Wow, that sucks.

RG100 I just dug through my email and found the invoice for my purchase of the 386DX40 CPU's.
I bought 3 of them from ic-china - the same seller you bought these from.

All 3 were the PQFP type mounted on a small PCB but it wasn't 'cut' or anything, and they definitely had pins on the bottom.

Here is the link from my invoice for the auction
http://cgi.ebay.com.au/ws/eBayISAPI.dll?ViewI … ME:L:OU:AU:1123

So unless he's changed it since I bought them, it appears to be exactly the same. I bought 3 of them, all were perfectly usable with pins on the bottom. I even left him positive feedback.

I can only assume that he had a stock of 'cut' pieces and working CPU's and he gave out the working ones first. However he should really have made this clear in his description. There is nothing to suggest they are for collectors only or are not working.

I'd be asking him for a 'please explain'.

Still, with due respect... On the ebay page, you can see in the picture that the PCB the 386 is mounted on has jagged edges and that the chip isn't centered. Also, on the right side, you can see printed numbers that were cut short and on the lower left side, you can see a surface mount component of some sort that didn't survive the hacksaw.

In the past, plenty of times, I've missed some pretty obvious red flags myself. However, those red flags are still present and fully visible. Though on the bright side, if you're handy with a soldering iron and encounter a slower 386 motherboard, maybe you could replace the chip and clock crystals.

You're right, but the seller could've at least mentioned these were cut out of the pcb. He wrote his advertisement to make it seem as if they were the "regular" AMD 386DX40's with a pcb.

I agree with you that you should always do your homework before making an investment, but frankly, even I didn't see it right away until I zoomed in the picture 😐

Edit:You'd need to be one helluva skilled solderer to solder those 100+ pins correctly. heck, you'd probably need a microscope or something!

Alphakilo470 wrote:

Still, with due respect... On the ebay page, you can see in the picture that the PCB the 386 is mounted on has jagged edges and that the chip isn't centered. Also, on the right side, you can see printed numbers that were cut short and on the lower left side, you can see a surface mount component of some sort that didn't survive the hacksaw.

In the past, plenty of times, I've missed some pretty obvious red flags myself. However, those red flags are still present and fully visible. Though on the bright side, if you're handy with a soldering iron and encounter a slower 386 motherboard, maybe you could replace the chip and clock crystals.

You're absolutely right. Of course a buyer should be more careful, and take account of the smallest detail, especially in this kind of trade. But there's something else that I insist that is important. If you search on ebay for a 386 cpu, you're going to find some ceramic pga ones, or some pqfp on that small board which converts them to pga. And sometimes you may find some desoldered pqfp ones(which usually mention that are for collectors). The fact is that our chinese friend's cpus are cut in such a way that resemble very much the converted pqfps. In my opinion this is misleading.

I agree with the 'buyer beware' sentiment but part of the blame lies with the seller as well. RG100 did ask about that ebay seller and I made the offhand comment that I think that's where I got mine from. I didn't check and confirm at the time but that's kind of irrelevant because I would have told him that I got 3 working CPU's from him in that case. Sorry RG100 if you bought these based on my advice.

Even the picture is difficult to judge, you'd have to look pretty hard to see that its been cut and its not just a PQFP CPU mounted on a PCB with pins on the bottom- which was a common configuration for these CPU's. Look at my picture in the OP of this topic for example. If you browse on a small screen or a mobile you could easily miss it.

There are plenty of sharks and dodgy sellers on ebay but 'others do it too so its okay' is hardly an excuse. This seller should have made it clearer exactly what he was selling, I think poor ebay feedback is warranted in this case. The fact is he is selling two different products on the same auction for the same price. One a non-functioning collectors item and the other a working processor.
Maybe ask contact the seller and if he has proper CPU's with pins he might swap them?

Tetrium wrote:

Edit:You'd need to be one helluva skilled solderer to solder those 100+ pins correctly. heck, you'd probably need a microscope or something!

No, and no. I think you misunderstand how it's done... you may want to look up a tutorial on QFP soldering sometime, it's really not difficult. A bit tedious maybe, especially if you don't have a proper rework station, but it's certainly doable with a regular soldering iron and a jeweller's loupe.

Very tedious! Still, I find it somewhat relaxing and a little fun. Something about melted metal that I find appealing.

As for the eBay auction, Donut, I do agree with you. Merely putting "on board" is insufficient considering (at least back in the 90s) several manufacturers would buy surface mount chips and put them on PCB. I have one example in my chip collection... It's a 233mhz Pentium MMX in tape carrier package mounted on PCB with pins to fit in Socket 7.

I thought Tetrium was referring to making a PCB adapter with pins on the bottom to fit it into a motherboard socket, like the board that RG100 bought. That pin grid array will probably be a pain to solder.

Otherwise these chips are useless unless you get a motherboard with an QFP solder pad for the CPU, I'd imagine most of these would already have a CPU occupying it.

RG100 if you want I will test and send you one of my spare 386DX40's for price of postage. I'm in Australia though so it might take a while to get to you.

Hey everyone! 😀 I appreciate everyone's commentary on this subject. @Donut King, no worries mate!

I'm still trying to shake off a winter bug, so I'll be brief. I submitted an offer to buy some of these AMD 386s on ebay. The seller wanted $30 each, and I offered $20 each. My offer was accepted, and I also bought some Intel 386DX-33s. Hopefully, this unfortunate situation has now been resolved, and I concentrate on the exciting prospect of building and testing some old vintage 386 PC hardware! 😀

Omg that's EX-PEN-SIVE!!!

Hopefully, this unfortunate situation has now been resolved, and I concentrate on the exciting prospect of building and testing some old vintage 386 PC hardware!

