Brief update:

1) The x222 Memory Read Burst Timing hack for EDO only seems to work with my single-sided 64 MB parity SIMMs. It did not work with my EDO double-sided 64 MB parity SIMMs. Unfortunately, I only have four sticks of single-sided 64 MB EDO parity simms and eight sticks of double-sided 64 MB EDO parity simms. The boot-up memory counter counts all 512 MB, but I haven't tested them in operating systems because each OS is set for x222 read timings. There is no way to skip the boot-up memory count, 256 MB is about as much patience as I have anyway so I won't be procuring more single-sided EDO SIMMs.

2) I tried to run a pair of Intel MMX Overdrive CPUs in this board (PODPMT66X200), and while the CPUs work individually, the system doesn't show the pair on boot - it just shows one. I guess Intel disabled SMP on their MMX overdrives. So looks like if you want to run dual MMX CPUs, you need to use upgrade VRM's, like those from Powerleap, Evergreen, Kingston, etc. Alternately, you can still use an MMX overdrive in the CPU1 socket and the Kingston in CPU2 socket and still use SMP.

3) I re-arranged my systems so that I can have matching Voodoo2 SLI's in the NEC system. It took some time to find reasonably priced Voodoo2's ($50-$60 each) as the market for this vintage garbage has gone COVID crazy.

4) re: r_smp at 0 and 1 in W2K. I kept running into trouble running Quake 3 v1.32 on this system in W2K. However, I did run it in NT4. r_smp 1 will not work with the Voodoo2 SLI for whatever reason, but I did try r_smp 0 vs. r_smp 1 using the GF2MX. r_smp 0 = 14.7 fps w/snd and r_smp 1 = 15.1 fps w/snd, so about a 3% increase. I did boot NT4 with the /onecpu flag and the r_smp 0 result was the same, that is, 14.7 fps. This implies to me that having the second CPU taking care of background operating system tasks didn't help any.

Thanks, good info. Wow you found two working Voodoo 2's for that ? Great find, seems like everyone wants too much and have read most Voodoo 2 do not work after receiving them (from the reviews/comments).
Have not messed with mine since Win2k puked a few weeks ago and will not allow me to do an in-place upgrade/rebuild to fix it. Win98 and DOS still run well.

Not quite. I found two Voodoo2’s of different brands, so I took the STB Voodoo2 12 MB from my IBM 5x86c and the STB Voodoo2 12 MB from my NexGen PF110 and put them into the Proserva. Then put the newly acquired Voodoo2’s into the IBM/NexGen systems. Luckily the newly acquired Voodoo2’s work. It can be risky buying Voodoo2’s on eBay as it really is hit or miss for functionality. A lot of sellers these days try to leave the eBay Condition field blank and say “as-is” in the description, but apparently no matter what is written to in the description to warn against “no returns”, merchandise is still returnable for a full refund.

I purchased one of these from the same seller and wanted to share my random findings, mainly just to document what I've learned for others who stumble upon this forum post as I did.

I have installed two SL27S 233Mhz MMX processors without interposers -- we used to do this quite a lot back in the day and there's really no threat to these running at 3.3v.

I've installed Windows 2000 SP4 for now. I have a 3c597-TX EISA Fast Ethernet card installed.

I currently have 4x64mb Parity SIMMs installed (waiting for more). I am waiting for an assortment of 128mb modules to show up to see if the board might support them.

For 233MHz operation I have set the bus to 66MHz and jumpered J1A and J2A to the 1-2 position (shown as invalid/not listed on the case sticker).

After upgrading to the 1.00.14WL14 BIOS (tried both the NEC or Zenith version -- which do differ slightly), Windows 2000 no longer saw 2 CPUs. The motherboard reported the stepping for both, but beyond that the operating system had no idea about the 2nd unit. I could not find any options, drivers or configuration that would fix this. I had to re-flash to the original (1.00.10 WLB08) to get it working again -- I was very glad someone prior had uploaded a backup of that to the forum.

I did not have any trouble using the newer EISA tools from the Zenith archive with the older NEC BIOS to configure my network card.

The maximum FSB frequency that can be generated using the stock clockgen IC is 66MHz, but this should be trivial to replace with a small mod board.

