PDA

View Full Version : Programmable engine



NocturnDragon
22nd April 2002, 08:06
If i'm not mistaken i remember then even the G400 had a programmable engine in it. It was called WARP.
Looking on this forums i read that probably it is used for triangle setup (even if nobody is sure about that) and it could be used to do almost anything but it lacked horsepower.
I am just speculating now. But what if the new card had a very powerful programmable angine. It would be posible to go out right now with a 8.1 D3D (openGL 1.3) card, and later on switch it to D3D 9 or 9.1 (OpenGL 2) just with a bios change.
What do you guys think?

TdB
22nd April 2002, 08:36
yeah, i renember that too, there was a lot of discussion about, if it would be possible to implement T&L on the g400.

back to topic:
if they plan on keeping the parhelia core for the next 2-3 years, then that aproach would make it very easy for them.
and if they would offer next-gen (and next-next-gen) features with a simple bios-update, then the parhelia would be a really really great card. :D
a bios-update like that would probably please alot of costumers, and about adding features with patches, isnīt that what they are already doing with the effects-addon for the rt2500? just on a higher level?
just saying the idea isnīt new to them. :D :D

Jammrock
22nd April 2002, 08:40
Depends on how programmable the GPU was. The BIOS just controls parts of the I/O, not the actual processing of information. If the GPU was highly programmable it would be possible, but it wouldn't be faster than say a dedicated DX9 card. They would have to emulate the additional DX9 functionality, instead of using hardocded paths. It would still potentially be pretty fast...depending on the implimentation.

Of course, if Matrox left space on the die for "future expansion" they could easily add functionality for future DX9/10 abilities on future cores.

Jammrock

Guru
22nd April 2002, 09:03
Originally posted by TDB
yeah, i renember that too, there was a lot of discussion about, if it would be possible to implement T&L on the g400.


Yes there is and even on the G200 if I remember things correctly!

NocturnDragon
22nd April 2002, 09:09
Yes it was present even on the G200

thop
22nd April 2002, 10:28
programmable engine makes sense to me.
what about swapable GPUs? like you keep the board but change your old GPU for an updated one for some $. wil probably save some money and will ensure people keep buying matrox and not nvidia and ati. sounds like an incredibly good idea to me :)

does it even make sense ?:confused: :p

Michel
22nd April 2002, 10:43
Originally posted by thop
programmable engine makes sense to me.
what about swapable GPUs? like you keep the board but change your old GPU for an updated one for some $. wil probably save some money and will ensure people keep buying matrox and not nvidia and ati. sounds like an incredibly good idea to me :)

does it even make sense ?:confused: :p

If i remember correctly there was some talk about swappable gpu's in the big G800 thread's a long time ago... They mentioned all future gpu's being pin combatible so they could use the same pcb and easily switch gpu's..

then again it doesn't makes sense to me, because you won't gain anything with replacing an gpu if the memory interface stays the same.

Guru
22nd April 2002, 13:07
Originally posted by Michel


If i remember correctly there was some talk about swappable gpu's in the big G800 thread's a long time ago... They mentioned all future gpu's being pin combatible so they could use the same pcb and easily switch gpu's..

then again it doesn't makes sense to me, because you won't gain anything with replacing an gpu if the memory interface stays the same.

So a AMD XP 1500+ is no faster then a AMD XP 2200+ using the same mobo and memory? :rolleyes:

Wombat
22nd April 2002, 13:18
No it isn't, not by much anyway. You're likely to do better by taking an XP1500 off of a standard SDRAM board and putting it on a DDR board.

P.S. The 2200 should be the faster chip there, buddy.

Ali
22nd April 2002, 16:52
think about this from a business point of view.

If Matrox say, buy this DX8.1 card now, and in 6 months we will upgrade it for free to DX9 specs, then MS does something strange to DX9, that means that they cant upgrade, where is Matrox now.

On the other hand, if Matrox sell us a DX8.1 part now, and say its upgradable, at a cost of course, then they could be on to a winner.

By the time the card is regulated to the bottom end (eg G200 at the moment), wouldnt it be nice to know you could give Matrox say $50 for a firmware upgrade that would make it at least able to play the most recent games, even if it was painfully slow.

Have updates/fixes free for 3 years, while the card is under warranty, then charge users for an upgrade. Its a win win situation.

The marketing guys would love it. They could sell a G200 type card (relative speed to newest card) and say its DX8.1 complient, even though it could only do .1fps in current games.

The thing Im worried about at the moment is power ussage though. These GPUs are getting to the stage that they are more complicated and powerful than CPUs that were current only a year or two ago. How can they do that much proccessing by only using the AGP power.

I still think 3dfx had the right idea with the external power supply. Costs more, but removes ALL the power problems.

Ali

az
22nd April 2002, 19:29
I think the problem would be the connection from the chip to the card - a connection without direct soldering works with processors, at bus speeds of up to 133 MHz (and no, P4 isn't more than 100 MHz Quad-Pumped ;)), but I doubt it'd work with the high bus speeds graphics chips use :)

As always, please correct me if I'm wrong :)

AZ

Greebe
22nd April 2002, 23:15
Ali, don't sweat the power issue... it'll be fine, no external power connector needed :)

NocturnDragon
23rd April 2002, 02:06
I completely agree with Ali on this topic!

knirfie
23rd April 2002, 03:42
I dont, and the new GPU's already have more transistors then the current CPU's, at least I think they do. and power is'nt a problem, the new gen is made at ,13 micron, at thus consumes less power.

Wombat
23rd April 2002, 09:39
The latest graphics chips aren't even close to the latest processors. Graphics chips haven't even broken 100 million yet.

I disagree with Ali. Socketing chips makes them slower, and more expensive. Also, can you imagine the CS nightmare when people f*ck up installing their own chips?

Greebe
23rd April 2002, 09:49
Knirfie what gives you that impression? Inside info?

Ali
23rd April 2002, 14:41
I wasnt talking abaout having a socket, a simple EEPROM update to the programable GPU.

I did a little microprossessor work when at school, and used Atmel gear. You just flash your .asm to the eeprom and your processor suddenly acts differently. should be as simple to implement as a BIOS update.

Ali

RedRed
23rd April 2002, 15:35
I remember having this discussion long, long ago.... when we hadnt heard of parhelia.... Socketed chips were struck out because they cant work reliably at the bus speeds required, and cost considerably more to produce & mount, for little benefit.

I remember having the discussion about a programmable gpu before, and I think it was Casey who pointed out that the overhead in making a completely programmable gpu was terrible.... you could not hope to come close to the performance. One thing that had popped out was perhap some sort of 'library' which could load into the gpu which might optomise the GPU to handle some specific task (depending upon what it was called to do at a particular time), T&L one time, Triangle the next.... (a von neuman GPU, if you will...) This was shot down too.... generally poor performance relative to dedicated circuitry, i think....

My tuppence....
RedRed

efty
23rd April 2002, 15:49
Actually depending on the granularity of the operation to be performed having a programable part of the gpu can be helpful.

This is currently a hot topic of research compining fixed function cpus and programable. It has it benefits but I seriously doubt it that it would make its first appearance in the graphics market.

Have a look here

http://www.cs.caltech.edu/research/ic/transit/reconfig.com/rc_for_gp.html

Wombat
23rd April 2002, 16:28
Having some sections as "programmable" may be useful, but that's the kind of thing that can be done at runtime, by drivers, no need for EEPROM.
Having a fully programmable unit would limit you to slower electronics.