Yes. It won't make any difference.

the potential is to help overclock, or you could lower timings by editing the firmware and so on....

That is a pretty neat idea, especially if it helps with overclocking.

A program called "MCLK" supported overclocking Cirrus Logic GD-542x/30/36/4x, S3 80x/86x/96x/Trio/Virge, Trident 9440/96xx, Tseng ET6000 and more: http://www.geocities.ws/liaor2/myutil/mclk.html

This was apparently a thing in the 90's.

The main benefit would be from overclocking, or trying to get the cards working on faster designed systems.

If say the FSB is too high for a graphics card with slow DRAM, the system may not boot as the DRAM won't refresh fast enough. I've seen this with VLB graphics cards... especially going any higher than 33MHz... many just will not boot at 40MHz.

33MHz systems.... 1/33MHz = 30ns clock cycle - if you have 80ns DRAM... you need 3 clock cycles (3 x 30ns = 90ns) until 80ns DRAM is refreshed
33MHz systems.... 1/33MHz = 30ns clock cycle - if you have 60ns DRAM... you need only 2 clock cycles (2 x 30ns = 60ns) until 60ns DRAM is refreshed

40MHz systems.... 1/40MHz = 25ns clock cycle - if you have 80ns DRAM... you need 4 clock cycles (4 x 25ns = 100ns) until 80ns DRAM is refreshed
40MHz systems.... 1/40MHz = 25ns clock cycle - if you have 50ns DRAM... you need only 2 clock cycles (2 x 25ns = 50ns) until 50ns DRAM is refreshed

It's basic algebra and depending on the DRAM latency you have will determine in how fast a system the graphic card will operate... it's the same as DRAM on a motherboard, which is why most folk can't get their system to boot at fastest settings.... they need faster DRAM!

kixs wrote:

Yes. It won't make any difference.

A small decrese... i.e. from 80ns to only 70ns might not make a huge difference.... but a larger change i.e. 80ns to 60ns... or even to 40ns could mean the difference from a machine booting to not booting.

Why would a machine not boot if RAM on the video card is too slow? You would either get display corruption or no picture at all, but since a video card is optional in any PC, it will not prevent booting.

derSammler wrote:

Why would a machine not boot if RAM on the video card is too slow? You would either get display corruption or no picture at all, but since a video card is optional in any PC, it will not prevent booting.

No picture at all = no boot!

You can use a computer with no screen images?

derSammler wrote:

Why would a machine not boot if RAM on the video card is too slow? You would either get display corruption or no picture at all, but since a video card is optional in any PC, it will not prevent booting.

Well, if a VGA adapter reports to BIOS in POST something BIOS doesn't like, you get a "VGA error" and beeps/no screen within seconds of startup.

If output ports of video card are damaged (or chip controlling them is), you will get past POST screen and PC will run, however you won't see any image.
In theory, you can use PC without image but it requires muscle memory of some degree (to make things faster), and knowing how/when to click on things (or use keyboard only with notes from another computer).

386_junkie wrote:

You can use a computer with no screen images?

Of course.

Putas wrote:

386_junkie wrote:

You can use a computer with no screen images?

Of course.

I'm curious... how do you navigate the command prompt with just a black screen?

If you're familiar enough, do it from memory.

gdjacobs wrote:

If you're familiar enough, do it from memory.

Do you do this often?

If a video card does not work (pc does not boot) on 40MHz PC for some reason, it can't be due to too slow video DRAM. VLB was difficult at high speeds anyway, so it might work on another 40MHz board just fine, or the video board or video chip or combination of those just does not work past 33MHz if it can't.

DRAM memory clocks are generated onboard from a crystal with a PLL, not from host bus clocks.
The video card would not even know if you switch its memory from 70ns to 60ns, it can't auto detect it, the video chip uses the card manufacturer settings for memory clock and access time and these settings usually come from BIOS code or onboard straps that are not user settable (not without soldering iron anyway).

