More R350 Details
- Thread starter lwells
- Start date
Very promising news 
i'll take it with a pinch of salt
0.15, 2 TMUs, and 400 Mhz would produce lots of heat i would expect.
I heard somewhere that doubling the TMUs would double the nr of transistors. That sound a bit far of, but how many transistors would you expect a chip like this to have?
i'll take it with a pinch of salt0.15, 2 TMUs, and 400 Mhz would produce lots of heat i would expect.
I heard somewhere that doubling the TMUs would double the nr of transistors. That sound a bit far of, but how many transistors would you expect a chip like this to have?
CeiserSöze
Newcomer
I doubt that R350 will be a 8x2 architecture. It's too costly and not needed to beat the GFFX (which is probably the sole purpose of this product ).
).CeiserSöze said:I doubt that R350 will be a 8x2 architecture. It's too costly and not needed to beat the GFFX (which is probably the sole purpose of this product
they already have the drop on the nvidia when it comes to AA performance, perhaps they want to kill them off once and for all by upping aniso performance with 2 TMUS =)
but yeah, I too doubt the 2 TMUs statement very much
jb said:
Well, it could turn out sorta like the 9500Pro/9700 comparison where multitexturing filrate is almost the same, but single texturing on the former is half the latter. Perhaps a 8x2 configuration would result in the same single-texturing performance, but much better multitexturing. . .
Still, one has to wonder how big of a jump the R350 is supposed to be. DDRII is likely, as is a higher clock speed, but the 2 TMUs. . . Perhaps it would work well with the increased bandwidth provided by the potentially higher clock speeds coming out of DDRII, but would that really be in the plans for a "refresh"?
Just why does everyone think that DDrII is going to be used? It's expensive, has high latiency and runs very hot. DDr I is available, has lower latency, costs about 50%less and runs cool...... just why would ATI use it? Nowhere has ATI said they were going to use DDrII, in fact, they have gone out of their way to say they would use GDDrIII whe it was available, that there was little to no speed gain on DDrII..... You guys need to stop taking the crap most of these online rags spew as the truth....
EDIT: wish I could spell......
EDIT: wish I could spell......
martrox said:
well, I still dont know how much of an effect the higher latency would have if the internal caches and timings were set accordingly. this "higher latency" thing is always thrown around by people like you there, but noones really shown me any reason to believe it really has that much of an impact on a VPU. These are not CPU and system memory interactions we're talking about.
And what are the thermal specs for 800mhz DDR vs 800mhz DDRII....do you know them?
I doubt it'll use DDR-II as well.
Of course it supports it (as does the R300) but we'll only likely see DDR-II adopted by select AIBs keen to market an "ultra-high end" card. With the right cooling the possibility of a 450/450MHz R350 from somebody looks likely.
MuFu.
Of course it supports it (as does the R300) but we'll only likely see DDR-II adopted by select AIBs keen to market an "ultra-high end" card. With the right cooling the possibility of a 450/450MHz R350 from somebody looks likely.
MuFu.
I find it abit funny that now many nvidiots are rising to barricades because R350 will not use DDR2. In the same time they are forgetting the most important fact; the memory technology itself doesn't really matter, but the result, that is GB/s, datapath width and latencies do. With 375 MHz DDR R350 will have 50% more raw bandwith than FX Ultra, is that really so unimportant? No, that is not.
What comes to TMUs. Well i don't think that they would take a lot of extra space. Remember that Parhelia has also 16 TMU:s, but only 4 pixel pipelines though. But still, it has "only" 80 million transistors.
What comes to TMUs. Well i don't think that they would take a lot of extra space. Remember that Parhelia has also 16 TMU:s, but only 4 pixel pipelines though. But still, it has "only" 80 million transistors.
