whitetiger
Newcomer
In what way was it unbalanced?
In what way was it unbalanced?
I don't remember GF114 coming close to GF110 in any game, so it's texturing power is a waste.
whitetiger
Newcomer
I'd say the reverse actually - the GF104/114 was better balanced than the GF100/110.
- the GF100/110 suffered from lack of TMUs - 4 per SM
- many reviews commented on this at time
The GF104/114 redressed this imbalance by doubling the numbers of TMUs per SM, while increasing the SPs per SM by 50%
So, the GF104/114 was better balanced than the GF100/110 is this respect.
I did read somewhere that the Fermi Architecture had other bottlenecks, such as the bus into the SM block, I think, meaning that the SPs could become starved of data easily
- so I should imagine that Kepler fixes this, and possibly other bottlenecks (hopefully)
whitetiger
Newcomer
560Ti -> 580
Memory Bandwidth: +50%
Also, the 48SP, Dual issue architecture meant that the actual per SM performance of the GF104/114 wasn't quite +50% compared the GF100/110 SM - so flop-wise the 580 would have more than +25% ...
Anyway, this all points the GF114 being better balanced than the GF110 (which was famous for running out of steam at higher res & AAs)
None of this says that the GF114 had memory bandwidth to spare as far as I can see.
I did not compare them or said that gf110 have enough TMUs. I see the balance somewhere between them.
Ah yes, my bad.
Yep, but it's hard to draw any sort of conclusion because the 580 isn't a symmetric upgrade in all areas.
whitetiger
Newcomer
No worries
The 580 is a bit a red-herring anyway
- my basic point was that the GF104/GF114 was perceived as quite well balanced (for a Fermi!), so it's difficult to see a GK104 with +25% memory bandwidth not being severely bandwidth limited in a lot of cases
- since clearly is should be able to blast past the GF110 & Cayman in terms of shader power, but has less bandwidth (GF110) or the same (Cayman)
- so even allowing for, say a better memory controller, with more ROPs etc, it still seems somewhat under-endowed in the memory department.
- so, just on a purely speculative, Sunday afternoon, kind of way, I was wondering why they wouldn't just have added more memory bandwidth...
- even a sneaky 64-bit controller (giving a 320-bit bus) would give +25% more bw, and probably not bust the pin-budgets....
- and +25% over the +25% already from the increased clocks would bring them right up to where they need to be....
The context, forum and thread from which you pulled that quote would be nice. That's just basic internet stalking etiquette
I would love to understand how you drew that conclusion from that statement though. Little tip for future reference - I don't know anything about what goes on behind closed doors at any graphics IHV, AIB or tech website. Anything I "know" is public domain rumor mill fare. Sorry to disappoint but it looks like you're chasing shadows....
Heh, isn't that the same thread you brought up earlier over here? It's really curious why you keep bringing up a discussion at [H] in a thread at B3D. Not to mention my comment you quoted was in reference to the futility of arguing over past generations and has nothing to do with the conversation here.
If there's a problem I can help you with feel free to PM me...on any forum
iMacmatician
Regular
Okay, let's say that GK104 comes with 256-bit 5 GHz GDDR5. How hard would it be to get a subsequent "GK114" to have 256-bit 6 or 7 GHz GDDR5 (if such speeds would be useful)? Would it be better to go 384-bit instead?
Man from Atlantis
Veteran
similar fonts and composition to original chiphell leak, doesnt make it any reliable
http://we.pcinlife.com/thread-1831876-1-1.html
DavidGraham
Veteran
wow , is the launch that close ? I didn't realize we are in the benchmark leak season!
whitetiger
Newcomer
similar fonts and composition to original chiphell leak, doesnt make it any reliable
http://we.pcinlife.com/thread-1831876-1-1.html
That would certainly justify the leaked marketing slide from a few pages back claiming it was the world's fastest DX11 chip!
But is it really Legit?
Man from Atlantis
Veteran
That would certainly justify the leaked marketing slide from a few pages back claiming it was the world's fastest DX11 chip!
But is it really Legit?
well it's first post of the guy who posted and register date is today.. if there is actual benchmarks in hands, release date would be very near but it's still too early.. i wouldnt say this if it happens 3 weeks later.. it may be end up truth but i wouldn't give a dime.. rule of thumb dont believe any benchmarks before card photos leaks
