VOGONS


Voodoo 1 vs. Voodoo 2 on a 486

Topic actions

Reply 100 of 124, by feipoa

User metadata
Rank l33t++
Rank
l33t++

The POD generally got a bad rap for being expensive, late, and with specs changing. It was also widely considered slow in everyday tasks at the time. It does alright in games, obviously. Anyone have 3D gaming experience with a POD83 from 1995-1998? If it was your main system at the time, it would be interesting to hear your take on the chip.

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

Reply 101 of 124, by Scali

User metadata
Rank l33t
Rank
l33t
feipoa wrote:

The POD generally got a bad rap for being expensive, late, and with specs changing. It was also widely considered slow in everyday tasks at the time. It does alright in games, obviously.

I think this was mainly the result of the 'underdog' that AMD was at the time.
For most people it was 'cool' to have a highly clocked AMD 486/5x86 chip, rather than an Intel.
I always loved the Pentium OverDrive because it gives you a 'real' Pentium. All the other CPUs claim 'P75 rating' or whatever, but they are tweaked 486 CPUs, no superscalar capabilities, not the massively improved FPU etc. Quake made it painfully obvious that the P75 rating clearly doesn't apply in all scenarios. The P83OD on the other hand does actually deliver P75 or better ALU and FPU performance, and is only limited by the 32-bit bus of the 486 infrastructure.

http://scalibq.wordpress.com/just-keeping-it- … ro-programming/

Reply 102 of 124, by Eleanor1967

User metadata
Rank Member
Rank
Member
feipoa wrote:

In the past, I had tested the Am5x86-150 with 1024K/128M, however with 256K/64M, I was able to get the system stable with a cache wait state of only 3-1-1-1. It would be fun to test the system with 10 ns cache (counterfeit) or 12 ns cache (real) to see if the wait states could be reduced to 2-1-1-1. Because of wait states, the L2 speed of the Am5x86-150 was the same as the Am5x86-160. The Am5x86-160 still has the benefit of the 40 MHz PCI bus, while the Am5x86-150 is running at 33 MHz. The resultant is that the Am5x86-160 is about 4% faster than the Am5x86-150.

I've got these settings stable on an Am5x86 @ 150 Mhz with real 12 ns cache. Only DRAM setting couldn't be maxed out, haven't tried with a 50ns stick though.
This is on an Acer AP43 with 256k dual banked. Stable means DOS, DOOM and Quake stable. No Windows testing.

Attachments

  • 20180913_232123.jpg
    Filename
    20180913_232123.jpg
    File size
    1.81 MiB
    Views
    1459 views
    File license
    Fair use/fair dealing exception
  • 20180913_232023.jpg
    Filename
    20180913_232023.jpg
    File size
    1.18 MiB
    Views
    1459 views
    File license
    Fair use/fair dealing exception

Reply 103 of 124, by feipoa

User metadata
Rank l33t++
Rank
l33t++

155/84/46 for 2-1-1-1, and I assume that "DRAM Faster" = 1ws/0ws?

I get 155/76/56 for 3-1-1-1 and DRAM 0ws/0ws. Data on page 4 of this thread.

Did your system pass HIMEM with these settings? Will you be able to install Windows and run a game? It is fairly easy to get faster settings going in DOS.

Are you also using 12 ns for the TAG?

I might have some real 12 ns cache coming. At which time, I intend to try 10 ns fake, 12 ns fake, and 12 ns real with IBM 5x86c, Am5x86-150.

Have you tried 512K or 1024K double-banked? How much RAM are you using? Might want to try a single stick of 32 MB. I have a single stick of 50 ns FPM, 32 MB, but it hasn't made any difference over my 60 ns sticks.

AP43 contains a SiS 496, which has a slight memory read edge on the UMC UM8881 for the lowest wait state.

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

Reply 104 of 124, by Eleanor1967

User metadata
Rank Member
Rank
Member

All test but quake I did with a single 8 MB FPM stick, I just tried a single 32 MB stick of EDO which allowed me to go with DRAM "Fastest" which I assume is 0WS/0WS, but I can't be sure. See picture below. All with a 12 ns TAG. The system also successfully booted Windows 95, but I haven't tried any games on that yet, it is quite late here so it will have to wait.

