Voodoo 3 and Voodoo 2 SLI should be pretty equal on a 486 I think and since we've already set it up (Markk actually 😜) I see no point in swapping the Voodoo 2 SLI for a Voodoo 3.
Time is running short unfortunately as we only have a few days to commence the experiment (:P), so another CPU is totally out of the question. However, we might give it another try in the coming months! Where can I get a Cyrix 5x86 cheaply enough (ebay.co.uk doesn't have any of them and .de has 3 very expensive ones!) ?

By the way, anyone ever run Dark Forces 2 on a fast 486? How does it play? I'd guess it'd be half playable, Dark Forces 2 was pretty light on system requirements compared to Quake and Quake 2!

If you got sims lying around try Longbow 2, Total Air War, European Air War, and give Grand Prix Legends a shot.. These all appeared in the V1/2 transition but were very demanding even on pentium systems. Would be interesting to see how they fly on a 486..

My memories aren't that fresh but for me a basic 486DX struggled with 3D games. It was only with a DX2 that 3D was getting smooth enough. So ideally for 3D you really want a DX4 bur personally I would go with a Pentium 😀

F2bnp wrote:

Voodoo 3 and Voodoo 2 SLI should be pretty equal on a 486 I think and since we've already set it up (Markk actually 😜) I see no point in swapping the Voodoo 2 SLI for a Voodoo 3.
Time is running short unfortunately as we only have a few days to commence the experiment (:P), so another CPU is totally out of the question. However, we might give it another try in the coming months! Where can I get a Cyrix 5x86 cheaply enough (ebay.co.uk doesn't have any of them and .de has 3 very expensive ones!) ?

Voodoo2 SLI is comparable to Voodoo 3 2000 in performance and that is absolutely amazing, considering. However, your 486 CPU is a bottleneck for those GPUs, as Voodoo2 was basically meant for Pentium II's and higher. So basically any 3D game that utilized Glide/OpenGL or Direct3D is "very demanding 3D to run on a 486". 😜

The plus side though is that your Voodoo2 cards provide you with VESA 2.0 compatability. 😀

Guys, you don't get this. This is an experiment, to see how much a 486 cripples the Voodoo 2 SLI and other similar video cards.
The idea is to go as fast as possible with a 486, just for shits. This is never going to be a permanent system, it is just an experiment!
I would be mad to go play 3D games on a 486 on a daily basis. I play most of my games (new or old) on my main PC anyway 😜

feipoa wrote:

A great idea! […]
Show full quote

A great idea!

F2bnp wrote:

Well, the point is to see how a 486 performs, so no Pentium Overdrive. The Cyrix 5x86 also doesn't clock as high AFAIK so it should be pretty equal in performance to a faster clocked AMD 5x86.

You may want to prefer to the comparative charts in the Ultimate 486 Benchmark Comparison. Only the ALU of an X5-160 has an advantage over a Cyrix 5x86-120. If the games you are testing are heavily FPU-based, a Cyrix 5x86-120 is a better choice.

If you are trying to max out the possibility of a 486 with 3D games, I am in agreement with Sliderider, a Cyrix 5x86-120's FPU is about 10 pentium ratings better than an AMD X5-160. A POD83-WB is 3 pentium ratings better than the Cyrix 5x86-120. If you are using a Cyrix 5x86, ensure that you set LSSER to 0, FP_FAST to 1, and BTB to 1, otherwise, you're preformance drops. Refer to Cyrix 5x86 Register Enhancements Revealed for performance improvements and how to enable these bits. LSSER and FP_FAST are probably the most important for this test, but if you can enable BTB, that's another bonus.

