This may seem like a dumb question, but why is everyone so hung up on the number of pipes? Isn't bottom-line the performance of the card and not how it achieves it? :|
I am sure Dave B will disagree with me on this one...
- Thread starter JoshMST
- Start date
Bouncing Zabaglione Bros.
Legend
digitalwanderer said:
I don't want to spend a huge wad of cash on a card that has been deliberately crippled in order that a company can squeeze more money out of me six months down the line. That's just artificially milking your consumers. If I spend money on top-of-the-line, that's what I expect to get, not paying out top-whack for a fully 32 pipe card where I can only use half the performance because some competitor company is lagging behind.
Give us your best efforts guys, and we'll come and buy it - don't try and screw us over. As I said earlier, knowing that there will be a 32 pipe card in the refreshes stops me from buying R520. If you gave me that refresh card now, I'd snap it up straight away as I jump to PCIE.
Of course, we are all assuming that ATI has the overhead to do all of this. What if G70 arrives and it's very fast indeed? ATI will not just be able to simply choose to ship a faster card, and will be in the position of trying an NV30-like overclock-with-a-massive-heatsink in desperation, or accepting a second place in the market for the next cycle. Not good when the last round between R420 and NV40 has been so indecisive.
How ATI got back in the game was because with R300 they aimed high, making the best they could, while Nvidia was aiming for adequate while they tried to milk the market. If ATI are not careful and try to get away with the least they can, then R520 could end up being in the same position of just not being good enough to consolidate on the ATI turnaround seen with R300.
if they aimed for 16 the high end chips would have to run at 600-650mhz or even higher.karlotta said:
thats not what you aim for when moving to a new proccess and redesigning your cards, chips that run at very high speeds have yield problems, i.e. the x700`s.
when you move to a new proccess you aim for a small to decent speed increase and more pipelines becasue you get a "free" increase in transistor count - same die size and hopefully same or even higher yeilds on the same size wappers.
Megadrive1988
Veteran
guys, what about the possibility of a 24 pixel pipeline, 12 vertex pipeline R520? that's 36 pipes in total, just not 32 pixel pipes.
and have the balance of vertex and pixel pipes back to where it was with R300 (which was 4 vertex + 8 pixel)
then the Rv520 has 8 vertex pipes and 16 pixel pipes.
and have the balance of vertex and pixel pipes back to where it was with R300 (which was 4 vertex + 8 pixel)
then the Rv520 has 8 vertex pipes and 16 pixel pipes.
Megadrive1988
Veteran
MuFu said:
dang, shot that down, heh.
Joe DeFuria
Legend
I still can't believe people are fixated on "pixel pipes" after all the hints dropped by Dave...
I think this whole "lets wait and see what NV does then we can adjust our performance to be slightly more than theirs" idea is a whole lot of bunk. Yes, NV and ATI do a bit of gamesmanship here and there with things like the GeForce 6800 Ultra Extreme edition, and creating a X800 XT (non-PE) to keep things in check with each other. But the idea that they can sit around, see what the competition has, then go out directly and adjust their fabrication mix to have a significantly faster product than the other is a bit ludicrous.
NV and ATI have targets at which they set things like clockspeed, and they design around those initial targets on the process they plan to be using. Once they have these massive chips taped out they send them to be manufactured, get the first silicon back, and see what test yields and speed bins look like. Revisions are done to improve speed bins and yields, but we know that this process can take months and months. So, in between when NVIDIA announces the G70 and when ATI announces the R520, do you really think that they can do such a revision to improve their overall clockspeed by a significant amount? I don't. I think ATI's endeavors at the moment are to get the R520 up as fast as they can yet still achieve sustainable yields. If that top speed is less than their target, then so be it. Increasing their yields by 10% while having a slower part could very well be a good tradeoff economically for ATI.
So, this "wait and see" and adjust the final clockspeed may be somewhat true to a point, but it is not the reason why ATI is waiting around. The fabrication process is far too complex to apply this simple motive to.
