The Future

[GT98]

Well hopefully we'll be able to get a Matrox SOON Video card in our PCs come the summer time...so This leads me to this...how long do you think we'll have bragging rights when it comes to benchmarks till say, Nvidia or ATI comes out with the next lastest and Greatest Video card? Do they have a chance even to meet the expected performance of the new card during the next refresh come the fall or will they have till wait till summer?

Next question....how long till you think till Nvidia comes out with some paperwork ala Kyro2 discrediting the Matrox card?

Anyway...it should be a pretty interesting Summer/Fall

Scott

[Ali]

If the rumoured specs are anything close to correct (19.2Gb/s bandwidth) and it comes out Soon® (couldnt remember the ASCII TM number), the it should be the fastest for a while.

The ATI R300 shouldnt touch it, so no worry from ATI for a year or so.

The Geforce4Ti series is not worth considering, and their update in 6 months probaly wont be able to make that much of a jump.

The 3DLabs/Creative chip is the only unknown at this point. It quite possibly could equal or beat Matrox, but we dont know enough about it to even start speculating (but since when has that stopped us?).

BitBoys are out (sorry Nappe1).

Sis are interesting, but I wouldnt expect a high end card from them.

Via/S3 (sonicBlue or whatever theya re called this week) might have something, but they never seem to be able to deliver quality hardware, not to mention drivers.

The next chip after the Nv25 is rumoured to NOT be called a geForce, so nV might have something comming to watch out for.

I would expect ATI to be very feature rich, but on raw bandwidth I dont see them getting close to Matrox.

Then again, my personal view is that Matrox will put off production untill they can get .13 micron proccess, then have an update when .10 is available. As far as I know, nV and ATI are still on .15micron (not sure about that).


Sorry for the badly organised post, but just brain dumping all my thoughts on the subject in the order they come out.

Ali

[Tempest]

I would say for about 1..4 months.

3DLabs, nVidia and ATI can all do a big surprise. Even Via (+?) PowerVR, SIS, the BitBoys and some other "black horses" might, but not that soon IMHO.
When it comes to fighting over the souls of people that would get Matrox because of it's special qualities, I won't count any of the three first out. They all have revised or completely new graphics cores nearing completion.

It's not like bandwidth is everything, but I hope Matrox hasn't gotten so carried away by their mighty bandwidth (if that 19 GB/sec is real and not "marketing" bandwidth) that they have forgotten to sport some BW saving features as well. The competitor's cards can very well have ~16 GB/sec of bandwidth - and with BW saving that is more like 21 GB/sec.

ps. I wonder who is the first to reach 32+ bit internal precision? NVidia, 3DLabs or..?

edit: Ali, the ™ is easy to remember... Just place your right hand's thumb on the keypad zero, the index finger on 1, middle finger on 5, and ring finger on 3. Now just press Alt and press the keys with each finger, starting from the thumb

[mdhome]

Originally posted by Ali
If the rumoured specs are anything close to correct (19.2Gb/s bandwidth) and it comes out Soon® (couldnt remember the ASCII TM number), the it should be the fastest for a while.

The ATI R300 shouldnt touch it, so no worry from ATI for a year or so.

The Geforce4Ti series is not worth considering, and their update in 6 months probaly wont be able to make that much of a jump.
Ali

If the rumoured specs for the RV300 are anything to go by then it should have 8 pipelines. Not too shabby. But then it all goes down to bandwidth. They aint much use if you cant feed 'em.

[Hyp-X]

Originally posted by Tempest
ps. I wonder who is the first to reach 32+ bit internal precision? NVidia, 3DLabs or..?

ATI with R200 (Radeon 8500)

[Novdid]

I think you're mistaken on that one, Hyp-X.

[roadie]

The next chip after the Nv25 is rumoured to NOT be called a geForce, so nV might have something comming to watch out for.

i am afriad that we will see the same situation that we did when the g400max came out. it was the fastest for a very short period of time, and then the geforce appeared and stole all the hype. nvidias next card is supposed to be using x3dfx tech isn't it??? they might have something special too

[Novdid]

The nv25 is using some x3dfx tech, but it seems like the next chip (not the refresh) from nv next year will contain alot of x3dfx tech, and that includes gigapixel technology. That means that it may have some sort of HSR tech that can measure with the PowerVR tilers.

