VOGONS


Reply 80 of 91, by Jo22

User metadata
Rank l33t++
Rank
l33t++

Performance seems fine in my eyes. Around 10 MB/s is quite fast.

Let's remember, early XTs and ATs were directly working through the front side bus.
In such PCs, using memory on the expansion bus was causing no bottleneck. 😀

I'm curious, did this card run on an 8, 10, 12 or 16 MHz bus?
- I'm asking, because quite a few of these vintage memory boards did officially support up to 12 MHz operation.

"Time, it seems, doesn't flow. For some it's fast, for some it's slow.
In what to one race is no time at all, another race can rise and fall..." - The Minstrel

//My video channel//

Reply 81 of 91, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Jo22, I'm only looking at the speed of the memory on the expansion card, not the 4 MB RAM on the motherboard. Half-speed memory perf. for the expansion card seems pretty lousy to me.

This system is running on a 16.7 MHz FSB with a SLC2 upgrade kit running at 66 MHz.

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 82 of 91, by Jo22

User metadata
Rank l33t++
Rank
l33t++

Thanks for your reply. I understand, but does thinking in percentage really matter, considering these dimensions ?
If I do have, for example, a CF card that transfers data at 2,7MB/s vs 5MB/s, does it matter so much ?
The numbers might seem drastically different in percentage, but not in actual MB/s.
With merely a 2,3MB/s difference, it's about as fast or slow in practice. 😉

With such limited hardware, I'm personally rather concerned about features such as DMA (on memory boards, this equals capabilities like bank-switching).
A slower card, but with a low access time or DMA, might cause better overall performance by not bothering the CPU with copy operations.

That being said, I do fully understand your point of view and it's perfectly fine, also.
It's just that I do perceive things differently, I guess.
Time is -or at least it's perception is-, relative, after all. 😉

Edit: Never mind. My apologies for writing down things like this, in case it's annoying.
I don't mean to upset people, I just simply think that way and often just realize how my thought process may seem strange to others when it's too late. 😅

"Time, it seems, doesn't flow. For some it's fast, for some it's slow.
In what to one race is no time at all, another race can rise and fall..." - The Minstrel

//My video channel//

Reply 83 of 91, by Predator99

User metadata
Rank Oldbie
Rank
Oldbie

Its new to me cachechk runs on a 286? If yes, I can try if I have a board installed next time.

For me it makes totally sense the RAM on the card (connected via ISA bus) is slower than the onboard RAM (connected direclty to CPU).

Reply 84 of 91, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I did not read anything here which could be perceived as upsetting. It really boils down to one's personal motivations. I am new to the memory expansion card arena and am mostly trying to fully optimise systems for the fun of it. Are you saying that some of these expansion cards do DMA while others do not? If so, do you know which do DMA and which do not? I find it hard to grasp my mind around DMA when considering a memory card.

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 85 of 91, by Predator99

User metadata
Rank Oldbie
Rank
Oldbie

No, the RAM is not copied/transferred to another location, therefore there is no DMA.

These cards are intended for use in 286 systems. There are usually other bottlenecks than the speed of the XMS. But yes, these cards are slow. You can already see in in the POST screen when RAM counts. The 1st part for the onboard RAM is fast, when switching to the ISA RAM counting gets much slower.

In most 386 systems you can put 16 mb onboard, so these cards are useless there.

Reply 86 of 91, by feipoa

User metadata
Rank l33t++
Rank
l33t++
Predator99 wrote on 2022-02-14, 08:47:

No, the RAM is not copied/transferred to another location, therefore there is no DMA.

That is kind of what I thought.

Predator99 wrote on 2022-02-14, 08:47:

These cards are intended for use in 286 systems. There are usually other bottlenecks than the speed of the XMS. But yes, these cards are slow. You can already see in in the POST screen when RAM counts. The 1st part for the onboard RAM is fast, when switching to the ISA RAM counting gets much slower.

Yeah, I can definately hear the memory tick sound difference once it passes 3712 KB (BIOS shadow enabled) or 4096 KB (BIOS shadow disabled).

Unfortunately, it looks like cachechk requires a 386 or better. I haven't been able to find a program which works on a 386 and can measure particular memory addresses like cachechk can. There's CCT, but the furthest out it can go is 1.5 MB.

