first command)
(on faster systems I don't use the result of the first test because closing console, but but with P100 you can fast enough)
after the test you got a full screen console, for a second test you can recall the previous command by pressing up arrow key than press "enter" key (grave key not needed in fullscreen console).

In Duke Nukem 3D the game will start automatically by the batch file above,
you only need to type in the code "dnrate"
It will shows your frame rate onscreen (on the top left corner),
than you need to wait some seconds to stabilize the number.

My results (in table on picture):

(I have attached this ODS table in ZIP)

My results (in text, some more details):

430FX: DTK PAM-0054I motherboard 256 Kb Asynchronous cache (bios date: 1995-Nov-24) […]
Show full quote

430FX:
DTK PAM-0054I motherboard
256 Kb Asynchronous cache
(bios date: 1995-Nov-24)

Quake 1.06
"fullscreen"
>timedemo demo3
320x200 = 20.8 fps
640x400 = 10.0 fps
640x480 = 8.8 fps

Quake 1.08
"fullscreen"
>timedemo demo3
320x200 = 21.9 fps
640x400 = 10.8 fps
640x480 = 9.5 fps

Duke Nukem 3D 1.3D
"fullscreen"
640x480
dnrate
Level 6, volume 1 = 27 fps
Level 8, volume 1 = 30 fps
Level 3, volume 2 = 43 fps
Level 7, volume 2 = 33 fps
Level 11, volume 2 = 45 fps
Level 6, volume 3 = 24 fps
Level 8, volume 3 = 40 fps
Level 10, volume 3 = 52 fps

430VX:
Gigabyte GA-586ATV motherboard

Quake 1.06
"fullscreen"
>timedemo demo3
320x200 = 23.9 fps
640x400 = 11.3 fps
640x480 = 9.9 fps

Quake 1.08
"fullscreen"
>timedemo demo3
320x200 = 25.0 fps
640x400 = 12.1 fps
640x480 = 10.6 fps

Duke Nukem 3D 1.3D
"fullscreen"
640x480
dnrate
Level 6, volume 1 = 30 fps
Level 8, volume 1 = 33 fps
Level 3, volume 2 = 46 fps
Level 7, volume 2 = 36 fps
Level 11, volume 2 = 48 fps
Level 6, volume 3 = 26 fps
Level 8, volume 3 = 42 fps
Level 10, volume 3 = 55 fps

430HX:
Intel TC430HX motherboard

Quake 1.06
"fullscreen"
>timedemo demo3
320x200 = 24.4 fps
640x400 = 11.4 fps
640x480 = 10.0 fps

Quake 1.08
"fullscreen"
>timedemo demo3
320x200 = 25.3 fps
640x400 = 12.2 fps
640x480 = 10.7 fps

Duke Nukem 3D 1.3D
"fullscreen"
640x480
dnrate
Level 6, volume 1 = 30 fps
Level 8, volume 1 = 33 fps
Level 3, volume 2 = 47 fps
Level 7, volume 2 = 36 fps
Level 11, volume 2 = 48 fps
Level 6, volume 3 = 26 fps
Level 8, volume 3 = 42 fps
Level 10, volume 3 = 55 fps

430TX:
ASUS TX97XE motherboard

Quake 1.06
"fullscreen"
>timedemo demo3
320x200 = 24.5 fps
640x400 = 11.4 fps
640x480 = 10.0 fps

Quake 1.08
"fullscreen"
>timedemo demo3
320x200 = 25.6 fps (25.5)
640x400 = 12.3 fps
640x480 = 10.8 fps

Duke Nukem 3D 1.3D
"fullscreen"
640x480
dnrate
Level 6, volume 1 = 30 fps
Level 8, volume 1 = 33 fps
Level 3, volume 2 = 47 fps
Level 7, volume 2 = 36 fps
Level 11, volume 2 = 48 fps
Level 6, volume 3 = 26 fps
Level 8, volume 3 = 42 fps
Level 10, volume 3 = 55-56 fps

So between FX and VX there is significant speed differenve so I say for the NASCAR Racing test the Intel VX/HX and later chipsets are not fear if we want a period correct comparison.

If you have Socket 5 board or early non Intel Socket 7 board your results are welcome here.
I have attached my table (.ods in zip).

Awesome testing! Who would have thought there would be such a performance difference on the Socket 7 Intel chipset boards? The FX chipset just bombs! I don't know how much it affects performance, but I tested with the GUI set to on in Quake. I also tested with demo1. The demo3 results can be compared against feipoa's 686 benchmark comparison.