To avoid DRAM memory to be the bottleneck, video chips implement buffers/FIFOs on host interface so only when those are full the host has to be signaled to add more wait states.
ISA boards sometimes had wait state selection jumpers.

Jepael wrote:

If a video card does not work (pc does not boot) on 40MHz PC for some reason, it can't be due to too slow video DRAM. VLB was difficult at high speeds anyway, so it might work on another 40MHz board just fine, or the video board or video chip or combination of those just does not work past 33MHz if it can't.

Have you extensively tested this theory with many VLB cards on many motherboards?... does this theory apply to all VLB cards or just the ones you have used? I'm interested to hear about your tests and findings if you are willing to share?

I have many VLB cards, some with wait states jumpers, some without. The reason I know faster DRAM makes a difference is because I have tested the theory and can confirm that with a couple of the VLB cards... they gave an image at 33MHz but not at 40MHz (this is with 70ns DRAM)... once switching out for faster DRAM, the same card/motherboard combo gave an image on the screen at both 33MHz/40MHz. The only change between tests were the DRAM latency... one gave an image @40MHz, the other did not.

386_junkie wrote:

Jepael wrote:

If a video card does not work (pc does not boot) on 40MHz PC for some reason, it can't be due to too slow video DRAM. VLB was difficult at high speeds anyway, so it might work on another 40MHz board just fine, or the video board or video chip or combination of those just does not work past 33MHz if it can't.

Have you extensively tested this theory with many VLB cards on many motherboards?... does this theory apply to all VLB cards or just the ones you have used? I'm interested to hear about your tests and findings if you are willing to share?

No, I have not tested this extensively. I think I've ever only owned one VLB motherboard (40MHz I recall) and some Cirrus CL-GD542X based VLB card. However I've skimmed through datasheets of various ISA/PCI/VLB video chips during recent years, and I recall from those that at least Cirrus 542X and Tseng ET4000w32p were supporting also VLB. There might be others that have VLB support in addition to other buses, but I don't remember.

I don't think any of them had any kind of DRAM speed autodetection, and they always generated their own memory clocks locally, as they also can't know or measure the bus clock.

What I do know is that for example ET4000w32p specs are only up to 33MHz bus clock and some of the CL-GD542X specs were up to 40MHz, so not all cards were guaranteed to operate on 40MHz bus.

386_junkie wrote:

I have many VLB cards, some with wait states jumpers, some without. The reason I know faster DRAM makes a difference is because I have tested the theory and can confirm that with a couple of the VLB cards... they gave an image at 33MHz but not at 40MHz (this is with 70ns DRAM)... once switching out for faster DRAM, the same card/motherboard combo gave an image on the screen at both 33MHz/40MHz. The only change between tests were the DRAM latency... one gave an image @40MHz, the other did not.

Well that is bizarre, I don't have an explanation for this phenomenon.
I'd be interested which cards or video chipsets did this, if you happen to recall?

I mean I totally believe you that changing DRAM chips made it work for some reason on 40MHz bus, so the card was clearly more stable with different memory chips.
I just don't see why would VLB bus clock speed be affecting onboard memory clock and timings so it requires DRAMs with faster access times.

What I do know about DRAMs is that faster DRAMs gulp more current than slower DRAMs, and on the other hand, older DRAMs gulp more current than newer DRAMs, and they can gulp currents with different profile (even if the average current consumption is same, other might consume current with narrow but large peaks while other consumes current with wide but shallow peaks). That could also make a difference, especially when the chipset itself is consuming more current at higher bus clocks.

386_junkie wrote:

gdjacobs wrote:

If you're familiar enough, do it from memory.

Do you do this often?

I do it at work alot.
Ever programmed old switches? Or old servers?
Its called a headless setup. Real common for rack mounted blade servers. Most of the time you ether use Remote Desktop or send it commands over a network or serial cable.

Edit
https://en.m.wikipedia.org/wiki/Headless_computer

On topic.
Unless if the ram can be ran at the stated speeds if anything the card will likely be the same if not slower do to the looser timings.