I'm pretty sure NexGen will be the slowest, I remember watching a video (possibly by HighTreason?) where the guy benchmarked it and found it to be competitive with fast 486, but noticeably slower than Pentium.

Could it be due to using floating point tests?

Nx586 has no FPU, so with emulation it will surely be disaster in Quake. However, its Pentium rating - 90 is quite ambitious if you can trust that.

mpe wrote:

Could it be due to using floating point tests?

Nx586 has no FPU, so with emulation it will surely be disaster in Quake. However, its Pentium rating - 90 is quite ambitious if you can trust that.

He didn't actually test Quake, it was slower than Pentium 66 across the board in integer tests (Doom, 3DBench, PCPlayer, etc).

Edit: found the video https://youtu.be/AnHx2TCvpSc

mpe wrote:

I am working on a project where I would like to compare early Pentium and Pentium-like systems and their alternatives roughly ar […]
Show full quote

I am working on a project where I would like to compare early Pentium and Pentium-like systems and their alternatives roughly around 1994/95.

My test platforms are currently these:

DSC_3217 (1).jpg

- Pentium 60 and 66 in ALI-based PCI Socket 4 board

Even by So4 standards, that's an unusual chipset. Can't say what to expect...

- Pentium Overdrive 83 MHz in VIA VL-Bus Socket 3 board

Big FPU but slow 32b RAM access. Results here will vary considerably depending on exact bottleneck in a given test. Note by the way that it clocks down to 1x multiplier if you remove the heatsink/fan, so it will run at 33MHz if configured as on this picture, not at 83MHz 😮

- NexGen Nx586-P90 on NxVL VL-Bus board (wish I had the PCI one)

Don't want the VLB one? Give it to me 😜

Performance might be slower, but I consider this one of the most beautiful boards ever made, where the PCI board superficially looks like just another 1995-ish motherboard.

- Pentium 75 MHz in VIA MPX Socket 7 PCI Socket 7 board

VPX, not MPX. Just use EDO and ignore the (flaky, very early) SDRAM support. However the PLB cache is a big game changer, this thing will run rings around the rest because of it.

I will be using usual late DOS and Windows 3.1 benchmarks for comparison.

I am still thinking about replacing the Pentium 75 board with a more period correct option. Like Socket 5 (Intel 430FX with async cache) as I believe the VIA VPX with pipelined-burst cache and SDRAM support is giving the P75 an unfair advantage compared to the rest.

Eventually I would like to add also - AMD Am5x86-P75 133 MHz, AMD K5 (5k86) PR75 (75 MHz), Cyrix M1 6x86 PR90+ to the mix, but I currently don't own these CPUs.

If you're shopping anyway, include a Cyrix Cx5x86, a very different beast to the Am5x86, and clock for clock significantly faster. The IBM-made ones tend to be cheaper and run cooler/can clock higher.

Before I start benchmarking. Any guess about which platform you think will prove to be the fastest and slowest?

As it stands I'd expect the P75-VPX to win most benchmarks hands-down, but to be beaten by the K5 and (particularly) M1 when you add those. However running PCI at 25MHz means PCI-sensitive stuff may run better on another platform, and some clock-sensitive but bandwidth-agnostic ALU stuff may run better on the 5x86 as well.

Any opinions about which graphic cards to use? I am thinking about using Tseng ET4000/w32p for VL-BUS and S3 Vision968 for PCI. Or perhaps should I use VRAM-based Mach64 GX for both? I would like to stay period correct, use what would be a typical choice for a given platform without giving any handicaps/unfair advantages.

If at all possible I'd suggest using the same chip on both PCI and VLB. If you have two Mach64s, that would be the way to go.

ShovelKnight wrote:

He didn't actually test Quake, it was slower than Pentium 66 across the board in integer tests (Doom, 3DBench, PCPlayer, etc).

Edit: found the video https://youtu.be/AnHx2TCvpSc

A really nice video. Thank you for sharing it!

dionb wrote:

mpe wrote:

- Pentium 60 and 66 in ALI-based PCI Socket 4 board

Even by So4 standards, that's an unusual chipset. Can't say what to expect...

It is actually a tiny bit faster than boards with Intel Mercury chipset (430LX) as far as I know. Unlike the LX it supports EDO RAM which gives it a small boost.

dionb wrote:

mpe wrote:

- Pentium Overdrive 83 MHz in VIA VL-Bus Socket 3 board

Big FPU but slow 32b RAM access. Results here will vary considerably depending on exact bottleneck in a given test. Note by the way that it clocks down to 1x multiplier if you remove the heatsink/fan, so it will run at 33MHz if configured as on this picture, not at 83MHz

Yes. The PODP5V83 will likely be the underdog here. Even the NexGen should be able to easily beat it except when using FPU. It has double the L1 cache of the Pentium and slightly higher clock. However, the 486 bus and disabled V pipeline is crippling it a lot.

dionb wrote:

If you're shopping anyway, include a Cyrix Cx5x86, a very different beast to the Am5x86, and clock for clock significantly faster. The IBM-made ones tend to be cheaper and run cooler/can clock higher.

