feipoa wrote on 2020-03-24, 08:13:

I'm curious what would have been the market for dual socket 5/7 systems based on the 430HX back in early 1996? According to wiki, the 430HX was released in February 1996, but dual Pentium Pro chipsets, the 450KX and 450GX, had already been released in November 1995. I have benchmarked the dual P233MMX against the PPRO233 and the PPRO is substantially faster in everything except DOS-only games, e.g. Blood and Sega Rally. This got me wondering why anyone wanting a dual CPU configuration back in early 1996 would choose the 430HX over the 450KX/GX.

Is that Dual P233MMX vs Dual PPro? Or vs. Single PPro?

If you are referring to the benchmarking I did, I compared them in Windows 98SE, so it would all be a single CPU comparison. I will begin installing games on the Proserva V using W2K shortly and will see if using dual P233MMX chips in W2K offers any advantage over a single CPU in Win98SE.

As for my main question, I was wondering why a company might have picked a dual socket 430HX system over a dual socket 450KX or 450GX system. Not to mention the 440FX chipset came out a few months later, so I can see why the dual 430HX system didn't seem so widespread.

feipoa wrote on 2020-03-24, 08:36:

If you are referring to the benchmarking I did, I compared them in Windows 98SE, so it would all be a single CPU comparison. I will begin installing games on the Proserva V using W2K shortly and will see if using dual P233MMX chips in W2K offers any advantage over a single CPU in Win98SE.

As for my main question, I was wondering why a company might have picked a dual socket 430HX system over a dual socket 450KX or 450GX system. Not to mention the 440FX chipset came out a few months later, so I can see why the dual 430HX system didn't seem so widespread.

My guess is, that in NT4 (or would it be 3.51 at the time)? for workstations, that dual Pentium (Non-MMX, probably, I don't think there are any splitvoltage dual S7 boards?) it would still have an advantage over a single PPro setup, and price-vise be somewhat closer to a single PPro system than a dual? Guesswork, I know, but that would've been my thoughts.

I share your opinion, however, that there must be a reason, that dual processor HX'es are not that widespread 😉

This was on the Zenith Z-CD and contains, amongst other things, overviews on product postioning & target audiences.

From the documentation, it looks like they were targeting small businesses of around 50 users or less; noting a price range of $3500 - $4000; and billing it as a File/Print Server and "providing e-mail and time management functionalities." The document talks about upgradability, but I question the cost-effectiveness of that. I guess the Proserva must have been a lot cheaper than even a single-PPRO system?

H3nrik V! wrote on 2020-03-24, 08:42:

...for workstations, that dual Pentium (Non-MMX, probably, I don't think there are any splitvoltage dual S7 boards?) ...

Some Tyan 430HX boards have split voltage to the socket, but I think the official support only went up to 200MMX. The limitation is probably due to how much current the onboard voltage regulator can source. When using upgrade modules, the majority of current is coming directly from the PSU via the molex connector, so there shouldn't be an over-current issue for two MMX CPUs. This is not true when using Intel's MMX overdrive chips though, which still take power from the socket. This is probably why Intel didn't produce a 233 MMX overdrive, being that most motherboard's couldn't source enough current.

Low cost duel socket is where I would have placed it as well. Price of 2 PPro's would have been a lot higher then price of 2 Pentiums I would have thought.
Much like duel Slot 1 systems existed alongside duel Slot 2

I keep saying low cost but that doesn't make it a budget system, Remember the mid 90's was before the internet really kicked off, Netware had a major market share and is a pretty lightweight OS only demanding more RAM then a typical PC so was common to find installed on a generic PC. If a company had a server (or network) at all.

I'd say 9 times out of 10 more then 1 processer ever really justified the cost financially until virtualisation came along at the end of 2003 era servers. Until then it was only databases that made good use of a 2nd CPU or mail servers but they are basically just a big database containing all the emails

feipoa wrote on 2020-03-24, 09:30:

Some Tyan 430HX boards have split voltage to the socket, but I think the official support only went up to 200MMX. The limitation is probably due to how much current the onboard voltage regulator can source. When using upgrade modules, the majority of current is coming directly from the PSU via the molex connector, so there shouldn't be an over-current issue for two MMX CPUs. This is not true when using Intel's MMX overdrive chips though, which still take power from the socket. This is probably why Intel didn't produce a 233 MMX overdrive, being that most motherboard's couldn't source enough current.

Crazy, I hadn't imagined that a 200MMX actually used more power than a 200 Classic .. The lower VCore shoud actually make dissipation decrease, but apparently the 1.2 million extra transistors more than compensate for that ..