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 87 of 91, by Predator99

User metadata
Rank Oldbie
Rank
Oldbie

Back to topic...next card I got is a NCR one. Its is labeled
NCR -85 017-0035806
ASSY 017-0035447 D
SCHM 017-0035448 B

IMG_9207r.jpg
Filename
IMG_9207r.jpg
File size
1.4 MiB
Views
232 views
File license
Fair use/fair dealing exception
IMG_9208r.jpg
Filename
IMG_9208r.jpg
File size
1.39 MiB
Views
232 views
File license
Fair use/fair dealing exception

User GuillermoXT from dosreloaded.de owns this card, too. He uploaded a manual.
https://dosreloaded.de/forum/index.php?thread … 3256#post133256

16147966017274678308628860829171.jpg
Filename
16147966017274678308628860829171.jpg
File size
157.5 KiB
Views
232 views
File license
Fair use/fair dealing exception
20210303_151649.jpg
Filename
20210303_151649.jpg
File size
230.35 KiB
Views
232 views
File license
Fair use/fair dealing exception

Reply 88 of 91, by Predator99

User metadata
Rank Oldbie
Rank
Oldbie

The card uses some very special RAM modules I have never seen before. On the top there is a normal 30-pin SIMM for comparision.

IMG_9210r.jpg
Filename
IMG_9210r.jpg
File size
1.36 MiB
Views
229 views
File license
Fair use/fair dealing exception

NEC MC-41256A4A-15 8632KD
NEC MC-41256A5A-12 8625KD

Unfortunately my card currently refueses to work. The RAM get very hot and nothing is detected. On the back there seems to be some kind of factory repair.

IMG_9208rep.jpg
Filename
IMG_9208rep.jpg
File size
251.08 KiB
Views
229 views
File license
Fair use/fair dealing exception

Reply 89 of 91, by Deunan

User metadata
Rank Oldbie
Rank
Oldbie
Predator99 wrote on 2022-02-14, 19:52:

The card uses some very special RAM modules I have never seen before. On the top there is a normal 30-pin SIMM for comparision.
NEC MC-41256A4A-15 8632KD
NEC MC-41256A5A-12 8625KD

Looks like NEC made a few variants of these modules, all based on the uPD41256 chips. Curiously they call them SIMMs even though the the PC memory with pins got called SIPP. The 22-pin one actually has a datasheet: https://datasheet.datasheetarchive.com/origin … DSAP0035671.pdf

Predator99 wrote on 2022-02-14, 19:52:

Unfortunately my card currently refueses to work. The RAM get very hot and nothing is detected. On the back there seems to be some kind of factory repair.

All of the chips get hot? These are NMOS but not that power-hungry, datasheet says each chip is about 30mW in standby. Perhaps the driving logic is faulty and more than one bank is trying to drive the bus? Look into /RAS and /CAS signals. Should not be low all of the time.

Reply 90 of 91, by Predator99

User metadata
Rank Oldbie
Rank
Oldbie
Deunan wrote on 2022-02-14, 21:43:
Predator99 wrote on 2022-02-14, 19:52:

Unfortunately my card currently refueses to work. The RAM get very hot and nothing is detected. On the back there seems to be some kind of factory repair.

All of the chips get hot? These are NMOS but not that power-hungry, datasheet says each chip is about 30mW in standby. Perhaps the driving logic is faulty and more than one bank is trying to drive the bus? Look into /RAS and /CAS signals. Should not be low all of the time.

Yes, the whole memory bank gets hot. I dont know I have the knowledge and energy to dive into that... 😉

But when looking on Pin 19 on the 1st bank (RAS) there seems to be some activity when powering on. On Pin 5 (CAS) there is nothing...

pins.jpg
Filename
pins.jpg
File size
417.61 KiB
Views
200 views
File license
Fair use/fair dealing exception
mem-RAS.jpg
Filename
mem-RAS.jpg
File size
15.41 KiB
Views
200 views
File license
Fair use/fair dealing exception

Reply 91 of 91, by Deunan

User metadata
Rank Oldbie
Rank
Oldbie
Predator99 wrote on 2022-02-15, 20:57:

Yes, the whole memory bank gets hot. I dont know I have the knowledge and energy to dive into that... 😉

But when looking on Pin 19 on the 1st bank (RAS) there seems to be some activity when powering on. On Pin 5 (CAS) there is nothing...

It would be an interesting project I think, but obviously there's only so much free time. Looks like it's all 74 series, except maybe those two chips in the upper corner near the mounting plate. Chips with stickers like that tend to be PALs or small ROMs. However some of these 74 are uncommon, like the '280 parity and '682 comparator. I actually had to look up that '682, narrow 20-pin ones are somewhat rare.
Anyway, there's some nice yellow resistor pack (or at least that's what I assume those are) and the usual delay line, so it should be possible to re-create the schematic. Obviously thats easier said than done though. I would probably start with removing all memory sticks except one complete bank (L+U on bank 0 seems a good candidate). Then check if they still get hot, and if both sides (partity and non-parity) is equally hot. If only one side or stick, it might be data bus contention.

You can also power up the sticks with 5V outside the circuit, I think there's only one VCC and GND pins? Measure the current draw per stick, see if any draw way more or less. For the test tie both /RAS and /CAS pins high, or add switches to see what current draw is in standby and when driving control lines. This would be without refresh so probably less power draw than on the card, but if you get about 30mW * chip count per stick then you know they are not completly shorted internally at least.