• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

GK110... where are you?

That's the million dollar question really... I can't see it topping 1Ghz given its size, but somewhere above 750Mhz should be a pretty safe bet.

Given how long they've had to respin and refine the chip, and given that the 28nm process at TSMC has had a year to mature, my guess would be something in the region of 850Mhz. At this speed it would have around 60% more pixel-crunching power than the GTX680, and the 384-bit bus should pump memory bandwidth up by 50% (no real reason for memory speeds to drop).

Another aspect,is also since the GPU is being optimised for compute purposes too,whether there will be a relative drop in gaming performance compared to gaming optmised cores like the GK104 or its replacement.

Edit!!

We saw this before with the previous GTX400 and GTX500 series and even to some degree with the current AMD cards.
 
Another aspect,is also since the GPU is being optimised for compute purposes too,whether there will be a relative drop in gaming performance compared to gaming optmised cores like the GK104 or its replacement.

Core-for-core performance should be equivalent. The penalty will come in the form of extra heat (from the "compute-related" transistors being active, and of not much use in gaming), and potentially reduced clockspeeds (larger chips are inherently less stable). So the penalty will be more indirect, in terms of restricted voltage and clockspeed, which relates back to your previous point :)

Perhaps of more concern is the "wasted" die space due to these extra features... They could further bump up the manufacturing price, not only due to increased die size but due to reduced yield as well (if a fault occurs in an SM block it can be disabled, but if a fault occurs within interconnect logic the GPU is as good as dead).


+1

(well put :p)

Thanks :D

I'll admit I did have a little chuckle when writing that!
 
That wccftech link is the only one that I've seen, and that doesn't read as being particularly concrete.


It's possible that GK110 will remain HPC-only, but if Nvidia can compete with AMD using the "sensible" GK114 chip (which is likely to be a refined and slightly beefed-up GK104 at a similar die size and price-point), then what do they have to lose in releasing a GK110 card at an even higher price-point?

They will already be manufacturing the chips for the HPC market, so it makes sense to leverage that into winning the annual GPU willy-waving competition.

True. Good point.

Consider it sold x2 here if it does come out and is the best card on the market!
 
Core-for-core performance should be equivalent. The penalty will come in the form of extra heat (from the "compute-related" transistors being active, and of not much use in gaming), and potentially reduced clockspeeds (larger chips are inherently less stable). So the penalty will be more indirect, in terms of restricted voltage and clockspeed, which relates back to your previous point :)

Perhaps of more concern is the "wasted" die space due to these extra features... They could further bump up the manufacturing price, not only due to increased die size but due to reduced yield as well (if a fault occurs in an SM block it can be disabled, but if a fault occurs within interconnect logic the GPU is as good as dead).

I suspect AMD will hit the same issues with its Tahiti sucessor too. I wonder if the AMD midrange will be bumped up in transistor count by a large amount, to better compete with the GK114 and its derivatives??
 
Last edited:
I suspect AMD will hit the same issues with its Tahiti sucessor too. I wonder if the AMD midrange will be bumped up in transistor count by a large amount, to better compete with the GK114 and its derivatives??

Possibly...

My guess is that both Nvidia and AMD will pull out slightly beefed-up and slightly more efficient versions of their existing lineup. i.e. GK114 to be fairly cut-down with regards compute features (as is GK104), coming in at around 300mm^2 and maybe 10-15% more powerful than GK104. I guess this chip will get the GTX780 name... Sea Islands at around 2300 shaders (as rumoured) to go up against GK114 in the "regular high-end" space, roughly matching GK114 for performance and price (at around £400 or so).

As for mid-range, I expect the Pitcairn replacement to once again be stripped of most of the compute features, giving it a similar "performance per mm^2" advantage over Sea Islands to that Pitcairn shows in comparison to Tahiti. Whether Nvidia's mid-range GPU (GTX760?) will be just a reduced version of GK114, or a separate GPU entirely, I don't know.

On top of this I'm expecting a dual-GPU card from both sides this time, and for Nvidia to release the GK110 as a "super-powered GPU" at a ridiculous price-point (as outlined above). It's always possible that AMD will come out with a monster GPU themselves, but they haven't gone that way for a long time (not since the 2900), and we haven't heard any rumours of big AMD chips cooking, so it seems unlikely.


These are just my educated guesses though, so feel free to throw them in my face in 6 months time :D
 
Last edited:
Possibly...

My guess is that both Nvidia and AMD will pull out slightly beefed-up and slightly more efficient versions of their existing lineup. i.e. GK114 to be fairly cut-down with regards compute features (as is GK104), coming in at around 300mm^2 and maybe 10-15% more powerful than GK104. I guess this chip will get the GTX780 name... Sea Islands at around 2300 shaders (as rumoured) to go up against GK114 in the "regular high-end" space, roughly matching GK114 for performance and price (at around £400 or so).

As for mid-range, I expect the Pitcairn replacement to once again be stripped of most of the compute features, giving it a similar "performance per mm^2" advantage over Sea Islands to that Pitcairn shows in comparison to Tahiti. Whether Nvidia's mid-range GPU (GTX760?) will be just a reduced version of GK114, or a separate GPU entirely, I don't know.

