For DOS, it makes no difference, since the ISA bus is the limit there. But when using Windows, it will speed things up when 2d acceleration is used (blitter mainly), which works directly in video memory, without having the ISA bus as a bottleneck.

Any benchmark results to back this up?

The case of the Voodoo is probably more interesting - no ISA, no 2D. Seems like a free for all with the RAM speed on these boards.

It has helped me to boot systems with a VLB graphics card, using the fastest RAM timings where at times prior to the upgrade, the system either did not boot or stalled during POST. Some setups with a VLB graphics card of say 70ns RAM will not boot on say the fastest RAM timings nor anything higher than a 33MHz FSB.

With a higher FSB... 40 or 50MHz, and by changing the DRAM for 40 or 45 ns, managed to get the systems to boot using the fastest RAM timings where they would not before.

If the BIOS or chipset isn't particulary tuned for these faster speeds, does it have any impact?

None.

The case of the Voodoo is probably more interesting - no ISA, no 2D. Seems like a free for all with the RAM speed on these boards.

Voodoo cards are clocked by the driver, better memory will allow some overclocking headroom, if rasterizer and TMU can handle it. 30-40ns memory is an overkill though, because it's nearly impossible to overclock this cards up to ~80mhz.

feipoa wrote:

Any benchmark results to back this up?

Backing up that local access is faster on faster RAM? Seriously? Just find yourself some info on how a blitter worked with early 2d acceleration cards in Windows and you'll understand why faster RAM makes a difference.

For the Voodoo cards, it's the same. The 3dfx chipset works with the local RAM on the card. The faster it is, the higher the performance is. Also, clocking the chipset higher requires faster RAM, too.

30-40ns memory is an overkill though, because it's nearly impossible to overclock this cards up to ~80mhz.

I've never seen a Voodoo card with RAM chips faster than 45ns either. Nor one with 60ns. I'd love to see pictures of those 30ns and 60ns cards.

why faster RAM makes a difference

Only if vendor actually made some impovements to the card BIOS/driver to utilise faster memory installed. Usually that's not the case and faster memory chips were added due to a price and/or availability of components at the time of production.

I've never seen a Voodoo card with RAM chips faster than 45ns either.

VGA Museum has some photos of 30-35ns memory cards.

derSammler wrote:

feipoa wrote:

Any benchmark results to back this up?

Backing up that local access is faster on faster RAM? Seriously? Just find yourself some info on how a blitter worked with early 2d acceleration cards in Windows and you'll understand why faster RAM makes a difference.

Condescending snap backs likes this are not all that helpful to online discussions. It appears that you may have interpreted my question as an attack on your verbal authority, or even character. This was not the intention. I am not doubting your claim, but it would be nice to see some examples. I would be interested to see some tangible benchmark data (non-synthetic) which identifies how much differing speeds of graphic card memory improve performance on an otherwise identical card. For example, on a non-overclocked Voodoo1, what are the benchmarks results in GLQuake with 30 ns RAM versus that with 50 ns?

derSammler wrote:

I've never seen a Voodoo card with RAM chips faster than 45ns either. Nor one with 60ns. I'd love to see pictures of those 30ns and 60ns cards.

I have attached some photos for your convenience. One contains 30 ns RAM, the other 50 ns RAM. I may have looked at too small of an image which contained 50 ns RAM when I quoted the "60" because I am having trouble finding it. A "50" can look like a "60" with low resolution and compressed images. Nonetheless, I see several images of Voodoo's with 30, 35, 40, 45, and 50 ns RAM. That is a fairly large range and I am wondering what benchmark benefit may come from this. Someone mentioned that the faster RAM would be useful for overclocking, which makes sense to me. However, you made the point that the card would be faster regardless.

The Serpent Rider wrote:

30-40ns memory is an overkill though, because it's nearly impossible to overclock this cards up to ~80mhz.

You still are struggling to do basic math!? I thought I had taught you this over in another thread... frequency (MHz) = 1 / wavelength (ns).

Please learn...

80MHz = 1/80MHz = 12.5 nano seconds

40MHz = 1/40MHz = 25 nano seconds

33MHz = 1/33MHz = 30 nano seconds

20MHz = 1/20MHz = 50 nano seconds

How many times does 12.5 go into 40?

or how about 25 into 40?

or 30 into 40?

... better yet, how many 40's, go into 40?