I recall having a similar issue with W2K not seeing the second CPU. The issue is due to flashing the BIOS causes APM to be enabled by default. You must disable APM. This was discussed previously in the thread. Re: NEC Proserva V Plus (dual socket 7 system) (have a lead on multiple of these)

Did you check the voltage regulator on the motherboard to ensure it can cope with running two Pentium 233 MMX chips at 3.3 V? Even if the P233MMX chips were running at
their nominal voltage of 2.8 V, I calculated that the onboard VRM cannot handle this to within a safe margin.

The regulator is a Power Trends PT6501A, which is a switching regulator at a fixed 3.3V output with 8A output max. This translates to approximately 26.4 W for running the CPUs. From the Pentium MMX datasheet using the worst case example (3.6V for Vcc3 and 2.0V for Vcc2), the max current for Icc3 = 0.75 A and Icc2 = 6.5 A. For two CPUs, this is 43.1 W. Similarly, for a P200MMX, the worst case requirement is 37.7 W, and that is with running them at 2.9 V, not 3.3 V. Even looking at just the ordinary Pentium P54C-200, worst case is 33.1 W, which is greater than 26.4 W. To fit the 26.4 W requirement, the maximum CPU to be used in SMP are P54C-150. Could it be that there are two outputs on the PT6501A which allow for 8 A each? Where's my mistake?

Do you think if we connect the motherboard's 3.3 V auxiliary connector to a PSU with 3.3 V output that the board has logic to bypass the PT6501A? Typical 300W PSU allows for around 20 A on the 3.3 V rail.

But if you use the P233MMX chips in an interposer module, they normally are powered from the power supply directly and have a VRM onboard, which bypasses the motherboard's VRM. So the issue becomes nil.

I don't think I've run P233MMX chips at 3.3V on both rails before. Do you need to add active cooling to the heatsink for this, or is the front case fan with stock heatsinks sufficient?

It sounds like you have confirmed that J1A, J2A at 1-2, 1-2 is the 3.5x multiplier? I haven't checked this myself because the interposers for my P233MMX chips have a mini DIP switch to set the multiplier.

You mentioned that you have some 128 MB modules on order. I can confirm that 128 MB modules work, but they must have parity. I tested up to four 128 MB FPM modules; it booted fine. However, it would be slightly more optimal to have EDO modules. Were you able to find unbuffered 128 MB EDO modules with parity? I was not. I tried 128 MB EDO and FPM modules without parity, but the display stays blank at power-up.

Feipoa,

I could have sworn I scoured the whole thread, but I must have missed the discussion about APM. I did actually notice that it was enabled, but didn't think anything of it. I will try that and see what happens.

You are correct that the PT6501A technically can't handle the power requirements of the two chips at 233MHz; the regulator can output at most 8000ma, where the CPUs can potentially draw 6500ma each (roughly, according to the datasheet) -- and that's running at the expected 2.8v, it will be potentially higher at 3.3v. I'm not sure realistically if the CPUs ever spend much time at this high of a current draw though, but ideally it should be addressed if you want a reliable system. A set of interposers is likely the best option if you have/can afford/find them. I think it's probably okay just to mess around with it like this, though.

As for the 3.3v aux connector, I would be surprised if they implemented such a thing. I suspect the voltage drop through the connector and over to the CPUs would be too great at such a low voltage. I need to spend more time investigating the board though in order to tell you for certain how it's set up.

As for cooling @ 3.3v, in the past I've mounted a fan cooler just to be safe, but in this Proserva I'm playing with I'm currently using the fan-less ones that came with it. The location of the chassis fan being so close helps. I haven't tested with any actual temperature probe, but based on touch they feel fine. I've left it doing some burn-in tests and nothing remarkable has happened (yet).

I believe the J1A/J2A = 1-2 setting technically puts the multiplier at 1.5x -- at least, if you were using the chips it was designed for. On the MMX chip this setting is interpreted as 3.5x.

As for the modules, I grabbed a few different listings, and past experience has taught me generally not to believe that the seller always knows or is truthful about what type of RAM it actually is. I will test them on my RAMCHECK and report back exactly which types I end up with and which ones work in the Proserva.

I am primarily interested in seeing what happens if you try to exceed 512mb -- any input on this?

RAMCHECK - what is this?