Didn't the 200 MHz AMD x5 end up beating the 5x85 133 FPU in your tests? (which should mean an 180 MHz x5 should beat a 120 MHz 5x86, and a 150 or 160 MHz x5 should beat a 100 MHz 5x86)

Actually, I was a bit surprised the POD and 5x86 performance gap over the 486 wasn't bigger, since (on paper) the common FPU operations take around 1/3 to 1/4 the number of cycles on 5x86 as the 486, and considerably faster still on the Pentium. (for add/subtract/multiply, at least -divide is about the same on the Cyrix/Pentium, and both about 2x as fast as the 486)

I love this thread and the first thing I thought of when I rebuilt my (extremely similar) 486 a few years back was "run the newest games I can"

I did try Unreal on a 486 - both SoftDrv and GlideDrv ran at the same 4-7fps framerate. 😳

F2bnp wrote:

Yes, Unreal (not UT) and Half-Life are already on the list 😜.
Half-Life skeletal animation will probably absolutely KILL performance though!

Actually HL's big killer is the 16-bit lightmapping you can't flashblend, and the DSP (software driven reverbs)

F2bnp wrote:

By the way, anyone ever run Dark Forces 2 on a fast 486?

It's a tad choppy 😜

For Quake2 it may help if you force your sound to 8-bit mixing.

kool kitty89 wrote:

Didn't the 200 MHz AMD x5 end up beating the 5x85 133 FPU in your tests? (which should mean an 180 MHz x5 should beat a 120 MHz 5x86, and a 150 or 160 MHz x5 should beat a 100 MHz 5x86)

Actually, I was a bit surprised the POD and 5x86 performance gap over the 486 wasn't bigger...

If you look again at the FPU chart, there is an AMD X5-180 (60x3) on there, and even that was 2.5 PR less than a Cyrix 5x86-120 (40x3). The key to a fast Cyrix 5x86 is to actually turn on the advanced features, which most people didn't didn't do in the past. When I disabled advanced features on a Cyrix 5x86-133, the consequence was 16.6 pentium ratings.

Finding an AMD X5 which will run at 180 or 200 MHz is very rare indeed, probably as rare as finding a Cyrix 5x86 which will run well at 150 MHz. If you are going to include obscure CPUs, then a Cyrix 5x86-150's FPU cripples that of the X5-200 by 10 PR.

I don't think any game is going to be exclusively FPU reliant, so the extra gain offered by the X5-160's ALU may equate to the Cyrix 5x86-120 in certain games. A good example is running Quake1 on the same system as the X5-160. The Cyrix 5x86-120 (3x40) got 16.8 fps, whereas the AMD X5-160 scored slightly higher at 17.3 fps. Note that these scores are were acquired on a system which added a 2/3 PCI multiplier, so the PCI bus was at 27 MHz. If you run your PCI bus at 40 MHz, expect these results to increase slightly. All Cyrix advanced features, including branch prediction, were enabled.

feipoa wrote:

Since you are going to use a 40 MHz FSB, did you confirm that this SiS board is not adding an automatic 2/3 FSB-to-PCI bus multiplier? If so, that would drop the PCI bus down to 27 MHz.

Have you confirmed this yet? Without getting too fancy, an easy way to check this is to run 3dbench on your X5 at 133 MHz (4x33) and at 160 MHz (4x40). If your 133 MHz score is higher than your 160 MHz score, then your motherboard is automatically manipulating the PCI frequency.

feipoa wrote:

If you look again at the FPU chart, there is an AMD X5-180 (60x3) on there, and even that was 2.5 PR less than a Cyrix 5x86-120 […]
Show full quote

kool kitty89 wrote:

Didn't the 200 MHz AMD x5 end up beating the 5x85 133 FPU in your tests? (which should mean an 180 MHz x5 should beat a 120 MHz 5x86, and a 150 or 160 MHz x5 should beat a 100 MHz 5x86)

Actually, I was a bit surprised the POD and 5x86 performance gap over the 486 wasn't bigger...

If you look again at the FPU chart, there is an AMD X5-180 (60x3) on there, and even that was 2.5 PR less than a Cyrix 5x86-120 (40x3). The key to a fast Cyrix 5x86 is to actually turn on the advanced features, which most people didn't didn't do in the past. When I disabled advanced features on a Cyrix 5x86-133, the consequence was 16.6 pentium ratings.