NV and ATI have targets at which they set things like clockspeed, and they design around those initial targets on the process they plan to be using. Once they have these massive chips taped out they send them to be manufactured, get the first silicon back, and see what test yields and speed bins look like. Revisions are done to improve speed bins and yields, but we know that this process can take months and months. So, in between when NVIDIA announces the G70 and when ATI announces the R520, do you really think that they can do such a revision to improve their overall clockspeed by a significant amount? I don't. I think ATI's endeavors at the moment are to get the R520 up as fast as they can yet still achieve sustainable yields. If that top speed is less than their target, then so be it. Increasing their yields by 10% while having a slower part could very well be a good tradeoff economically for ATI.
So, this "wait and see" and adjust the final clockspeed may be somewhat true to a point, but it is not the reason why ATI is waiting around. The fabrication process is far too complex to apply this simple motive to.
As cpu's used to have MHz that people could grasp video cards have MHz and pipes they can grasp.
I blame Ati for having 8x1 in a fast card than nvidia's 4x2 ! People then started equating speed with number of pipes as well as MHz, before that it was really just MHz.
Really now instead of IPC for cpu's we need SPC ( shades per clock ) for gpu's to be able to see how powerful a video card is.
I'm going to love it is r520 is 16 pipes and yet still blindingly fast. There's about 20 forums where I am going to post " told you so " messages. If it's not of course I will gentle melt into the very quiet background
I blame Ati for having 8x1 in a fast card than nvidia's 4x2 ! People then started equating speed with number of pipes as well as MHz, before that it was really just MHz.
Really now instead of IPC for cpu's we need SPC ( shades per clock ) for gpu's to be able to see how powerful a video card is.
I'm going to love it is r520 is 16 pipes and yet still blindingly fast. There's about 20 forums where I am going to post " told you so " messages. If it's not of course I will gentle melt into the very quiet background
dizietsma said:As cpu's used to have MHz that people could grasp video cards have MHz and pipes they can grasp.
I blame Ati for having 8x1 in a fast card than nvidia's 4x2 ! People then started equating speed with number of pipes as well as MHz, before that it was really just MHz.
Really now instead of IPC for cpu's we need SPC ( shades per clock ) for gpu's to be able to see how powerful a video card is.
I'm going to love it is r520 is 16 pipes and yet still blindingly fast. There's about 20 forums where I am going to post " told you so " messages. If it's not of course I will gentle melt into the very quiet background
Did you see the Huddy interview about Xenos? The part I found really interesting was his asertion that current high-end (hard to say if he was including R520 in that or not, as it might already be "current" in his mind) are only about 50-60% efficient.
Joe DeFuria said:
digitalwanderer said:Joe DeFuria said:
On the one hand I agree with both of you --on the other I understand waverers given that nearly the whole rest of the graphics world has been rumor-mongering 24/32 heavily for months and months. Even caboosemoose, who first broke the roadmap with pipeline config on it, that Wavey left-handed gave some credibility to, has been drifting 24/32-wards of late.
I was very happy to see xbit dancing on that issue, as reading between the lines they are clearly leaning Wavey's way on this issue.
Edit: I should add that the reason I mentioned the Huddy thing upstream is that I haven't given up (as my sig testifies) that R520 owes more to R500 than the "it's a R420 tweak" crowd are allowing. In my view, this is part of the problem in untangling the competing theories --the more "R520 is just a minor refinement of R420" you believe, the more you are lead towards the 24/32 path as a matter of inexorable logic, so long as you also believe that ATI will have a part that is at least competitive with a 24-pipe G70. A signficant increase in efficiency (tho not as much as R500) helps close that gap.
There is MuFu's path --and I don't entirely discount it-- I just think that getting a significant increase in efficiency, plus sm3.0 with the kind of performance I believe that ATI has promised-- makes 650+mhz hard to deliver.
But then, who knows --maybe that's why there was a delay; they had trouble getting the clocks high enuf. I tend more towards optimization of some of the "efficiency" pieces.
Ok, I agree with your assessment of what "the graphics scene" thinks and what the rumors say....but there is this nagging little voice in my head screaming, "Remember that the same people/scene were talking about EXTREME PIPELINES the same way at the launch of the R420". :?geo said:
Fooled once, shame on you; fooled twice, shame on me. I think all the pipeline talk is smoke & mirrors and it isn't anywhere near as important/critical to end performance as everyone seems to think.
Similar threads