I am now curious of the wait states I can get at 60 FSB instead of 66 with the Cyrix. Will try that, too.

Attachments

  • 20180914_020148.jpg
    Filename
    20180914_020148.jpg
    File size
    1.16 MiB
    Views
    1445 views
    File license
    Fair use/fair dealing exception

Reply 105 of 124, by feipoa

User metadata
Rank l33t++
Rank
l33t++

The 56 vs. 59 MB/s for RAM read is the edge I was talking about in favour of the SiS 496. I want to say that the UMC had a slight edge in L2 cache read speed, but don't recall. Are you able to run your system with 3-1-1-1 and see what the L2 result is in cachechk?

At 60 MHz FSB, I had to use 3-2-2-2 for L2 cache. Perhaps some real 12 ns cache could bring that to 2-2-2-2 or 3-1-1-1, I don't know. My issue is that I previously only tested with 1024K, which doesn't work as well with tight timings on borderline stable systems.

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

Reply 106 of 124, by The Serpent Rider

User metadata
Rank l33t++
Rank
l33t++

At 60 MHz FSB, I had to use 3-2-2-2 for L2 cache. Perhaps some real 12 ns cache could bring that to 2-2-2-2 or 3-1-1-1, I don't know.

2-2-2-2 - might work. 3-1-1-1 - impossible. In the first place, such tight timings were only possible due to drastic speed difference between 486 and Pentium buses.

I must be some kind of standard: the anonymous gangbanger of the 21st century.

Reply 107 of 124, by Eleanor1967

User metadata
Rank Member
Rank
Member
feipoa wrote:

Are you able to run your system with 3-1-1-1 and see what the L2 result is in cachechk?

I played with all 3 cache settings a bit.
- Write Cycle on 3 doesn't affect L2 at all, main memory speed drops to 46.1 though.
- Burst Cycle on 2 drops L2 perfromance so hard that cachechk doesn't recognize it anymore, although speedsys still shows a bit of difference in perfromance (was it 65 MB/s I can't remember?)
- Single Cycle on 3 hangs te system on post. I really don't get why.

Reply 108 of 124, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Yeah, I found that intersting as well with cache write cycle and main memory speed. ----> Help with SiS 496/497 Tomato 4DPS 486 motherboard

The cache/memory and other settings on SiS496 boards are more difficult to figure out and optimise compared with Um8881 boards.

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

Reply 109 of 124, by chublord

User metadata
Rank Member
Rank
Member
derSammler wrote on 2018-08-08, 07:39:

Multitexturing happens on the Voodoo card, it doesn't matter for the CPU or the bus. The Voodoo 1 needs 2 passes for multitexturing, while the Voodoo 2 only needs one pass. While, in theory, you could get more frames in games that use multitexturing, in reality it won't happen since a slow CPU isn't able to provide enough additional geometry data for more fps. On a 486/5x86 the CPU is the limit, not the Voodoo card. Only something like hardware T&L would help.

I know this thread is a couple years old, but has anyone tested a 486 system + graphics card with hardware T&L? Something like a early Geforce PCI or the economy Geforce MX's from later eras?

IBM Valuepoint 486 DX4-100, Opti 802G, 50 MHz FSB, Voodoo1+S3 864, Quantum Fireball EX 4.0 GB, Seagate Medalist 1.6 GB, 128 MB FPM, 256k L2

Reply 110 of 124, by The Serpent Rider

User metadata
Rank l33t++
Rank
l33t++

Something like a early Geforce PCI or the economy Geforce MX's from later eras?

You need T&L support from games, but 486 simply can't render such new games fast. You probably could gain +1.0-1.5 fps to 5 fps score in, for example, MDK2. Which would be quite impressive in percentage gains, but won't change anything realistically.

I must be some kind of standard: the anonymous gangbanger of the 21st century.