Finding an AMD X5 which will run at 180 or 200 MHz is very rare indeed, probably as rare as finding a Cyrix 5x86 which will run well at 150 MHz. If you are going to include obscure CPUs, then a Cyrix 5x86-150's FPU cripples that of the X5-200 by 10 PR.

I don't think any game is going to be exclusively FPU reliant, so the extra gain offered by the X5-160's ALU may equate to the Cyrix 5x86-120 in certain games. A good example is running Quake1 on the same system as the X5-160. The Cyrix 5x86-120 (3x40) got 16.8 fps, whereas the AMD X5-160 scored slightly higher at 17.3 fps. Note that these scores are were acquired on a system which added a 2/3 PCI multiplier, so the PCI bus was at 27 MHz. If you run your PCI bus at 40 MHz, expect these results to increase slightly. All Cyrix advanced features, including branch prediction, were enabled.

feipoa wrote:

Since you are going to use a 40 MHz FSB, did you confirm that this SiS board is not adding an automatic 2/3 FSB-to-PCI bus multiplier? If so, that would drop the PCI bus down to 27 MHz.

Have you confirmed this yet? Without getting too fancy, an easy way to check this is to run 3dbench on your X5 at 133 MHz (4x33) and at 160 MHz (4x40). If your 133 MHz score is higher than your 160 MHz score, then your motherboard is automatically manipulating the PCI frequency.

It's not really fair comparing the AMD and Cyrix 5x86 designs, though. The AMD chip was still very much based in the 486 generation while the Cyrix part was a scaled back version of their Pentium compatible chip modded to work in socket 3 motherboards.If the POD can't be considered for purposes of this discussion because it's not a 486, then neither should the Cyrix 5x86. A Cyrix DX4 chip should be considered instead.

Thinking about it now, the Cyrix 5x86 shouldn't even be included in the fastest 486 benchmarks, either, since it has more in common with a Pentium than a 486.

sliderider wrote:

It's not really fair comparing the AMD and Cyrix 5x86 designs, though. The AMD chip was still very much based in the 486 generation while the Cyrix part was a scaled back version of their Pentium compatible chip modded to work in socket 3 motherboards.If the POD can't be considered for purposes of this discussion because it's not a 486, then neither should the Cyrix 5x86. A Cyrix DX4 chip should be considered instead.

Perhaps so, unless the goal is to see how much abuse a socket 3 platform system can take, whereby the Intel Pentiums are exluded. I often use "486" loosely to simplify titles. If I were to run this, I'd probably include the best socket 3 CPU from each vendor, an AMD X5-160, a POD-83/100, and an IBM 5x86c-133 (2x66). If Win95 offers even the slightest speed advantage over Win98 for the employed games, I'd probably use Win95. I'd also ensure the PCI frequency is not reduced when using FSB's other than 33 Mhz. You also want to be sure all RAM is cached and that all BIOS timings are optimised. Ultimately, it is up to the tester to decide what he/she wants to include or not.

sliderider wrote:

Thinking about it now, the Cyrix 5x86 shouldn't even be included in the fastest 486 benchmarks, either, since it has more in common with a Pentium than a 486.

If you are refering to the Ultimate 486 Benchmark Comparison, the goal there was for an Ultimate Socket 3 Benchmark Comparison, however that title isn't as catchy. For many people, the ears really perk up when '486' is uttered.

feipoa wrote:

Perhaps so, unless the goal is to see how much abuse a socket 3 platform system can take, whereby the Intel Pentiums are exluded. I often use "486" loosely to simplify titles. If I were to run this, I'd probably include the best socket 3 CPU from each vendor, an AMD X5-160, a POD-83/100, and an IBM 5x86c-133 (2x66). If Win95 offers even the slightest speed advantage over Win98 for the employed games, I'd probably use Win95. I'd also ensure the PCI frequency is not reduced when using FSB's other than 33 Mhz. You also want to be sure all RAM is cached and that all BIOS timings are optimised. Ultimately, it is up to the tester to decide what he/she wants to include or not.

