No offense intended, but as far as I can tell, Wavey's post made perfect senses while your reply made none. Gross margins are equal to (Revenue)/(Cost of Revenue). Unless you know what the costs are, the price is meaningless.
NVIDIA GT200 Rumours & Speculation Thread
- Thread starter Arun
- Start date
- Status
According to nHancer, you are correct.
For some strange reason, their online documention says supersampling cannot be turned on when using an 8x00 series card, but I can get it working on my 8500 GT...
Yeah i fully agree with the cluster count. Maybe i should've clarified but i was basing the number on how the G80/G9x "grouped" its scalar ALUs i.e 8x2 per cluster. So possibly you could be correct. IMO 15 sounds rather alarmingly high and unreasonable tbh.
16x is a combination of 4x RGMS and 12x CSAA; and no coverage sampling isn't "edge AA" (there's fairly plenty of documentation by now available how coverage sampling exactly works). In any case let's not split hair here what is what, truth is that in terms of IQ 16x lies somewhere in between 4xMSAA and 8xMSAA.
As for the hybrid MS/SSAA modes and pure SSAA modes they weren't operational with the first drivers for G80; they enabled them though later on and you can now enable them with 3rd party applications like nHancer.
Yeah my bad. You're right 16xCSAA isn't supposed to be all that hard on the hardware. I was thinking 16xQ. Those are the settings I use as well but I'm only on 1680x1050. Well not CSAA I just use plain jane 4x.
You can still do much better than a 640MB GTS today though. And if you're running a single monitor the GX2 is a compelling option. But there are always gonna be guys like you with the 24 and 30 inch monitors who genuinely need faster hardware. It's just that this time around the bored whiners far outnumber them.
Early on I was expecting 8 clusters at 8x4 or 8x3 per cluster for 256 or 192 SP's respectively. G80 has so much overhead per cluster and per SP-SIMD that scaling at that level has got to be very expensive.
You're talking 30 instruction and operand issue units across a 240 SP chip (assuming 8-wide SIMDs). On a 480 SP RV770 you've only got 6 (granted they're much bigger). Then you have the per cluster TMU's, PDC and L1 as well. G8x pays a high price for its flexibility and throughput.
I know a lot more than Fudzilla (among others) about both GT200 and RV770. That doesn't mean I'm supposed to let you know about it, damnit!
AHHA!!! I KNEW IT!!!!!!
Just tell me one thing......... Should I buy NVDA stock?
Uhhh, I don't have precise die size data and I obviously have no insider info about yields (here's a hint: if anybody claims that they do, they're lieing). And even raw specs don't always tell you everything about real-world performance; plus it's not like I had the RTL or anything, there are plenty of things you can't estimate before actually running synthetics. So I really don't know how I'm supposed to know that? Plus, that'd amount to insider trading even if the info wasn't 100% reliable.
Plus, that'd amount to insider trading even if the info wasn't 100% reliable.Uhhh, I don't have precise die size data and I obviously have no insider info about yields (here's a hint: if anybody claims that they do, they're lieing). And even raw specs don't always tell you everything about real-world performance; plus it's not like I had the RTL or anything, there are plenty of things you can't estimate before actually running synthetics. So I really don't know how I'm supposed to know that?
Can I bribe you?
So what is it called now?
OK I'm gonna go get lostNo? And I said I know more than Fudzilla and others (that's not an incredibly high standard is it?), not that I have every single tidbit of information you'd need to make a perfectly informed investment decision. So stop being ridiculous and get lost!
Those that are really in the know are usually bound to a NDA too, which trash tabloid sites like Fudzilla aren't obviously.
You better tell that Fudo cause he's the one that bounced from G100 to GT200. Anyway he isn't even sure on which manufacturing process it'll appear and he just a couple of days ago sounded somewhat secure that it'll be a single chip GPU. I'd call that rather a large pile of jackshit than anything else.
Oh, I'm pretty damn sure GT200, G100, NV55 and D10U are all perfectly correct codenames. I don't know if I should be saying that either, but I'm so sick and tired of people saying this or that isn't a proper codename that I just thought I had to let that out. And the latter pretty clearly indicates it won't be called 9900 anyway...
- Status