Reply 111 of 124, by leileilol

User metadata
Rank l33t++
Rank
l33t++

There's also the fact most of the early HWT&L requiring games use DX8 for their API backend, which is a DirectX that cannot be installed on 486s whatsoever, and still eat at the CPU anyway (hungry for that GHz) to make 4th gen PCs a non-starter.

I believe feipoa already benchmarked a Geforce2MX in a Cx586 before...

apsosig.png
long live PCem

Reply 113 of 124, by chublord

User metadata
Rank Member
Rank
Member
leileilol wrote on 2020-07-05, 12:43:

There's also the fact most of the early HWT&L requiring games use DX8 for their API backend, which is a DirectX that cannot be installed on 486s whatsoever, and still eat at the CPU anyway (hungry for that GHz) to make 4th gen PCs a non-starter.

I believe feipoa already benchmarked a Geforce2MX in a Cx586 before...

I had my old 486 Valuepoint running Quake 3 (I know that's OpenGL) but I believe this game can use hardware T&L. It was maybe 3-4 fps, but my configuration was far from the best for a 486 machine. (DX2-80 with 4MB Voodoo1)

IBM Valuepoint 486 DX4-100, Opti 802G, 50 MHz FSB, Voodoo1+S3 864, Quantum Fireball EX 4.0 GB, Seagate Medalist 1.6 GB, 128 MB FPM, 256k L2

Reply 114 of 124, by darry

User metadata
Rank l33t++
Rank
l33t++
feipoa wrote on 2020-07-05, 13:10:

I have tried but couldn't get the GF2 drivers working with a Cyrix 5x86; worked with an Am5x86 though.

Did the GF2 work for 3D in OpenGL and/or DirectX on that AM5X86 ?

I would have expected the 3D acceleration handling part of the driver to have been designed to take advantage of (and likely require) Pentium instructions by this point in time .

Reply 115 of 124, by The Serpent Rider

User metadata
Rank l33t++
Rank
l33t++

I had my old 486 Valuepoint running Quake 3 (I know that's OpenGL) but I believe this game can use hardware T&L.

Only partially.

I must be some kind of standard: the anonymous gangbanger of the 21st century.

Reply 116 of 124, by leileilol

User metadata
Rank l33t++
Rank
l33t++
chublord wrote on 2020-07-05, 16:52:

but I believe this game can use hardware T&L.

it doesn't. Every lighting and transform in Q3 is strictly software calculated. Any T&L that happens are likely more simple things coming from the OpenGL driver's end, and Q3's renderer doesn't offload any of the transform / lighting work to it. A lot of the renderer speed comes from how everything's batched into primitives/strips (that lack data T&L would work with, like normals) which benefit compiled vertex arrays. also that fast inverse square root thing used on a couple things (envmaps and entity lightgrid calc IIRC)

apsosig.png
long live PCem

Reply 117 of 124, by Mvickers03

User metadata
Rank Newbie
Rank
Newbie

This is great.

78 FPS on Descent 2 on a 486 is nuts. What FPS did you get without any Voodoo?

I guess your motherboard is able to provide 66MHz FSB to the CPU and 33MHz FSB to PCI for this to work reliably, is that right?

I was staying away from PCI boards as they don’t like being overclocked. I may look to upgrade later down the line for if the difference is that great.

Reply 118 of 124, by The Serpent Rider

User metadata
Rank l33t++
Rank
l33t++

I was staying away from PCI boards as they don’t like being overclocked.

Are you joking? PCI motherboards are the only option for extreme overclocking. VLB boards can't achieve even 50 Mhz FSB reliably.

I must be some kind of standard: the anonymous gangbanger of the 21st century.

Reply 119 of 124, by Mvickers03

User metadata
Rank Newbie
Rank
Newbie

No I’m not joking mate. I’m actually looking for a board that can run the PCI bus at 33 and CPU at 66 so it’s stable. From what I’ve read even at 40 the PCI bus is unstable.

If you want to teach me something go ahead. Or even better sell me a board suitable for overclocking 😉