"fair" has other areas of contention anyway . . . like AMD's access to more advanced manufacturing cprocesses for the 486 vs Cyrix's 5x86. (AMD also released it later, so the DX4 120 was the more contemporary chip of the Cyrix 5x86 -since Cyrix discontinued production in favor of the 6x86)

Also, the 5x86 may be a scaled-down 6x86, and it may have the same FPU (with very similar performance), but given Fiopa's actual tests, the faster clocked (overclocked) x5 486s pushed ahead in many tests/applications, including floating poiin5 in some cseveral cases (approaching the POD 83, when that technically should have been much faster).
The only area the 5x85 consistently obeat the x5 486 was I/O and cache performance. (which is also the biggest area the 6x86 had over the P5)

Some of those results may be due to compilers not using pentium optimizations or bottlenecked by areas other than floating point or I/O . . . or others that are pentium optimized, but don't end up benefiting the CX5x86 over the 486.
Quake, which should be very pentium-optimized, performed almost as well on a x5 200 as a POD, and significantly faster than a Cyrix 5x86 133.

The performance gap likely would have been wider had the 5x86 continued production longer (and transitioned to later processes and higher speeds), or more so had it actually been released as a socket 5/7 part, but as it is, the performance is much more in line with the 486. (especially for mainstream apps of their day ~95/96, hence the x5 133 and Cyrix 133 both being P-rated 75)

I know some of the guys arent interested in the POD83 but anyway.

I got the Quake 2 timedemo up to 15.2fps with sound, basically I turned the gun animation off and particle effects, this is blood, smoke, etc, the particle effects took too much away from the game but the gun I could live without which is about 14fps.

I played through the entire Dark Forces 2 demo and it was'nt too bad, definately better than Quake but still had big slowdowns although I ran in high detail, funny when we talk about whats playable Goldeneye on the N64 did'nt have high frame rates most of the time and had huge slow downs.

F2bnp wrote:

Yello! Me and Markk are going to try some games on a beefed up 486 machine, specifically an Am5x86 at 160MHz... Time is running short unfortunately as we only have a few days to commence the experiment...

How did it go?

Well, unfortunately we had to postpone the experiment due to health complications (nothing too serious though) ! We'll definitely report back with the final results once we get together again to do it 😜

feipoa wrote:

If you look again at the FPU chart, there is an AMD X5-180 (60x3) on there, and even that was 2.5 PR less than a Cyrix 5x86-120 (40x3). The key to a fast Cyrix 5x86 is to actually turn on the advanced features, which most people didn't didn't do in the past. When I disabled advanced features on a Cyrix 5x86-133, the consequence was 16.6 pentium ratings.

Finding an AMD X5 which will run at 180 or 200 MHz is very rare indeed, probably as rare as finding a Cyrix 5x86 which will run well at 150 MHz. If you are going to include obscure CPUs, then a Cyrix 5x86-150's FPU cripples that of the X5-200 by 10 PR.

I don't think any game is going to be exclusively FPU reliant, so the extra gain offered by the X5-160's ALU may equate to the Cyrix 5x86-120 in certain games. A good example is running Quake1 on the same system as the X5-160. The Cyrix 5x86-120 (3x40) got 16.8 fps, whereas the AMD X5-160 scored slightly higher at 17.3 fps. Note that these scores are were acquired on a system which added a 2/3 PCI multiplier, so the PCI bus was at 27 MHz. If you run your PCI bus at 40 MHz, expect these results to increase slightly. All Cyrix advanced features, including branch prediction, were enabled.

