I think it will depends on how fast your processor speed is.(and how fast the motherboard chipset is). On a 386 DX 25 or 386 SX 25 it doesnt make any differents, lower NS memory will do to.. But for the average machines the best way it mostly go with 70ns if you can

Robin4 wrote:

I think it will depends on how fast your processor speed is.(and how fast the motherboard chipset is). On a 386 DX 25 or 386 SX 25 it doesnt make any differents, lower NS memory will do to.. But for the average machines the best way it mostly go with 70ns if you can

Lets see the processor is a AMD 386 SX running at 33mhz with a IIT 3C87 running at the same speed, and the chipset is an OPTI 82C493/495. They're both the same price so I honestly wouldn't mind having slightly slower memory as long as it works correctly.

How would faster memory cause issues? Just curious what you find with your research.

PhilsComputerLab wrote:

How would faster memory cause issues? Just curious what you find with your research.

It may be some misinformation from "Arasirsul" on arstechnica:

Machines in the 486 era tended to use 80ns SIMMs, which is what you got. Often, memory faster than what the machine expected would cause issues because the DRAM refresh didn't happen soon enough.

https://arstechnica.com/civis/viewtopic.php?f=8&t=1163711

I honestly asked the same question Phil, I've put faster memory into other machines before and it preformed to the maximum speed the bus would allow, I just didn't know if older motherboards had limitations and needed exact NS RAM modules to work correctly.

Yea I found the same.

On a 386, using faster memory, and also cache, can let you use lower timings in the BIOS. But it does also depend a little bit on the chipset.

So with fast RAM and Cache you might be able to use the tightest memory timings that the BIOS lets you select.

I mostly use 60ns memory in my old systems. I've never had an issue. More importantly, when buying 30 pin Simms, try to get 9 chip (parity). I've found older boards don't like 3 chip ram sometimes.

Btw, is your chipset really opti495? That's usually for 386dx or 486s.

PhilsComputerLab wrote:

Yea I found the same.

On a 386, using faster memory, and also cache, can let you use lower timings in the BIOS. But it does also depend a little bit on the chipset.

So with fast RAM and Cache you might be able to use the tightest memory timings that the BIOS lets you select.

I'm actually test running the system with the DRAM wait state both read and write set to 0 and it actually seems to be running slightly faster. BlakeStone Aliens of Gold is running at a slightly higher framerate, I'm actually about to do some benchmarks to see if it indeed made a difference.

Anonymous Coward wrote:

I mostly use 60ns memory in my old systems. I've never had an issue. More importantly, when buying 30 pin Simms, try to get 9 chip (parity). I've found older boards don't like 3 chip ram sometimes.

Btw, is your chipset really opti495? That's usually for 386dx or 486s.

You know I think Dr. Hardware has that wrong somehow, here's the actual chips on the motherboard, looking at it, unless it was stamped incorrectly, the software is reporting the wrong chipset.

Whats funny is I'm running 8 CHIP Non Parity SIMMS in the machine right now with the Parity check turned off in BIOS however, I've been told that 9 CHIP is best for AT PC's and 8 CHIP is best for Macintosh. Yea I've had no luck with 3 chip SIMMS before in any older board I've tried, so 60ns would be my best bet it looks like.

You can use this benchmark pack: http://www.philscomputerlab.com/dos-benchmark-pack.html

I've added a 386 friendly Doom low resolution / details option that works well. It replaced Wolfenstein 3D that I used for a while, but it's easier using Doom because it's shareware and interestingly enough Doom at low resolution / details gives you similar FPS than Wolfenstein 3D 😀

PhilsComputerLab wrote:

You can use this benchmark pack: http://www.philscomputerlab.com/dos-benchmark-pack.html

I've added a 386 friendly Doom low resolution / details option that works well. It replaced Wolfenstein 3D that I used for a while, but it's easier using Doom because it's shareware and interestingly enough Doom at low resolution / details gives you similar FPS than Wolfenstein 3D 😀

I am actually using your benchmarks right now since I found your youtube channel Phil. I'll download the newest one you linked.

Sure enough with Superscape I went from 5.4 to 6.1, I haven't even taken the 50mhz crystal osculator out either, how odd.

Great 😁

Playing around with the ISA clock speed can also give you decent gains, but push it too far and you get issues.

PhilsComputerLab wrote:

Great 😁

Playing around with the ISA clock speed can also give you decent gains, but push it too far and you get issues.

I saw that in your 386 video when you were changing the ATCLK to CLK2/5, the lowest mine will go is CLK2/4, I'm going to do more experiments to see what are the optimum settings in CMOS are.

Currently my RTC is not working correctly its actually 3 seconds slow in counting time (instead of 1,2,3,4... its counting 1,1,1,2,2,2,3,3,3,4,4,4...)I'm guessing its because there's no battery in the system, or the 50mhz Crystal which should yield a 25 mhz clock speed, in Landmark is reporting 33 mhz, hmm maybe it's the new IIT 33 mhz 387 I put in today, glad I got a 66 MHZ crystal on the way, maybe that will fix it.

Thanks to everyone for their input on SIMMs, I think I will go with the 60ns 😀

Great timing (double entendre intended), as I'm likewise looking at upgrading and maxing out the RAM on my new old 486 system I'm building. The motherboard specs call for 70ns, but what I can find in 72-pin, non-parity, double sided, 5V, 32MB per stick for 128MB total on matching sticks is 60ns.