On top of this I'm expecting a dual-GPU card from both sides this time, and for Nvidia to release the GK110 as a "super-powered GPU" at a ridiculous price-point (as outlined above). It's always possible that AMD will come out with a monster GPU themselves, but they haven't gone that way for a long time (not since the 2900), and we haven't heard any rumours of big AMD chips cooking, so it seems unlikely.


These are just my educated guesses though, so feel free to throw them in my face in 6 months time :D

The Compute performance is rumoured to go UP on AMD's next line of GPU's (8### Series)

http://www.tomshardware.co.uk/forum/346081-15-radeon-8000-series-complete-discussion

As you can see from the leaked slide, Tenerife would be an Enthusiast level product replacing the current flagship HD7970 GPU so we can assume that it would be called HD8970. It is said to feature 2304 Stream Processors, 128TMU’s and 48 ROP’s which would result in over 4.50 TFLOPS of processing power and over 20% better compute performance compared to HD7970. Display outputs include an HDMI and 2 x MiniDisplay ports, DVI is not listed however the card is suggested to support a maximum of 6 displays.
 
Last edited:
GK110 was the Tesla/Workstation card, there was no evidence of it ever becoming a GeForce, just rumours and hope.
Does look as though GTX780 will be something different or a cut down version of it.

The Compute performance is rumoured to go UP on AMD's next line of GPU's (8### Series)

http://www.tomshardware.co.uk/forum/346081-15-radeon-8000-series-complete-discussion

"Display outputs include an HDMI and 2 x MiniDisplay ports, DVI is not listed however the card is suggested to support a maximum of 6 displays."

LAME.
 
They're withholding it until they can release it alongside a decent, working set of WHQL drivers.

So expect it 2016. :p
 
The Compute performance is rumoured to go UP on AMD's next line of GPU's (8### Series)

http://www.tomshardware.co.uk/forum/346081-15-radeon-8000-series-complete-discussion

So 20% faster than a 7970. OK that makes it about 10% faster than a 7970 GHz edition (seeing as the GHz edition didn't arrive until June 22 and that slide is dated march) not exactly a huge increase.

Nvidia will do a similar thing with similar performance increase and we will be right back where we are now, the top two cards very close, vying for our cash.
 
So 20% faster than a 7970. OK that makes it about 10% faster than a 7970 GHz edition (seeing as the GHz edition didn't arrive until June 22 and that slide is dated march) not exactly a huge increase.

Nvidia will do a similar thing with similar performance increase and we will be right back where we are now, the top two cards very close, vying for our cash.

let's wait to see what the actual performance is before we start making predictions.

Don't forget that AMD have consistently upped the performance of their GPU's while using the same number,- or less cores than the previous generation, and half the time on the same architecture.

The rumour is the next card will actually be the 8870, which is said to be a contender to the GTX 680 http://videocardz.com/34981/amd-radeon-hd-8870-and-hd-8850-specifiation-leaked

It's due out early next year.

It will be interesting to see it, that's for sure.
 
NVidia lucked-in with GK104. Small die and narrow memory memory bus compared to fromer top end cards and AMD's 7900 series. Even NVidia proudly confessed to being surprised at how well the GPU and VRAM interface scaled. For it's size this card should never have been a GTX580 successor, but everything fell nicely into place. AMD didn't help themselves by underclocking the much more expensive to produce 7970.

I believe the success of GK104 has held back a fast release of the GK110 desktop version. Until AMD counter with a faster card, NVidia would only cost sales for GTX670/680's, which are very cost efficient and profitable parts.
 
That slide is based purely on the number of added cores, nothing else.

The other link i added (for the 8870) has far more detail. http://forums.overclockers.co.uk/showpost.php?p=22890116&postcount=34

But i would still stress, they are all just rumours.

Ah yes that slide, the one that predicts the new 8870 giving (7970 power?) for £160 (20% cheaper than currant £200 price)to be honest this is not going to happen, i'd love to see it but it wont happen.

pretty much what i said two weeks ago when that slide first surfaced.

edit:


Personally its not the more transistors, the larger die, faster core and memory or even the using less power that I don't believe, but all that for 20-25% cheaper yeah right.

Somebody has been pulling figures out of their behind, but at least unlike some previous rumours for past generations most of those figures actually sound plausible.

160 GFlops to 246 GFlops isn't 60%, 60% would be 256 GFlops but that could just be a typo.
 
Ah yes that slide, the one that predicts the new 8870 giving (7970 power?) for £160 (20% cheaper than currant £200 price)to be honest this is not going to happen, i'd love to see it but it wont happen.

pretty much what i said two weeks ago when that slide first surfaced.

edit:

Oh yeah, It would be extremely inconvenient to Nvidia if can match the 7970, i agree.

In anycase it's not £160, it's $250, the price of current 7870's (£210)
 
Oh yeah, It would be extremely inconvenient to Nvidia if can match the 7970, i agree.

In anycase it's not £160, it's $250, the price of current 7870's (£210)

Why are you saying Nvidia? He was saying '7970 power'.

His argument is coherent in my eyes.
 
Back
Top Bottom