OK, good to know about the 1.5x setting.

It is too bad the PT6501As' are fixed 3.3 V regulators. But perhaps there is still a voltage set resistor in there which can be changed to achieve 2.8 V output.

socket 5/7 interposers aren't expensive if you have patience. $25-$50 usually.

I don't think I had the guts to exceed 512 MB. I only have four parity 128 MB SIMMs, but I suppose I could add two 64 MB modules to it. I'll leave that for someone else though.

EDIT: Have you run anything in Windows 2K on your Proserva yet? I found that Quake 3 in W2K doesn't like this system.

RAMCHECK is a piece of test equipment for testing DRAM, made by the (as of earlier this year) now defunct Innoventions, Inc. In the most basic mode it will do a function test and report back the basic stats of the stick -- both what it's jumpered/programmed for, and what it actually tested as. A more advanced (slower) test can be ran from that point that goes through dozens of algorithms, including under/over voltage and load tests.

I have a RAMCHECK LX along with the adapters to test almost everything from 30 pin SIMMs up to DDR3 on the same machine, including some specialized ones for testing individual chips. It's fast, testing SIMMs in seconds and large DDR3 sticks in a few minutes. You can also use it to view/reprogram SPD data, test a stick at voltage and frequency settings that would technically be way out of spec (overclocking, undervolting, etc), generate and print reports -- just generally a handy tool if you're running a repair shop or otherwise selling/manufacturing RAM sticks.

If you're patient you can usually find older SIMMCHECK (30, 72 pin SIMMs and early SDR DIMMs) machines and adapter sets for <$100. Used RAMCHECK and RAMCHECK LX machines and adapters do occasionally show up, but it can get pricey fast. The SIMM adapter and the DDR3 and DDR4 adapters are rare to find used.

I don't know if the PowerTrends module is only powering the CPUs or what all components the board is powering from it. That would need to be determined first, I think, before we could even consider adjusting the voltage.

Personally, I'd be tempted to design a little board to pop in it's place based on some modern switching regulator, but I don't know how willing most people would be to install one in theirs. I did grab a spare motherboard for the Proserva that I intend to mutilate with my experiments.

Can your RAMCHECK LX test the response time of DRAM and SRAM? It would be interesting to see what drift there is as these modules get up in age. Will a 60 ns stick of EDO memory still guarantee a 60 ns response 30 years later?

The ability to reprogram SPD could be useful. A lot of memory manufacturers would use, e.g., PC133 CL2 IC's but program SPD for CL3. I've even seen some that use PC133 CL3 IC's but only have PC100 CL3 in SPD. While most consumer motherboard BIOSes let you overwrite SPD, a lot of OEM ones do not, like Intel and Dell Workstation boards.

I recall looking into this Innoventions company 10 years ago. I was considering buying their SRAM tester, which according to their support e-mail, was able to test the SRAM's response time. I think to within 8 or 10 ns. This would be useful for quantifying the response time of these Chinese counterfit 128kx8 SRAM DIPs. The guy wanted way too much for the unit. I think it was around $1000.

But now that they are defunct, who's taking over their brand and service?

I'd be willing to remove the Proserva's VRM if there is a suitable solder-in replacement. However, the motherboard's CPU traces aren't wired for split rail, so the VRM should be adjustable so that one can find a happy medium between the 2.8 V core and the 3.3 V I/O.

I admittedly didn't read all 18 pages of this thread, so apologies if this was already answered -- Could you run other Socket 7 processors in this system like a K5 or K6?

chublord wrote on 2020-09-04, 17:31:

I admittedly didn't read all 18 pages of this thread, so apologies if this was already answered -- Could you run other Socket 7 processors in this system like a K5 or K6?

feipoa wrote on 2020-02-21, 09:14:

When using a split-rail VRM interposer, I am pleased to report that a AMD K6-III-450AFX chip will function properly on this motherboard at 400 MHz. I tried a K6-2+ chip, but the screen stayed blank, so perhaps only the non-plus chips work. I am also pleased to report that my Pentium 266 MMX Tillamook chip also functions properly with L2 cache enabled. Seems like very few socket 5/7 boards run properly with a Tillamook (SL2Z4), but this one does. Tested it in DOS and Windows with a few games. Note that the ADS# and ADSC# pins need to be bridged on the the SL2Z4, which can be done on the top surface. I also have the 4x pins bridged (pins from grid W33 & W34), but not sure if those were necessary.