Mulciber said:
You might not believe it but it's true...the advantage of DDRII is not latencies--that's a disadvantage--it's clockspeed (which is precisely why the latencies are higher.) At some threshold the advantage in clockspeed overcomes the latency deficit, but I don't know what it is right now, as work on both DDRI and DDRII is progressing to make them better, so things are in flux. As you no doubt have been reading lately mass production of DDRII for the greater computer market has been pushed back to 2004/5. Apparently, some advances in DDRI have been made in which clocks as high as 500MHz have been reached (1GHz DDR.) And the DDRI is outperforming current DDRII at that speed handily because of the latency differences. Basically, they are shooting for lower latencies in both DDRI and II at the moment, but especially DDRII (hence the need for GDDR3 for gpus.) Right now Intel is pushing JEDEC to ratify a DDR400 design as they'd like to use it in a quad-pumped 800MHz fsb to offer when AMD is aggressively pushing Athlon 64 and HyperTransport later this year to the mass market.
IMO, ATI was light-years ahead of nVidia in designing in an economical 256-bit bus which provides the company with maximum flexibility when choosing ram for its products, a luxury nVidia does not have due to its short-sightedness in this regard. Basically, a 9700P already has as much bandwidth as a GF FX 5800 Ultra, even if you add in the dubious and unspecified "gpu caching advantages" nVidia claims add ~4 gigs/sec to its ~16 gigs/sec physical bandwidth. And without the need for DDRII. As it is obvious you don't know the thermal specs, either, why ask the question? *chuckle*
WaltC said:
He's not asking to be told "it's true", he's asking for support for that statement. VPUs tend to access data in typically much more sequential "stream" oriented ways...the type of situation that a P4, for example, is designed to exploit with a higher latency and higher speed RAM type. This means your statement could do with more clarity than "You might not believe it but it's true". Specifically, it also means that your use of a discussion of System RAM does not help clarify your comments about the possible "tradeoff point" for a VPU (which is the actual question asked).
[OT criticism]
While I'd tend to make similar observations, I'd maybe consider skipping a reply altogether if I felt I couldn't provide the information he wanted or a pertinent analysis, and in any case I'd try to do those things ahead of being condescending.
For example:
What, instead he should state it as fact if he doesn't know? What are questions for, then?
Maybe I just misunderstood, but if I replied to all your statements in that post that said "I don't know" but went ahead and made statements with such a comment, what kind of tone would that convey to you? My $0.02.[/OT criticism]
Any case, my own, possibly flawed (please feel free to correct with any pertinent info at your disposal) take:
AFAIK the specification of "GDDR-II" that nVidia is using on the nv30 is "overvolted" compared to the volt spec of "System DDR-II". IIRC, it has also been stated that it doesn't suffer from the latency concern of DDR-II. In fact, as far as I understand, it is a bastardization of DDR "I" with some signaling changes grafted on, based on DDR-II design ideas, maybe as simple as using DDR-II modules in a DDR "I" way, if such is possible. I'm not sure at all it warrants being considered as "DDR-II" for the purposes of this discussion. If indeed my understanding is correct, that would make some of this speculation sort of like chasing a red herring.
Check out my reply to Mulciber on these very things here:
http://www.beyond3d.com/forum/viewtopic.php?t=4169&start=20
Top of page 2. He has been busting my chops on this. He has yet to respond.......
http://www.beyond3d.com/forum/viewtopic.php?t=4169&start=20
Top of page 2. He has been busting my chops on this. He has yet to respond.......
martrox said:
yes, its a conspiracy
actually I work until 7am, so dont expect responses from me until evenings regardless, you still haven't properly answered any of my questions.
a quote from CMKRNL doesn't help much. obviously the only reports of price comparisons we're going to get are from insiders, but until I get substantiating info from another insider, it can't hold much weight. Also, I dont know the context or time of that quote. I link to the thread would have been helpful, he could have said that 6 months ago, in which case it could be a moot point by now.
the link to lost circuits just doesn't prove anything. As demalion was saying, we already know the affects of higher latency on CPU / system memory interactions, and also, in the PIVs case, how these can be mitigated. My question is, do you know its effects on VPU/ram interactions before stating, as if its a fact, that it will have a dramatic negative affect, such that would offset the advantages of higher clock frequency. and also if you know that samsungs DDRII specifications for desktop use are the same as their specifications for graphic card memory. I dont know the answer to any of these, but instead of asserting that I do, I'm asking. Hopefully someone will come to my rescue
The heat dissipation at equal clock frequencies has yet to be addressed. But according to WaltC...since I dont know myself, I shouldn't be asking, apparently... *chuckle*
