9970 to be twice as powerful as GTX 780/Titan ???

[oldestgregg]

The size of the new AMD bundle suggests to me that the new cards will be on par with a 780 as if they were that powerful why would they need to bundle as many games with them? Just a theory.

 

[MG42Maniac]

I'm tempted by the new generation but cannot justify spending ~£350 on the new version of the 7950 (9950?) when I can get a PS4 for the same price. Bad timing really as otherwise I'd almost certainly buy one. Other than BF4 are there any other FPS games on the radar in the next 12 months that will require a decent amount of GPU power?

 

[VincentHanna]

GTA V if they actually port it.

 

[VIPER258]

I think I'll wait until 20nm appears, unless next gen titles require significant gpu grunt.

 

[Boomstick777]

Volcanic Islands are AMD’s codename for next-gen GPUs, but the next-next-gen parts may end up with a more colourful codename – Pirate Islands.

Of course, this is just a rumour at this point, so take it with a grain of salt. According to 3Dcenter.org, the new names are indeed legit and they include three GPUs, Bermuna, Fiji and Treasure Island. All three appear to be 20nm parts, which means they’re not coming anytime soon.

As for Volcanic Islands, the codenames are New Zealand (again), Hawaii, Maui, Iceland and Tonga. They should appear in late 2013, but details are still sketchy. Let's just hope they're more Captain Sparrow than Capitan Morgan.

TSMC’s 20nm node won’t be ready for mass production as soon as GPU makers would like, hence it should take at least six months before we see the first 20nm Radeons and Geforces.

Looks like a refresh this year, followed by pirate Island GPU's next year when TSMC can provide the 20nm node.

 

[Jono8]

Looks like a refresh this year, followed by pirate Island GPU's next year when TSMC can provide the 20nm node.

Makes sense.

Just bought a GTX 670 on the OCUK recent offer as a stop gap between my 480 and a 20nm GPU. I will wait till late 2014 when prices on Maxwell/Pirate Island (or whatever they are calling it) before upgrading again.

 

[dmferrett]

I'm tempted by the new generation but cannot justify spending ~£350 on the new version of the 7950 (9950?) when I can get a PS4 for the same price. Bad timing really as otherwise I'd almost certainly buy one. Other than BF4 are there any other FPS games on the radar in the next 12 months that will require a decent amount of GPU power?

BF4 is the only FPS that counts IMO...

We also have Rome 2 next Month....Which is gonna be exceptional.....Fall pf the Samurai was pretty amazing....A great contrast to BF3 when you need to use a bit more brain over brawn...

And later in the year we will get Command and Conquer Generals 2 on the Frostbite 2 engine which looks Awesome....

Other than that............Titan Falls which looks like a glorified console port...

Lets hope AMD bring out a monster card and pray for value..............

 

[Orangey]

Pirate Islands is 2015.

Volcanic could be anywhere from October to Q1 2014, AMD won't release the next Islands for at least a year after that. They do love their long gens so there is no hope of Pirate in 2014.

 

[bru]

For some strange reason I can see this being a disappointment of epic proportions, imagine a 7970 with boosted clocks, maybe 1250Mhz, a few other tweaks to boost levels and power metering etc. But essentially the same as we have now but without the overclocking headroom as they are already tweaked a lot, just like the lower couple of 700 series.

Whatever they are doing it really can't come soon enough, 20 months (and counting) since the 7970 first arrived is just too long between generations, even same node refreshes in my opinion.

 

[Duff-Man]

It's the fake bus width that's the problem. The cost of producing a 512bit card would be much higher than a 384bit in the long run.

A 512bit interface is certainly more expensive to produce, and adds complexity to the entire design. But it's still a potential option for an "all out maximum performance" type design. I doubt we'll see it bolted to any of the next-gen cards, but it's still possible.

The real red flag here is the combination of 512bit interface with only 5Ghz RAM speed. If you're going to go to the trouble and expense of fitting such a wide memory bus, you sure as **** don't run the memory at such a low speed. 7Ghz GDDR5 is very much attainable, and a 384-bit interface with 7Ghz RAM would outperform the suggested 512-bit + 5Ghz solution while being much cheaper and simpler to implement.




... Having just railed against wide memory buses, should perhaps backtrack a little here - I have a feeling that Nvidia may investigate a 512-bit bus (or even wider?) for the 20nm Tesla cards. In a huge swathe of scientific computing (CFD, FEA - anything that relies on sparse matrix solutions really), the limiting factor is not processing speed but memory bandwidth. The speedup from GPU-acceleration comes from the better GPU-VRAM interface speed compared to CPU-DRAM (so roughly 3x with today's hardware). To a slightly lesser extent this is also true for GPU-physics.

How much this would filter into the desktop gaming market I don't know. The GPGPU market growing rapidly, and eventually I expect Nvidia to start developing distinct GPUs only for the scientific compute market. Maybe a Tesla-only variant of the top-end Maxwell chip, with a fatter memory bus, will appear. Certainly the extra bus width would be of only minimal use for gaming.