Friday, May 22nd 2015
NVIDIA GeForce GTX 980 Ti Core Configuration Revealed
It looks like NVIDIA's GTX 980 Ti launch, which is imminent, won't be a repeat of the GTX 780 Ti, in that it won't be faster than the TITAN product at the time of launch. According to Korean tech publication HWBattle, the GTX 980 Ti will feature fewer CUDA cores than the GeForce GTX TITAN-X, at 2,816. NVIDIA gets that count by disabling 2 of the 24 SMM (streaming multiprocessor Maxwell) units on the GM200 silicon. The texture memory unit (TMU) count will be proportionately lower, at 176 (compared to 192 on the GTX TITAN-X). The ASIC bears the model number GM200-310, according to older reports.
We can't take a call on the ROP count and L3 cache amount. Normally we would deduce that it has a full complement of 96 ROPs, but given that Maxwell allows SKU designers to disable components in a way they previously couldn't, it's possible that the GTX 980 Ti could have a different ROP count than the GTX TITAN-X, just as the GTX 970 has a lower "effective" ROP count at 56, compared to the GTX 980, despite the same memory bus width. We know from other reports, that the GTX 980 Ti will feature 6 GB of memory. The TDP is a very arbitrary number, and 250W shouldn't surprise us. What also wouldn't surprise us is NVIDIA reusing the PCB and NVTTM (NVIDIA Time-to-Market) cooler design from the GTX TITAN-X (and several older SKUs). NVIDIA could allow its AIC (add-in- card) partners to come up with custom board designs from day-one.
Just a quick recap.
We can't take a call on the ROP count and L3 cache amount. Normally we would deduce that it has a full complement of 96 ROPs, but given that Maxwell allows SKU designers to disable components in a way they previously couldn't, it's possible that the GTX 980 Ti could have a different ROP count than the GTX TITAN-X, just as the GTX 970 has a lower "effective" ROP count at 56, compared to the GTX 980, despite the same memory bus width. We know from other reports, that the GTX 980 Ti will feature 6 GB of memory. The TDP is a very arbitrary number, and 250W shouldn't surprise us. What also wouldn't surprise us is NVIDIA reusing the PCB and NVTTM (NVIDIA Time-to-Market) cooler design from the GTX TITAN-X (and several older SKUs). NVIDIA could allow its AIC (add-in- card) partners to come up with custom board designs from day-one.
- SKU name: GeForce GTX 980 Ti
- ASIC: GM200-310, 28 nm
- 2,816 CUDA cores
- 176 TMUs
- Unknown ROP count
- 384-bit GDDR5 memory interface
- 6 GB standard memory amount
47 Comments on NVIDIA GeForce GTX 980 Ti Core Configuration Revealed
1) the Fiji XT is not as good as Titan X, or
2) the Fiji XT is better than Titan X and Nvidia want to get it to market before AMD steals the sales.
Oh dear, flip a coin - what is it?
EDIT: either way, looks like Titan or Fiji XT for me
It's definitely gonna be an exciting couple of weeks in Videocard Land :D
I take all the info from now on as a fact. Thanks
With the 780 Ti, it was the same GPU as the Titan Black, but now they're giving us less, so the only way to get the top GPU is to spend £900. No thanks NVIDIA, I'm not that big a sucker.
Red team says this is a ploy by Nvidea to sell off its cards before the 3xx series comes and kicks its teeth in within a month.
How about some real world numbers from either side? Until then a fanboy will draw whatever conclusion supports their particular delusion. I just want to know which card will be better for the money.
Releasing a cut down card this soon may be a move to clear out the inventory of defective GM200 chips before AMD releases its new flagship card, either that or Nvidia anticipates Fiji is equal in terms of performance to a cut down GM200 or the high end model will be priced at around $800, making it more of a direct contender to Titan-X and/or a potential fully featured GM200 based card in the same price range.
Meanwhile on Johan Andersson's twitter:
This new island is one seriously impressive and sweet GPU. wow & thanks @AMDRadeon ! They will be put to good use :)
twitter.com/repi
(Fiji xt picture included?)
AMD in to us new generation made a lot of technological advances and the 20 nm so I waited. On GTX980ti is nothing new, yet the new carrier substrate, they could not make even a year to us throw away lies the head, not to mention the price.
However, if nvidia cuts prices on the 980 to $375-450, amd might be in trouble from sales hemorrhaging after such a big product launch (i.e. Fiji).
Knowing that TitanX Average/Peak power numbers where up like 42/32% respectively over a 980, moving the clocks on something like this won't make those numbers improve. The upside if they use Samsung modules instead of what came about from 12Gb of inefficient (aka hot) Hynix, Nvidia could find some the saving there with half and better memory. It probably going to come in slightly better perf/w, but as the TitanX wasn't a great 4K, this won't appear to fair better.
When Fiji XT drops and Titan X looks threatened, Nvidia green lights Titan X-treme. This is very simply GM200-400 with all intact plus a balls to the walls unlimited power limit. Core clocks at 1400Mhz, Boost 1500+.
It'll cost £1000 and be made only to say - we're still the best. 10 will be made.
At which point, AMD release Bermuda (390X2) and because the memory chips are on die, it'll be small and easy to manufacture (relatively speaking pcb wise). The Bermuda dual card will be called Cronos, as this was the boss of the Titans.
I rest my case and I'm off to the patent office.
www.hardocp.com/article/2015/04/14/nvidia_geforce_gtx_titan_x_video_card_review/13#.VV9iWJdViko