I originally guessed as you did, that the reduced voltage would make up for the difference, but according to Intel's datasheets for worst case current draws,

P54C - 200 with 4.6 A at Icc3, or approximately 16.6 W.
P55C - 200 MMX with 5.7 A at Icc2 and 0.65 A at Icc3, or approximately 18.9 W.
P55C - 233 MMX with 6.5 A at Icc2 and 0.75 A at Icc3, or approximately 21.5 W

Multiply each of those by two CPUs and its pretty clear that the onboard VRM's might not be able to handle the P233 MMX. I recall reading someone trying dual P233MMX chips on those Tyan boards and running into trouble.

For those who acquire parity EDO RAM for this system, the BIOS is limiting the EDO read speed to x333. I brought this up in another thread and found some solutions to get it running at x222 in DOS and Windows (w98se, NT4, w2K). EDO vs. FPM on 430HX

I ran some benchmarks on the Proserva running two Kingston P233MMX chips. I ran several game benchmarks in Windows 98SE, which only uses one processor, and compared the results on the same system running Windows 2000 SP4. The same graphics card was used on both systems. I also compared the results to a Pentium Pro 512K chip running at 233 MHz using the same graphics card.

The data, plus some notes.

The data in chart form.

A chart showing which operating system, W2K or Win98, yield better benchmarks on the Proserva V Plus running dual Kingston P233MMX chips. The number of games which run faster on Win98SE vs. W2K is about split in half, but the game which prefers W2K buy at least 25% is Shogo, and the game which greatly prefers Win98SE is Unreal (classic). I don't know if the second CPU in W2K is having any benefit in the games. I did a brief check with GLQuake and Quake 2 to see if the results dropped if I set the CPU affinity to 0, but in both cases, the frame rate increased 1fps when the second CPU was disabled. This may be within the noise threshold though.

Lastly, I compared the same games using my PPRO system at 233 MHz. Clock-for-clock, the PPRO excelled by a good margin, with strong exceptions being DOS-only games, such as Blood and Sega Rally. In this chart, I am also comparing results taken by user F2bnp, which ran another fantastic 233mmx, ppro233, and PII233 comparison here: Pentium Pro 233 vs Pentium MMX 233 vs Pentium II 233 . Since we both didn't use all of the same games, there will be some gaps in the chart. The two areas which struck me the most were by how much faster the PPRO was compared to the MMX and also by how much more benefit my results show favour for the PPRO compared to F2bnp's results. In his tests, he is using a Voodoo3, which as I understand it, is a lot less CPU demanding than the Geforce2MX400 that I used, particularly with slower CPUs like the P233MMX. So it seems that the GF2 really needs a PPRO or faster to come to life.

Nice work, good to see some "official" figures of the PPRo vs MMX
More interesting to me was 98 vs 2k, and 2k winning at times. I always assumed I was taking a hit for running 2k, not that I cared the cool factor of 2 CPU's was more important to me.

If I'm honest with myself I know having 2 CPU's doesn't help, but I'm not so going to keep believing freeing one up while the other deals with the OS is gaining me a few fps 😉

I think at 233 MHz the second CPU is quite busy, taking care of all "non-game" things, that's why Windows 2000 is faster. Can you test Windows 2000 with one CPU disabled (you can do it via the /numproc option in boot.ini)? The Pentium Pro is running at 66 MHz FSB?

Yes the Pentium Pro system is using a 66 MHz FSB. There is no higher option, hidden or otherwise.

What exactly do I put into boot.ini and where? I've never heard of this option. I didn't want to remove a CPU because sometimes Windows gets stuck in uniprocessor mode. Also, for some games, I couldn't get the Task Manager open to set affinity, so this command you mentions may solve both of these problems.

I think /numproc only applies to OS above Vista like Xp and Win2003 but could be wrong. Available boot.ini switches:
/numproc Instructs the operating system on how many processors it is capable of using.
/onecpu Causes the operating system to utilize one of the processors in a multi-processor computer.

https://support.microsoft.com/en-us/help/8337 … dows-server-200
/numproc=number
This switch sets the number of processors that Windows will run at startup. With this switch, you can force a multiprocessor system to use only the quantity of processors (number) that you specify. This switch can help you troubleshoot performance problems and defective CPUs.

see also this: https://www.pantz.org/software/windows/bootin … upswitches.html

added: ok maybe am wrong, looks like those two swictches work in NT4 and Win2k:
http://franck.kiechel.free.fr/dbr_eng/BootIni.htm

always found it very curious how poorly build performs on P6 clock-for-clock, the engine is very much P5-optimized. according to ken silverman, the partial register stalls were responsible for this: https://github.com/jonof/jfbuild/blob/master/doc/build2.txt

so he did fix this in late versions and also added some drawing optimizations for P6, but all the retail games are stuck on the earlier engine version of course. however windows source ports should include those optimizations and be free of any 16-bit code i suppose, so i'd expect the ppro to do much better there.

feipoa wrote on 2020-03-31, 01:09:

Yes the Pentium Pro system is using a 66 MHz FSB. There is no higher option, hidden or otherwise.