Cheers. 😎

WR3ND wrote:

Great timing (double entendre intended), as I'm likewise looking at upgrading and maxing out the RAM on my new old 486 system I'm building. The motherboard specs call for 70ns, but what I can find in 72-pin, non-parity, double sided, 5V, 32MB per stick for 128MB total on matching sticks is 60ns.

Cheers. 😎

I'd shop around on eBay see if you could get a better deal but, this is where I got my 30 PIN SIMMs, and they have tons more memory modules at decent prices: http://www.memoryx.com/72pinsimm.html

I actually went with that same seller too. http://www.ebay.com/itm/32MB-1pc-72-pin-FPM-n … CoAAOSw4A5YxkEa Or at least they look to be the same...

Prices seem reasonable. Haven't gotten the memory yet though, so I can't vouch for it yet.

I've only ever seen refresh issues on VLB cards, when the card was set to run at a >33MHz bus but then run at 33MHz. The board's refresh circuitry didn't refresh the video memory quickly enough and the screen output quickly became corrupted.

I suspect with system memory there's a bit more room to move, because the advice I've always read is that the memory timing is the maximum speed it can operate at and doesn't affect the lower limit. Presumably they are designed to work at a certain minimum speed of some original ancestor chip for backwards compatibility, and they can just operate faster if needed.

If you work it out (or use a calculator) then:

• 120ns = 8.3 MHz
• 100ns = 10 MHz
• 80ns = 12.5 MHz
• 70ns = 14.28 MHz
• 60ns = 16.6 MHz

Any idea how this links in with the bus speed? Do you make sure the RAM you pick is equal to or faster than the system bus? That can't be right because 286 machines run pretty high bus speeds (the CPU multiplier is often 0.5x) so how is the RAM speed calculated? Surely it's not taken off the ISA bus?

EDIT: Just looked up the datasheets for a few RAM chips. Looks like each model has a minimum speed too, which is roughly almost half the maximum speed. So 80ns RAM will go as slow as 150ns, and 70ns as slow as 130ns. So you probably won't run into problems until you start looking at chips in the 30ns range for 286-486 machines, but you'd have to find a datasheet for the chips used on a given stick of RAM to know for sure.

When dealing with older 386's/486's... the Front Side Bus (FSB) determines how fast your system can run and what cache/DRAM you need... all you do is divide into 1.

If you have an FSB of 33MHz... then 1/33,000,000 = 3.03 x10 to the power of -8 = a system timing of 30ns, = 30ns cache!
If you have an FSB of 40MHz... then 1/40,000,000 = 2.50 x10 to the power of -8 = a system timing of 25ns, = 25ns cache!

At 33MHz, with 70ns DRAM simms... cache will be 2.33 times faster
At 33MHz, with 60ns DRAM simms... cache will be 2.00 times faster!

At 40MHz, with 70ns DRAM simms... cache will be 2.80 times faster
At 40MHz, with 60ns DRAM simms... cache will be 2.40 times faster!

As you decrease latency, you will need faster DRAM i.e. 50ns!... and so on.

zakurowrath wrote:

It may be some misinformation from "Arasirsul" on arstechnica: […]
Show full quote

PhilsComputerLab wrote:

How would faster memory cause issues? Just curious what you find with your research.

It may be some misinformation from "Arasirsul" on arstechnica:

Machines in the 486 era tended to use 80ns SIMMs, which is what you got. Often, memory faster than what the machine expected would cause issues because the DRAM refresh didn't happen soon enough.

https://arstechnica.com/civis/viewtopic.php?f=8&t=1163711

I think the statement is quite wrong. First of all, during that time, I remember the standard was 70ns. (in 1992, even my very cheap 386SX-16 came with 70ns SIMMs, and the specs were saying 70 or 80ns is required). In addition to that, faster memory on a machine that has specs for a lower speed memory does not create any problems in general as far as I know. My 386DX-40 rig at the moment runs with 64MB of 60ns memory very happily and without any problems, just like an AWE32 card that I used on it with 2x4MB 60ns memory.

> I've read in some research having faster SIMMs could cause problems in the system
Not with the 32-pin SIMMs. At a slower speed the fast DRAM behaves as the slow one. With the 72-pin SIMMS there might be a problem, because they have a method of indicating speed to the mainboard. The mainboard might not know all standards.

> Machines in the 486 era tended to use 80ns SIMMs
70ns was the common speed for 32-pin SIMMs before they died out.

> because the DRAM refresh didn't happen soon enough.
The refresh rate is a parameter that is independent from the access time.

> I've only ever seen refresh issues on VLB cards
Video RAM has its own video-clock dependent refresh and doesn't need the system board for that.

h-a-l-9000 wrote:

> I've read in some research having faster SIMMs could cause problems in the system
Not with the 32-pin SIMMs. At a slower speed the fast DRAM behaves as the slow one. With the 72-pin SIMMS there might be a problem, because they have a method of indicating speed to the mainboard. The mainboard might not know all standards.

Faster SIMMs i.e. 40ns and 50ns can be used without problem in slower systems. There is no inhibitions or restrictions that prevent the use of faster DRAM... I really don't understand why people think faster SIMMs would be a problem.

If you have a 40MHz system i.e. 25ns system speed...

80ns DRAM requires = 25ns x 4 cycles = 100ns until it is ready for access

Whereas

70ns DRAM requires = 25ns x 3 cycles = 75ns until it is ready for access

Using faster RAM just means it will be ready for access sooner... where's the catch and what is so tricky about it!?