Some chip makers actually marked their EDO ram in mhz and ns, like Tmtech memory for example. So 83mhz = 35ns.

Many years ago i did some benchmarking of S3 cards with some comparations of same chips with different clocks....
See archived version here - http://web.archive.org/web/20071115222718fw_/ … anku=2007030001

There are DOS tests of 3 Vision 968 (86C968-P) cards - Elsa Winner 2000Pro/X, Number 9 Motion FX771, Spea V7-Mercury P64V

and 2 Trio64 (86C764X) cards - Dataexpert - 1MB FPM 54,89MHz 0ns vs Expertcolor 1MB FPM 69,8MHz 0ns

Cards memory chips can be seen here - http://vgamuseum.info/index.php/cards/item/355-s3-vision968

I'm really looking for benchmark data with the same graphics card which has different speed RAM chips.

Is the Number 9 Motion FX771 yours? Does Speedsys report 4 MB?

and 2 Trio64 (86C764X) cards - Dataexpert - 1MB FPM 54,89MHz 0ns vs Expertcolor 1MB FPM 69,8MHz 0ns

You mean these two? Expertcolor Dataexpert
So basically they're using the same 50ns memory chips, but one is clocked higher. Interesting.

I have 35ns Trio64V2/DX from ASUS lying somewhere and now I am interested how high they've actually clocked its memory.

386_junkie wrote:

It has helped me to boot systems with a VLB graphics card, using the fastest RAM timings where at times prior to the upgrade, the system either did not boot or stalled during POST. Some setups with a VLB graphics card of say 70ns RAM will not boot on say the fastest RAM timings nor anything higher than a 33MHz FSB.

That's interesting. I'll have to try that out some time. One other person also claimed that certain Diamond S3 based VLB cards were able to overclock to 50MHz because they had surface mount memory, whereas identical boards with SOJ type memory could not.

TOPBENCH gave significantly higher score to me on a Pentium1 board when I used a Mach64VT with 45ns memory compared to a Mach64VT with 60ns memories.
One would have to reprogram the card to accept the higher memory timings when one replaces chips of some random card, or the faster chips will be ran at slower speeds if the design is async to bus speed which I imagine most stuff is.

Tiido wrote:

TOPBENCH gave significantly higher score to me on a Pentium1 board when I used a Mach64VT with 45ns memory compared to a Mach64VT with 60ns memories.

Same brand and model of graphics card, or different? Did the first card know it has 45 ns RAM? Would you be willing jot down some benchmark numbers?

Tiido wrote:

One would have to reprogram the card to accept the higher memory timings when one replaces chips of some random card, or the faster chips will be ran at slower speeds if the design is async to bus speed which I imagine most stuff is.

This is what I was originally thinking as well.

Same brand and model of graphics card, or different?

Probably early ATI shenanigans with retail/OEM cards being differently clocked.

So I've found my ASUS Trio64 and also bought ASUS Virge DX:

ASUS S3 Trio64V2/DX with 35ns is clocked at 60mhz core/memory by default. Overclocking pushed it to 75mhz.
ASUS S3 Virge DX with 35ns is clocked at 66mhz core/memory by default. Overclocking pushed it to ludicrous 84mhz.

Both cards are clearly not using fast memory to their benefit. S3 Trio64 core can't even reach nominal speed of 35ns memory.

Thanks for this info, guys!

386_junkie wrote:

You still are struggling to do basic math!? I thought I had taught you this over in another thread... frequency (MHz) = 1 / wave […]
Show full quote

The Serpent Rider wrote:

30-40ns memory is an overkill though, because it's nearly impossible to overclock this cards up to ~80mhz.

You still are struggling to do basic math!? I thought I had taught you this over in another thread... frequency (MHz) = 1 / wavelength (ns).

Please learn...

80MHz = 1/80MHz = 12.5 nano seconds

40MHz = 1/40MHz = 25 nano seconds

33MHz = 1/33MHz = 30 nano seconds

20MHz = 1/20MHz = 50 nano seconds

How many times does 12.5 go into 40?

or how about 25 into 40?

or 30 into 40?

... better yet, how many 40's, go into 40?

Agreed but really cant get the arythmethics idea...
100Mhz => 10ns, 133MHz => 7ns ok, no problem
83MHz => 35ns, 33MHz => 70ns these values do not connect to previous line.
May be making conversion I have to take into account memory organisation (SDRAM/EDO/FPM)? Or it depends on some another things?
Just wanna put the things in order