I did not try the K5 or original K6.

chublord wrote on 2020-09-04, 17:31:

I admittedly didn't read all 18 pages of this thread, so apologies if this was already answered -- Could you run other Socket 7 processors in this system like a K5 or K6?

I dug through all the other Socket 7 CPU datasheets I could find and near as I can tell the Pentium and Pentium MMX are the only ones that ever supported dual processor configurations. That said, other chips might work fine in a single processor configuration, though there are plenty of other hurdles you'll potentially run into with that.

I personally haven't tried any others, as I have no interest in running this system with only one CPU.

feipoa wrote on 2020-09-04, 12:26:

I recall looking into this Innoventions company 10 years ago. I was considering buying their SRAM tester, which according to their support e-mail, was able to test the SRAM's response time. I think to within 8 or 10 ns. This would be useful for quantifying the response time of these Chinese counterfit 128kx8 SRAM DIPs. The guy wanted way too much for the unit. I think it was around $1000.

Haha, yeah, they were certainly proud of their work. I guess be glad you never asked about a RAMCHECK kit ($6-8k at one point). To be fair, those prices aren't too bad for a manufacturer or medium sized computer shop -- we certainly made our money back with it over the years.

I did manage to pick up a used SRAM tester right as they were closing down for about $500. That's exactly what it's great for -- I ended up with a huge batch of Chinese re-marked 128kx8 SRAMs that, surprisingly, are actually what they say on the label, but don't know how to get rid of them since most people seem to be afraid of re-marks out of principal. I use them around the shop without any trouble.

feipoa wrote on 2020-09-04, 12:26:

But now that they are defunct, who's taking over their brand and service?

I asked about this when they were closing down, and specifically wanted to take over repair and service for their older products since we'd already been doing it unofficially for years, but there was no interest at all. As far as I know after the owner died everything fell apart and the current executor either doesn't know how or doesn't care to deal with the company.

feipoa wrote on 2020-09-04, 12:26:

I'd be willing to remove the Proserva's VRM if there is a suitable solder-in replacement. However, the motherboard's CPU traces aren't wired for split rail, so the VRM should be adjustable so that one can find a happy medium between the 2.8 V core and the 3.3 V I/O.

Need to check and see what all runs off that regulator first. If I recall correctly, the datasheet for the MMX implies you could get away running both rails right at 3 volts and still just barely be in spec. Got to account for voltage drop too, not sure if there's any provisions for sensing on the board.

That is certainly interesting - you bought the SRAM tester and they halved the price for you! It is too bad that they are gone for good.

Could you take a photo of the re-marked Chinese SRAMs that actually conform to their datarate? It might put people at ease.

Hello all! I'm the guy that Cyclone3d bought the first computer from that started all this. It appears as though you've been having quite a bit of fun with these computers with just enough frustration thrown in to keep it interesting. Furthermore, I have recently acquired some DEC Talk speech synthesizing cards that were used in these machines. Also, I'm going to post a photo of one of the NWS AWIPS work stations where the NEC ProServa was used. Enjoy!

Proserva: I suggest you donate those 3 faceplates to John as he has 3 systems nobody wants because of the damaged bezels you sent him. He isn't even going to break even on the whole lot.

No sales or trading allowed on Vogons. Please respect this rule.

Dominus wrote on 2020-09-24, 18:39:

No sales or trading allowed on Vogons. Please respect this rule.

have you considered adding this rule to the rules page so people will actually know about it?

maxtherabbit wrote on 2020-09-24, 20:23:

Dominus wrote on 2020-09-24, 18:39:

No sales or trading allowed on Vogons. Please respect this rule.

have you considered adding this rule to the rules page so people will actually know about it?

Yeah.. we get these posts all the time from mods because noobs are not aware of the rule that everybody that has been here a while knows about but isn't actually posted on the rules page.

How much time does it take for the mods to post about the rule all the time compared to a 15-second edit of the rules page?

AFAIK it's one of the rules new users need to agree when they register. But it could be that it is not explicitly mentioned and only the "no auction" rule is there.

I cannot edit the rules page