[mdhome]

I read somewhere that Tilers efficiency diminishes as the number of polygon increases. Unfortunately I cant vouch for that but seeing as the next generation of games will increase the number of polygons used by several orders of magnitude over current games, that may be important. Does anyone know if this is true.

Regards MD

[CHHAS]

I read about that too MD, but if you look at Anand's look at the new Unreal performance test you'll see the Kyro 2 is keeping up with the GF2 just fine. That test has about 50-100 times the polycount of Unreal Tournament according to Anand.

http://www.anandtech.com/showdoc.html?i=1580&p=7

[Schattenreiter]

-----------------------------------------------------------------------
Tempest wrote:

It's not like bandwidth is everything, but I hope Matrox hasn't gotten so carried away by their mighty bandwidth (if that 19 GB/sec is real and not "marketing" bandwidth) that they have forgotten to sport some BW saving features as well. The competitor's cards can very well have ~16 GB/sec of bandwidth - and with BW saving that is more like 21 GB/sec.
------------------------------------------------------------------------

As long as there is that bottle-neck called Northbridge between the Cards and the rest of the computer (which is still very slow), the high bandwith won´t help them very much.

If the mainboard manufacturers could directly connect the AGP with the CPU (like in all the game-stations like Xbox, PS2, Cube, etc) we should see an enormous amount of acceleration.

But as long as all the data has to go through the northbridge...

[Hyp-X]

Originally posted by Schattenreiter
As long as there is that bottle-neck called Northbridge between the Cards and the rest of the computer (which is still very slow), the high bandwith won´t help them very much.

I wonder how the northbridge is a bottleneck between a GPU and the memory on the video card...

If the mainboard manufacturers could directly connect the AGP with the CPU (like in all the game-stations like Xbox, PS2, Cube, etc) we should see an enormous amount of acceleration.

Since when does any of those consoles have an AGP bus?

But as long as all the data has to go through the northbridge...

I wonder what kind of data you are talking about...
Normally all data needed to render a scene / game level should be uploaded to the card, and then rendered without using the AGP bus at all.
That is if you have sufficient memory on the card, but you can already buy cards that have twice the memory than any of those consoles...

[Wombat]

As long as there is that bottle-neck called Northbridge between the Cards and the rest of the computer (which is still very slow), the high bandwith won´t help them very much.

Yes it will. This bandwidth is all on the video card. From video RAM to output. Doesn't much matter how fast the system bus is. What you're saying is an example of why higher AGP speeds aren't too useful.

If the mainboard manufacturers could directly connect the AGP with the CPU (like in all the game-stations like Xbox, PS2, Cube, etc) we should see an enormous amount of acceleration.

No, actually. The point of AGP (and bus-mastering, and DMA) is to <I>remove</I> the burden from the CPU, and the latency that such a middleman causes. North bridges are good for you, honest. Besides, not all data has to go through them anyway. One PCI device can talk to another once the bus-mastering transaction is approved.

[Nappe1]

Originally posted by Hyp-X


ATI with R200 (Radeon 8500)

yep, you are right. they have per pixel precision scale from -8 to +8

but I think they ment, when chips will start calculate pixels internally with 48 or 64 bits...

well, there is already one chip on the market that does it. It is 3DLabs WildCat III.

and another one with 64 bits Internal floating point precision is coming at least to press tests, but because I don't want to cause more trouble than I have already done, I'll leave company name as a secret...

and no one (except me, MK and maybe Insomnia here) likes them anyways.

[Nappe1]

Originally posted by Tempest
I would say for about 1..4 months.

It's not like bandwidth is everything, but I hope Matrox hasn't gotten so carried away by their mighty bandwidth (if that 19 GB/sec is real and not "marketing" bandwidth) that they have forgotten to sport some BW saving features as well. The competitor's cards can very well have ~16 GB/sec of bandwidth - and with BW saving that is more like 21 GB/sec.

oh well... I could just have wro te this one too to Muropaketti, but because some circumstances it would cause a flamewar, I leave it up to you Tempest.

anyways... Matrox isn't carried way with bandwidth so there will be at least some pretty impressive bandwidth saving tech if I am not mistaken.

too bad it looks like that Matrox will be cost too much for me this summer... oh, well... back to the Mainstream I go. Time shows what it has to offer.