What exactly do I put into boot.ini and where? I've never heard of this option. I didn't want to remove a CPU because sometimes Windows gets stuck in uniprocessor mode. Also, for some games, I couldn't get the Task Manager open to set affinity, so this command you mentions may solve both of these problems.

Open boot.ini, copy the line that you have under operating systems and add /onecpu at the end. Like in the screenshot:

Restart and Windows should display a boot menu, choose the option you have added and hit Return:

This shouldn't affect the hal Windows 2000 chose. You still should be running the multiprocessor HAL.

Unfortunately, when I use the /onecpu switch, Windows 2000 hangs either just after logon or just before. The very first time, it was able to boot and run GLQuake & Quake 2, but hung on Quake 3. Several attempts at rebooting failed, but if I used 2 CPUs, it booted up fine. Perhaps there is some kind of CPU affinity table that I can preset the game executables to use just one CPU?

feipoa wrote on 2020-04-01, 09:32:

Unfortunately, when I use the /onecpu switch, Windows 2000 hangs either just after logon or just before. The very first time, it was able to boot and run GLQuake & Quake 2, but hung on Quake 3. Several attempts at rebooting failed, but if I used 2 CPUs, it booted up fine. Perhaps there is some kind of CPU affinity table that I can preset the game executables to use just one CPU?

This is odd. Setting affinity won't matter, because what I was curious about was OS overhead. GLQuake and Quake 2 are single-threaded, Quake 3 without r_smp also, so they will only use one core. Windows 2000 will use the second CPU, that's something Windows 98 is unable to do. That's why I wanted to compare, whether the Quake's are still faster on 2000, when both the OS an game share CPU time.

AFAIR; on NT4.0, there was this "uptomp.exe" or something that like command, that made the OS MulitProcessor upgraded - (used to upgrade a system, where an extra CPU had been installed). But after running that command, there was no way to run the system uniprocessor again, as the whole HAL was replaced. Maybe something similar is the case with Win2k?

Perhaps. Or maybe something related to the motherboard's hardware configuration. I don't know, but before the system refused to work with /onecpu, I noted the Quake 1 & 2 scores and they were 1 fps faster than in dual mode, which is within the error bars, so might not mean anything. I'd have thought that at least Quake 2 would have slower results with 1 CPU, but that wasn't the case. Maybe a test like this needs to be run in software mode.

Quake 1,2 and 3 engines is using only one cpu, even they are launched under NT kenel OS which know 2 cpu's. Quake 3 has an patch to use the 2nd cpu's but it was experimental and the second cpu was used only for audio. It was until the Quake 4 engine when 2 cpu's could be used in games.
In regards with PPro vs Pentium performance, the FPU in PPro is more powerful than the one in Pentium processors, that's why in games which demand a lot of FPU power the P6 is the winner.
Speaking the prices, at launch, the cheapest Pentium Pro costs around 1000 usd ( 150 MHz) and 200 MHz 256k costs 1200 usd. When Pentium 233 MMX was launched, it costs around 600 USD, but price dropped in 2 months to 300 USD . In the middle of 1997, when Pentium Pro 200 with 1MB cache L2 was launched, it costs over 2000 USD, way above the costs of the most expensive Pentium II cpu. Of course, high price was because of low yelds but other reason was the fact that only P Pro was able to be used in 4 way cpu servers.
Not to mention the chipsets, for 1-2 cpu P Pro had 440 ZX, and after FX, which was more expensive than 430 HX. An tipical 2 cpu 440 FX board costs around 400 USD, only the chipset costs around half of that price.
For 4 way servers, the chipset used was 440 GX.

That was the reason for 2 way Pentium machine, the costs.

To resume, in presented games, 2nd cpu is not used by the game itself. The gain by disabling one cpu in OS for running games could be because the OS kernel doesn't have to manage two cpu's and is possible some access to ram to be also quicker. But from my point of view, is not worth the trouble for 1 fps or so.

Anyway, congrats for the tests, P Pro it was an great cpu overall.