Amen mate! 😉

The 3 boards I ordered are finally in my living room. All 3 batteries show no sign of leakage! 😁
I'll remove those batteries later when I got a couple hours to spare 😀

Tetrium wrote:

The 3 boards I ordered are finally in my living room. All 3 batteries show no sign of leakage! 😁
I'll remove those batteries later when I got a couple hours to spare 😀

Having fun? 😁

I have just unpacked mine. I put 4 sticks of 1MB SIMMs in to it. They have 70 written on them. I guess that's the speed. I tried a Tseng 4000 ISA video card, and with the mobo's bus clock jumpers set to "/6" mode, I get 16.1 in 3DBench. I tried the very fast "/4" bus clock option, and the BIOS POST message area tells me that the mobo is upset about "bad cache". It subsequently disables it. I guess the 20ns cache isn't fast enough to cope with the "/4" bus clock option. Pity.

The evil barrel battery is leak-free. It even works! After lots of trips to the BIOS to fiddle with the options, it remembers all of my changes.

Edit: I ran Speedsys. I've only got 4MB installed at the moment, and so Speedsys tells me it can't run several tests. I'm waiting for some more memory to arrive, so I will rerun it then. But it did give me the overall system speed. It says 9.18, which is extremely respectable. That's using the "/6" bus clock option. BTW, "/6" = 13.333 system clock speed.

Here is Speedsys, with the bus clock jumper set to "/6", which is 13.333 MHz.

Edit: I found some more memory. Another 4MB. So, I've got 8MB now. 8 lots of 1MB. The other 4 lots were labelled 80, so I guess it doesn't matter if bank 0 = 70ns, and bank 1 = 80ns. The mobo seems to behave OK.

Also, I can boost the cache chips from 128MB to 256MB, but I don't know what to do about the Tag chip. Do I need another one, if I stuff the empty cache sockets with more memory, to boost it from 128 to 256?

Interesting that your CPU score is significantly higher than mine. What have you set your wait states to in the BIOS?

I couldn't get my system to boot with the /6 divider but then I've got a fully loaded system, with the SB Pro and Music Quest MPU401 card. The actual error I was getting related to Video RAM refresh failure so I can only assume the video card didn't like it. Bit strange since I have an ET4000 as well but I guess that older hardware for you.

To upgrade to 256KB of cache you will need 9 32KB SRAM chips. The TAG RAM needs to be replaced with a 32KB SRAM chip as well, it can be the same as the 8 memory chips. Look on page 2-7 of your manual. The 10th smaller chip can be left as is.

Thanks very much for the cache info! I will be able to do this tomorrow, and then I'll retest everything. I have a suggestion about your hardware. Looking at the SIMMs photo in your original post, they may be 100ns. It looks like they say "bla bla 10". I think the 10 means 100. If they are "slow as slugs", my suggestion would be to replace them with with FPM/80ns memory. Also, here are my BIOS settings -

Edit: I've just managed to improve my 3DBench score, from 16.1 to 16.6. I have done this by using a Diamond Speedstar 24x 1MB video card.

I have upgraded the 386 mobo to use 256Kb of cache. This has improved my 3DBench score from 16.6 to 16.9.

Hmm that's very weird, I didn't get any increase from upgrading the cache.

I think you are right about the memory, being 100ns they wouldn't be FPM and the manual actually says on page 1-2 that it supports 80ns FPM memory.
My understanding of the old SIMMs before FPM and EDO came along is that they ran at the clock speed of the system bus and the ns rating was simply what clockspeed they could handle. So if 100ns SIMMs could run at the same clockspeed as 80ns SIMMs (with the same latencies/wait states) then performance would be the same. Of course the lower ns rated SIMMs could handle lower latencies/wait states and could be made faster. But those 100ns SIMMs I'm using run at minimum wait states already. Maybe someone could confirm this? Of course, once you add FPM and EDO into the equation things are different.

I have a stack of 30 pin SIMMs but I will have to dig through it to find 8MB of FPM SIMMs to test. Plus getting to the RAM slots in this case is a total nightmare 🙁 They are right under a metal beam and I'll have to remove the power supply as well.

I also found another ET4000 in my pile of bits, it looks older but I'll try it anyway and see if it is happier at the faster ISA bus speed.

Regarding the weird increase/no increase in 3DBench, with more cache. I have an idea. Try rebooting several times, or powering down then powering up again. Rerun 3DBench, and other tests. You may find the results change in an unpredictable way.

I think I've probably "hit max" on this system. I have got it to behave normally, with the bus jumper set to "/4". That's a silly fast 20 MHz bus speed setting. I got it to work by using a Cirrus Logic 5422 video card. 3DBench still gives me 16.9, but I do get a tiny increase from SpeedSys. It's now 9.23.

I wonder if my Speedsys score is ~9.2 because I'm using this thing? I doubt it. I expect it's because I'm able to increase the bus speed successfully. BTW, I'm waiting for some of those 386s mounted on special pin packages, which we discussed earlier. As soon as they arrive, I'll retest everything, just to make sure it isn't the type of CPU which is affecting the benchmark results.

I've just tried changing the osciallator, and it works. I removed the 80MHz osci, and replaced it with a 66MHz unit. The 386DX-40 chip now runs at ~33MHz. I could not use the mobo at the maximum 20MHz bus speed, with this slower osci. The BIOS POST display complained about the cache being "bad". Consequently, I lowered the bus speed to ~13.333. That worked. 3DBench = 13.3.

Here is the new Speedsys screen, showing slower results. I also have a 50MHz osci, which I will try. I will then do a big test, involving all 3 oscis, and adjusting various other settings. My hope is to get 3DBench to cover all values from 1 to 16.

BTW, I have found a website that sells 14 pin 100 MHz oscis. Hehe! 😈 😁