I have my own 430FX and a Socket 5 board on the way, and once I receive them I'll test them and then post the results in this thread. I will also be using one or the other for the video. It will all be period correct hardware from Q4 1995.

vetz wrote:

The demo3 results can be compared against feipoa's 686 benchmark comparison.

I think you mean't to say can't. I also used all default settings.

What criteria are you guys using to establish "period correct" hardware? How many months apart can the products be released, e.g. the video card and motherboard, to still be considered "period correct"?

feipoa wrote:

vetz wrote:

The demo3 results can be compared against feipoa's 686 benchmark comparison.

I think you mean't to say can't. I also used all default settings.

What criteria are you guys using to establish "period correct" hardware? How many months apart can the products be released, e.g. the video card and motherboard, to still be considered "period correct"?

That's a very good question. Period correct could range anywhere from a span of weeks to a span of years depending on who you ask.

Period correct is a lie.
The day I installed a 9800pro in my P3-S 1266 because I was to poor to get a full Athlon 64, was I still period correct ? 😁

I don't understand how that would not be obvious? When the cutoff date is November 1995 we can just use hardware that were available up to that point.

May the criteria is not a clear line/limit for first view.
That time period we try to define is when 3D Blaster (VLB), NV1 and Millennium were released. That time these cards were true high end and expensive toys and the targeted systems are also the highend systems in that times. It is very likely that NV1 and Millennium (and NASCAR Racing 3D versions for them too) were optimalised, developed on Pentium processors and Intel Triton chipset (or at least that was the most tested platform).
So Intel 430FX chipset is period correct because there was available in Q4 1995 (at that time this was the "high end" Intel chipset) but Intel 430VX and 430HX are have introduced on 1996-Feb-12, later 430VX scheduled for the end of March. To compare the 486 VLB based 3D Blaster to a Pentium "top" machine in Q4 1995 is much more fear as to compare to a Q1 1996 Pentium "top" machine.
1996-Jan-04 "Intel release Pentium 150 and 166 processors." for example.
So in these pretests for NASCAR Racing 3D versions the "period correct" criteria is about the availablity in Q4 1995.
Later hardware is enabled if it does not affect the result. For example my 2x32 MB EDO ram modules are much later than Q4 1995 but perfurm full the same result in my the 430FX board as my good old FPM modules. Although it was sayed that times:
EDO is 5% faster than FPM DRAM, which it began to replace in 1995, when Intel introduced the 430FX chipset that supported EDO DRAM.
I can't find this 5% in my tests. Later Intel chipsets are faster with EDO (with VX: Quake 1.08 320x200: 28.6 fps with FPM but 29.1 fps with EDO) so to overstate the speed difference between these chipsets I have used EDO (and SD for TX) modules.

I think a Pentium 133 Mhz would be much better. I played SO many good games on my old Pentium 133 Mhz laptop and smoothly too. It still works actually, it is just unplayable now as the system was not reinstalled since god know when, and I broke the external "caddy" CD ROM drive the first day my mom bought it (well, I was a clumsy 9 year old...) which was not a problem as all the games were already there, but is a problem now because Windows 98SE does not run pretty when it was not reinstalled for more than 10 years.

But if you insist on strict period accuracy to 1995 as opposed to building a general mid-90s style rig, I'd go with Pentium 90. Real Pentium over those 486-style franken-pretend-Pentiums any day. You don't want a bottlenecked system. I don't know much about overclocking, but I think you could overclock it to at least 100 Mhz, which would give you Pentium 100 performance without violating strict period accuracy.

Anyways, if you want really high performance while being period accurate, try a Pentium Pro. Even the slowest one is 150 Mhz and it wa around since November 1995. Obviously, it depends on how you define period accuracy. Just because the average user in 1995 had a 486 does not mean high performance CPUs were not available, but only some people could affrord them. The ultimate 1995 machine would probably be a dual Pentium Pro machine, and that one could run Quake 3 (yes, Quake 3) quite well http://www.youtube.com/watch?v=bK5QFySGpy4 . It is a bit cheating tho because he used a modern graphics card, but still awesome. NFS Shift runs much worse on my friends laptop than Quake 3 runs on that old Pentium Pro.

m1so wrote:

I think a Pentium 133 Mhz would be much better.

That was the top of the line CPU in Q4 1995 and was very expensive. You have to remember that the three cards have to be compared against each other on a similar system. The problem is that the 3D Blaster runs on the VLB bus, while the NV1 and Matrox are PCI. I can test the NV1 and Matrox on the P133, which I also plan to do. There are ofc methods of testing the 3D Blaster on a Socket 5 system, but those boards (with VLB and PCI) are rare and expensive.