I'd missed this post before:

Yes, I was partially thinking of the Quake scores (where the AMx5 200 had a decided edge over the Cyrix 5x86 133). However, the 486 FPU scores (clock for clock) still seem to fare much closer to the Cyrix (and P5) than I'd have guessed by the execution cycle times and latencies I've seen listed for common FPU instructions on those 3 processors. (as seen on wikipedia's x87 page -which is backed by some pretty good documentation iirc)
Maybe it's an issue of optimized coding (or lack thereof) deflating the performance advantage of the Cyrix or P5, though I'd think not given the vintage of the benchmark. (and P6 optimization rules share a lot with P5, so there shouldn't be much of a conflict there either; k6 FPU optimization shouldn't be in conflict either since that doesn't depend on the scheduling the P5/P6 do to achieve high throughput since there's no added pipeline-related latency overhead on the K6 -though other rules for optimized FPU/ALU parallelism may apply, as they do for the 6x86 with its FPU FIFO)

For Quake, it could certainly be an ALU issue favoring the 486 (which is also probably why the K5 beats the K6 in the 133 MHz challenge -exceptional ALU performance, though the K5 FPU multiplier also likely plays a role). The 6x86/MII certainly runs quake faster at the same clock speed in spite of using the same (or virtually the same) FPU. Not just raw integer execution performance either, but also I/O and cache speed may come into play, though the 486 has a massive disadvantage there, so it's probably tied more to the ALU execution times.

kool kitty89 wrote:

I was partially thinking of the Quake scores (where the AMx5 200 had a decided edge over the Cyrix 5x86 133).

I don't quite follow. An IBM/Cyrix 5x86-133 (2x66) beat the AMD X5-200 (4x50) in Quake by 0.1 fps (18.4 fps vs. 18.3 fps). If you are refering to the Ultimate 486 Benchmark Comparison concerning the Cyrix 5x86 at 4x33, that score is with branch prediction disabled.

To get some idea for how much improvement branch prediction improves the Quake score, refer to column's D and E; the improvement is 3%. For the Cyrix 5x86-133 at 4x33 with branch prediction enabled, count on a Quake score of 17.8 fps.

Remember you are still comparing different FSB's, namely 40, 50, and 66 MHz. The 66 MHz FSB Cyrix had a 1/2 PCI divider, so the PCI clock was at 33 MHz. The X5-200 w/50 MHz did not have a PCI divider, so that PCI clock was run at 50 MHz. Had the 66 MHz Cyrix 5x86 been run with a 2/3 PCI divider, or a 44 MHz PCI clock, the 0.1 fps gain of the Cyrix would have been even greater.

Quake on an AM5x86 performs better with memory r/w waitstates set to 0 or 1. Goes from 8fps to 12fps at Start.bsp, catching up to Cyrix686MX performance @ 80MHz. 😁

PhaytalError wrote:

The plus side though is that your Voodoo2 cards provide you with VESA 2.0 compatability. 😀

I cant imagine that as v2 cards have no 2d capability

Davros wrote:

I cant imagine that as v2 cards have no 2d capability

Actually, Voodoo2 has features for 2D / GUI acceleration.

http://darwin-3dfx.sourceforge.net/voodoo2.pdf

2D Features · Direct memory-mapped access to frame buffer and texture memories via linear address mapping · 2D BitBLT engine s […]
Show full quote

2D Features
· Direct memory-mapped access to frame buffer and texture memories via linear address mapping
· 2D BitBLT engine supporting CPU-to-Screen and Screen-to-Screen transfers
· Separate programmable strides for Source and Destination areas during BitBLT transfers
· Solid Fills
· Monochrome text expansion with optional byte-packed glyph format
· Ultra-fast full-screen clears using SGRAM color-expansion capability*
· 16 Raster Operations (ROPs)
· Source and Destination Chroma-range functionality
· Scissor rectangle clipping
· 2D BitBLT registers and state independent of 3D rendering registers and state

You can in fact use one to run your Linux desktop accelerated.
http://www.x.org/archive/X11R7.5/doc/man/man4/voodoo.4.html

The Voodoo2 has 16bpp acceleration and the driver provides accelerated versions of most operations except angled lines and stipples. Accelerated alpha blending with the Render extension is also supported as is DGA.

Voodoo 2 can reach 1024x768 providing it has at least 2Mb of frame buffer memory.

However, there is no legacy PC stuff or VESA support AFAICT.