Hmmm, I think this is going to be a lengthy post
Napalm-
"I think that implementing a multichip solution is fantastic as it allows a 3d card maker to pack one core with all the feature flavours of the month and then hit all power segments with cards featuring 1, 2, 4 or more chips. If NVidia were doing it, I'm sure you'd be praising them"
I wouldn't praise nVidia for doing it, it is a cop out, and a waste, on several different levels. Check through my older posts if you like, I nearly always reccomed a single faster CPU then a SMP rig unless faster chips are not available which in this case they are. Hitting power segments, drop the core voltage/chip speed or memory speed and you hit all target segments. What will 3dfx do when the Rampage launches? What will they be using for a low cost option? The V4 is already a waste, several cards available or will be shortly that are both less expensive and faster with more complete feature sets.
"It is no more of a cop out than moving to a lower die so that the core operating frequency (and overal speed) can be increased with less heat."
I disagree. Moving to a lower die size is, the majority of the time, limited by available technology not inability to design competitive components. The counter is of course that designing a multi-chip capable offering is a factor to compete, skyrocketing production costs and the risk of supply shortages are real world factors that are right now causing 3dfx some serious problems.
Bober-
"Come on Ben, your smarter than that. Is it just a cop out to build an 8-way Xeon system?"
Absolutely and in many ways. Which would you rather have, a single CPU IR2 or Cray or an eight way Xeon? The 8-Way Xeon systems shouldn't even classify as a cop out, they are complete hacks. Because of the antiquated GTL+ architecture you have an effective bandwith of 12.5MHZ or 16.6MHZ per chip, that is a hack. What level performance boost does an eight way Xeon offer over a single Xeon? Here is a question, which would you chose, an eight way 700MHZ Xeon rig or a single 4GHZ Xeon(if they were available of course)?
What I would consider a cop out is a four or eight way Cyrix system to run against the Xeon/PIII or Athlon, wouldn't you? For an example of this currently in the works look to consumer PPC offerings, they are counting on SMP to pull them within striking distance of x86 consumer offerings.
Now I do think there are certain exceptions for instance the Obsidian monster VSA-100 offering. They are utilizing the only technology available the only way available to reach their goals. In certain ways you could also look at the 8-way Xeon rig as such, if you want the fastest x86 machine there is then look no further, but if you want a true "SuperComputer" then look to SGI, Compaq(Alpha) or Sun. They also utilize SMP but they do so using the best available technology, and they do spend enormous amounts of time and money building the fastest single unit they can, not adding a few features like SIMD to there last generation and relying on improved scalability to carry them.
If ATi were to ship a Maxx Radeon now, then I would consider it a great offering, using the best available technology to place themselves firmly in the lead, if they use it because they need to to keep up with the competition then I view it quite differently.
Also as of right now no 3D grapihcs card is shipping with multiple rasterizers that doesn't have problems directly because of it, either AGP transfer rates or incompatibilities. Perhaps a finished product that was designed
properly to be run in multi chip configurations would change much of my view on them, but so far you pay a price not just in cost but in other factors as well versus single chip solutions.
OneEng-
"Since ATI seems to be on the cutting edge when it comes to optimising bandwidth usage, it would appear that not only do they have 1 up on nVidia with their multi-chip solution capability, but they additionally have the jump on bandwidth compression (for lack of a better term)."
My hats off to ATi, they have what certainly appears to be the best card on the market. If the pro OpenGL benches come through looking as I hope they will, it would be the card I would buy if I was in the market. I don't care who makes the best card, I just want to own it
nVidia announced at the launch of the GF that it was capable of being in multi chip solutions, the question is what reason would they have for doing it? Until September, when the Radeon should be shipping in volume(not the Maxx, just the regular flavor) nVidia has a lock on the mass market performance title. People on this board will probably be able to find a Radeon before then so assuming that we consider that wide availability nVidia has just now seen the first real threat to its' performance title in over a year, why would they want to launch a multi-cip board?
Bandwith compression is a better move then multi chip and again I applaud ATi for supporting it even if in a very rudimentary way. The next gen part from nV(Matrox possibly?) along with 3dfx's chip scheduled for after Rampage and the Kyro all are also supporting memory saving measures. ATi was second with support(PowerVR has them all bested by years), and they should be admired for it, but there support is very limited compared to what is coming.
"ATI is an enormous company. Their OEM connections are legendary. Now that their eye has turned to the performance retail 3D accelerator market, I can't see nVidia (or 3dfx incidentally) beating them out."
ATi is nearing losing their number one position to nVidia, they aren't the giant they once were. ATi can't overpower nV, though either of them could with 3dfx(gouge prices and watch them sink and then return to business as usual). ATi has been losing OEM deals very quickly, mainly to nVidia, what is there to indicate that this will change? A strong product offering for
this generation means it could be a couple of years before the mass market sees any benefit. RagePro, TNT and Riva128 are still the volume parts for ATi and nV, and nV has a couple years worth of dominating ATi before the product cycles line back up.
In the performance market, nVidia now has full access to all of SGI's IPs and now has the rest of their engineers, ATi is relying on ArtX by comparison, who would you bet on? I hope they both stay neck and neck, but ATi blowing nV away is highly unlikely in the performance segment.
"This SHOULD have a positive impact on 3D accelerator prices!!!! I desperately hope to see a price fued between nVidia and ATI. I can't see 3dfx surviving or participating in this war since its product is signifigantly more expensive to manufacture."
Agreed. To add- I think the biggest blow from the Radeon is to 3dfx, not nVidia. NV and ATi can go head to head in a massive price war and spend enormous resources to win it(hehe, that would be nice), 3dfx would be long buried before it was over.
I also agree on your comments about a low end card with my only reservation being OpenGL application performance. If they come out with solid numbers, besting the GF2, I would have no problem saying they have the best available card in all areas(save 2D which is still Matrox's to lose) for the consumer market. I'm not defending nVidia here, I think multi chip solutions are foolish the overwhelming majority of the time, I happen to agree with them. I'm not a nVidia zealot, I am however a hardware T&L zealot and from the specs the Radeon has the best available unit
Facebook
Twitter