But if you insist on strict period accuracy to 1995 as opposed to building a general mid-90s style rig, I'd go with Pentium 90. Real Pentium over those 486-style franken-pretend-Pentiums any day. You don't want a bottlenecked system. I don't know much about overclocking, but I think you could overclock it to at least 100 Mhz, which would give you Pentium 100 performance without violating strict period accuracy.

Yes, this is what the results in this thread has shown. The 486 system running the POD at 100mhz with the 3D Blaster and the NV1 and Matrox running on a real P90. Performance is very similar and you also get each card tested on the system they were designed to run on.

Anyways, if you want really high performance while being period accurate, try a Pentium Pro. Even the slowest one is 150 Mhz and it wa around since November 1995.

The Pentium Pro were crazy expensive when it came out. Somewhere we have to draw the line. Another problem is that I don't own a Socket 8 system for testing. It would also not allow testing of the VLB card.

vetz wrote:

The Pentium Pro were crazy expensive when it came out. Somewhere we have to draw the line. Another problem is that I don't own a Socket 8 system for testing. It would also not allow testing of the VLB card.

They may have been crazy expensive in 1995 but now with so many people selling them for their scrap gold value, prices are climbing and they getting rare. Better grab one now while you still can.

Just tested my Gigabyte GA-586AP Socket 5 motherboard based on the ALI M1451B chipset. The results are shocking! It is even worse than the 430FX. The Pentium motherboards and chipset really was crap until the release of the VX/HX series in the beginning of 1996. Even late 486 chipsets beat the early Pentium ones even though the Pentiums have double bus width. I wish I could get hold of a 430NX Socket 5 board and benchmark that as well. Socket 5 UMC boards are also interesting.

Yes, is seems ALI M1451B a bit slow. I agree the Intel 430NX would be the most interesting to compare to the 430FX or the ALI M1451B.
I say the 430FX and Pentium 100 would be the most fear and period correct for Matrox Millennium and NV1. This is on the back side of the box of the Matrox Millennium:

Windows and DOS performance tests conducted by Matrox Graphics Inc. using Ziff-Davis Winbench 4.0 and PC Bench 9.0 on an Intel Triton P100 system (256k cache, 8 MB EDO RAM, Windows for Workgroups 3.1) at 1024 x 768 in 256 colors, 72 Hz, large fonts. MGA Millennium was configured with 2 MB of memory for the PCI bus using driver releases 1.0.

Here is the pictures about the box of the Millennium http://www.vgamuseum.info/index.php/pirxs-col … trox-millennium

Vetz,

I think you noted some of my benchmark scores incorrectly. For the POD100, it should be:

L1 / L2 / RAM
195 / 56 / 44

L2 is at 56 MB/s.

What L2 and DRAM timings are you using with your 1024K VLB board? I've noticed recently on a SiS 1024K board that the timings need to be reduced at 40 MHz which make the L2 slower than at 33 MHz. Do you have a snapshot of the chipset BIOS settings?

Gona wrote:

I say the 430FX and Pentium 100 would be the most fear and period correct for Matrox Millennium and NV1.

I agree. It is just too bad that the Iwill P54Ti 430FX board I received was DOA! I need to get another one now 🙁

I think you noted some of my benchmark scores incorrectly. For the POD100, it should be:

L1 / L2 / RAM
195 / 56 / 44

Sorry, fixed the L2 score!

What L2 and DRAM timings are you using with your 1024K VLB board? I've noticed recently on a SiS 1024K board that the timings need to be reduced at 40 MHz which make the L2 slower than at 33 MHz. Do you have a snapshot of the chipset BIOS settings?

See picture:

I haven't seen any need to reduce the timings when going to 40mhz FSB.

Maybe VLB/ISA-only boards were designed with 40 MHz operation as a primary objective? Which chipset is on that board? I wonder if POD-100/40 cache/dram timings are naturally less sensitive than Cyrix 586-120/40 timings...?

Your BIOS is missing one of the cache options in some SiS 496/497 chipsets I have. It is the L2 Cache/DRAM Cycle WS settting. Is your system only stable with Cache Write Cycle on 3T? What happens if you use 2T?

Could you run cachechk -w -d -t6 to check your memory write speed and cachechk -d -t6 to check your read speed?

Would you mind sending a binary image of your BIOS file? I want to explore which settings they have hidden.