NVIDIA GF100 & Friends speculation

TDP for the 512CC part is 295W.

 

Naah, spfp rules in that domain.

 

TDP is actually one thing I couldn't care less about.
Inbetween my i7(at 3.5Ghz), 12GB RAM, 2 x 21" CRT's, 7.1 soundsetup ect....I couldn't care less about TDP, "performance/watt" ect...it's high end....like talking about miles per gallon in F1 racing: meh!

 

What is the upper limit of UAVs in terms of size? Textures in d3d are 16k x 16k, so my guess is somewhere in the region of 28 bits for addressing.

 

SPFP alu's are *assuredly* a big fraction of SM area in G80 and GT200. Fermi adds equal no. of ALU's, each of which is ~40% larger (this neglects the area overhead of rounding modes in spfp alu's).

 

Well it depends on what kind of performance you can get from those 295W no?

For me, its all good, as I dont buy neither ATI or nVIDIA xD. I enjoy the race though

 

However, if 295w results in marginally better performance than a 5870 then it's still a problem in general. Those extra billion transistors need to count for something.

 

Also, even if the address needs 32b mul for address generation, it sure as hell doesn't need it @full speed. Synthesized 32b mul, as in g80/gt200 or a 1/5th ratio as in evergreen seems quite fine considering it's frequency of occurrence .

 

Well...I really don't care.
The bottomline performance is where I look.

Like I said...this is F1 racing class...not Toyota Prius class.
Just turning on my amplifier for my speakers cost me 50-60 Watt...but I don't care.

If it was a laptop I would care..but high end gaming....don't care about watt, TDP...or CO2

 

If it was even 290W, they would have said <295W.
If it's < 300W, you can be sure that 295W<=GTX480 TDP<=300W.

Well, as long as you get enough performance, you could be right. But, as the supposed GTX480 MSRP price may suggest, there are some chances that you could get a good deal of additional performance @ the same TDP by buying an HD5970, which is something quite peculiar, given that we are talking about a dual versus a single card. I actually think that this could be the first time in which a dual has a better perf/watt ratio than a single card of the same generation...

 

NVidia could have implemented 1/4 rate DP and foresaken full-speed MUL32 (lower order bits only) in the primary ALU, yes. I won't disagree.

The choice made by NVidia appears to be much like that made by Intel for Larrabee as the apparent throughput for SP MUL, DP MUL and MUL32 seems to be equivalent.

Jawed

 

Well obviously the 480 isn't going to cost $699 as some funky pre-orders wanted to show. I have the very same doubt for TDP values.

There's an official picture for the 480 floating around the net indicating a minimum PSU of 600W/42A for it. Now considering that NV itself recommends for its GTX295 a 680W/46A PSU (minimum) with a 289W TDP, there's something not adding up with the 480 having a hypothetical 295W TDP.

Can I set the TDP for the 470 at 220W and take it for granted? Now I'd love to read a reasonable point that suggests what the 480 exactly has that justifies a 75W difference in TDP.

 

Only thing that would make me wonder is if TPD goes up more than the performance.
My GTX285 1GB has a TDP of +200 watt...if GTX408 has a TDP of >400 Watt but ~2x the performance it is really: meh!

Besides dual GPU's seldom scales 100% so I always aim for the fastest single GPU when purchasing.

 

Good question, I don't know what the maximum size of a UAV is.

16384x16384 is the maximum count of elements for a texture, but if the elements are 128-bit (i.e. 4x 32-bit), then the surface is 4GB

http://forums.amd.com/devforum/messageview.cfm?catid=390&threadid=128449

Don't know of any other, useful, information.

Agreed, though, that 28-bits is a potentially reasonable upper bound. But 4GB+ graphics cards are coming...

Jawed

 

Except in this case, Perf/W limits maximum perf....

 

Quarter rate DP seems fine to me. Overall, the VLIW approach makes a *lot* of sense. It makes dpfp in evergreen about as cheap as it can be, while still having higher perf/mm than nv. Rounding modes/denormals might change this though.

If nv had no int pipeline, their dp throughput would be 1/8 sp from just the fp pipes. . No wonder they had to add the int path just to reach respectable dp performance. May be Fermi3 will make all the 4 pipes identical (both spfp and int32) and quad-issue warps every clock. Otherwise, the INT32 pipe seems rather unnecessary. And

 

The problem I see with the GTX 480, is that it could end up at the same price and the same TDP as two 5850s. Is Nvidia so confident that their GTX 480 can match two 5850s? If so then kudos to them!

 

Like I said:

So if GXT480 performs ~2x my GTX285, but has a TDP <4-500 Watt...I really, really don't care

 

http://vr-zone.com/articles/nvidia-geforce-gtx-480-final-specs--pricing-revealed/8635.html

 

/YAWN

Since when those are THE final specs? Where are the clocks? Whats new there really? Just the 512 or 480 debate. The rest was pretty much known. So much fanfarre for nothing, IMO.