Will try to get one. Thanks.

dionb wrote:

As it stands I'd expect the P75-VPX to win most benchmarks hands-down, but to be beaten by the K5 and (particularly) M1 when you add those. However running PCI at 25MHz means PCI-sensitive stuff may run better on another platform, and some clock-sensitive but bandwidth-agnostic ALU stuff may run better on the 5x86 as well.

Yes. That's why I would like to reduce its advantage by using a 430FX board with async cache which was still the norm in 1994 I believe.

i430FX is a 1995 chipset. If you are aiming for 1994, it's i430NX to look for - which is old, slow and not supporting EDO. The introduction of the i430FX was the moment that the Pentium actually became an interesting CPU to buy in technical terms, it was vastly faster than the i430NX, and slightly but relevantly faster than competing SiS, Via and ALi designs.

I think the first batch of i430FX boards mostly didn't have the pipelined burst cache or EDO fitted. I'm curious how much better an i430FX with async cache is than an i430NX.

Later I may be able to provide some data from another weird Pentium board I own. It's an Acer M5 PCI/EISA backplane (with intel chips), combined with a P54 CPU card with a highly integrated Acer chipset. I can't seem to find an image of it to get the chipset model, but it was a 1994 product and an alternative to the Mercury/Neptune. The name of the CPU card is "SP54C-ALi". I just noticed there is another CPU card called "SP54C-ECC" which also uses the ALi chipset. Only difference should be two extra memory slots, so I guess the chipset might be ALi M1451 like your board...only on mine it uses an Intel chip for PCI.

Thats a pretty cool board you have there

Anonymous Coward wrote:

Later I may be able to provide some data from another weird Pentium board I own. It's an Acer M5 PCI/EISA backplane (with intel chips), combined with a P54 CPU card with a highly integrated Acer chipset. I can't seem to find an image of it to get the chipset model, but it was a 1994 product and an alternative to the Mercury/Neptune. The name of the CPU card is "SP54C-ALi". I just noticed there is another CPU card called "SP54C-ECC" which also uses the ALi chipset. [...]

I have the same base board in storage, but with an i430NX based dual socket 5 P54C Cpu card. I'll help with the benchmarking if that could be useful. Let us pray the board still works.

Could you provide me with some photos of your dual CPU card? I've been trying to find one.

See attachment.

I'll check my boxes to see what other early Pentium board I have. There's certainly one Socket4 430LX I have somewhere and a Siemens dual socket5 NX which I never got round to testing.

Thanks! That's great to have. At least now I have the part numbers for it.

I wonder why they used one LIF and one ZIF socket? It's pretty annoying to attach heatsinks without the plastic tabs.

I suppose they used the Pentiums with the glued-on heatsinks but I agree it is a strange sight.

I've seen a 430FX board with both SRAM DIP sockets as well as the COAST slot. I'd like to use one of these for Socket 5 CPUs. I can only hope it works just as I expect it (SRAMs in socket -> async L2 cache, module in COAST -> pipelined burst L2).

I am still shopping for the AMD K5-PR75 and a Cyrix chip (most likely Cyrix5x86-100GP if I can get one).

In the meanwhile I arranged CPUs roughly in the order I expect them to finish in my tests.

I think I found a more appropriate Socket 7 board for my tests.

This nice Intel Endevor board I purchased has an early P54C chipset (430FX Triton):

Please note the ultra-rare gold cap Pentium 100 in the socket 😉

I love many things about this board. It uses well-described DIP switches rather than fiddly jumpers. It has coin battery, no dead Dallas chips, cache in COAST slot which can be either asynchronous or pipelined burst and even integrated SB16 with true Yamaha OPL-3 (although I don't need one).

I dunno both of my COAST sticks are of pipelined burst type so I only hope I will be able to find asynchronous module to have more fun. Do they really exist?

Very nice board! I will be watching this thread with interest.

My very first Socket 7 motherboard (Abit PT5) was based on the 430FX chipset, so I have a soft spot for it.

Please note the ultra-rare gold cap Pentium 100 in the socket 😉

Still pretty uncommon, but not ultra rare anymore. A seller on eBay found a bunch of them. I have four.

mpe wrote:

I dunno both of my COAST sticks are of pipelined burst type so I only hope I will be able to find asynchronous module to have more fun. Do they really exist?

My FIC PT-2003 (430FX chipset) came with an async. cache COAST module. In fact, it's the only cache stick that works on my board, with all of the PB modules I tried it just doesn't detect any cache memory. Must be a motherboard revision thing, since other users with this board have working PB cache.

If I have time during the weekend I can post some pictures of it.

Just curious, do you thik that async module would work in an M919 486 motherboard as well?