wccftech:AMD Fiji XT Leaked For The Third Time On Zauba – 20nm Looks Promising

[Silverforce11]

HBM is a big deal not just for bandwidth.

1. Better latency. XBOX esram is a fancy cache its not comparable to how dGPUs use their vram.
2. Less power usage for the memory sub-system which is a large chunk of the TDP budget.

 

[dacostafilipe]

HBM is a big deal not just for bandwidth.

1. Better latency. XBOX esram is a fancy cache its not comparable to how dGPUs use their vram.

HBM is far, far from the latency that ESRAM has. And even with ESRAM, the improvements are not what you'd expect. GCN has a lot op tools to "hide" the latency.

2. Less power usage for the memory sub-system which is a large chunk of the TDP budget.

True, but going down to 384 like Tahiti with more L1/L2 cache also reduces TDP. Just look at Maxwell.

PS: Don't get me wrong. I'm not against HBM. It's just that the Gen1 does not look that great at the moment. The 2Ghz Gen2 is an other story.

 

[Enigmoid]

It seems after AMD lost mobile dGPU design wins with HD7000 series, they lost focus. They should make a super efficient low and mid-range chips to cover everything up to GM204 (970/980M) and share this design with the desktop parts.

I think its more AMD simply doesn't care. They never released Bonaire for mobile which would have plugged the gap between the 7970m and the lower end cards.

Tonga = Maxwell in terms of memory efficiency architectural gains & front end for tessellation.

My bet is nowhere close. The 960 will probably release at just under the 285's performance level and a 128 bit bus (though higher clocked).

 

[Mondozei]

They only have 2 advantage, HBM and GloFo/Samsung.

Fair, but those two are not "only". Having just one of them is a big deal. Having two is huge. Not saying it's a done deal, but AMD's 2015 stack has a higher potential delta than Nvidia's. But that's really a must by now. 2016 could, if all goes well, see a continuation of this with HBM 2nd gen and a 14 nm transfer. Although the latter will probably be negated if TSMC gets its house in order this year and NV can go full Pascal with 16 nm in Q1 2016. But TSMC being TSMC, you'd never know.

And AMD really needs two back-to-back wins for both years in a row to regain their momentum.

 

[exar333]

Hopefully AMD releases in early Q2 (I know this is probably unrealistic) but with Witcher 3 arriving in May, that will definitely drive some upgrades. I think Dragon Age: Inqusition drove a decent amount of sales in Q4, and much of that was probably 970s/980s along with 290s.

Releasing too late into the summer will hurt AMD unless they are releasing an absolutely stellar product that has been sampled and reviewed early-on to generate excitement.

On this note, I think AMD needs to come guns-blazing with their 300-series release and they need to get the gaming community excited with what's arriving. Essentially we already know what GM200 will deliver (~50%) so AMD has their work cut-out for them. Hopefully they can demonstrate some awesome performance with HBM and GCN efficiency gains...

 

[RussianSensation]

OK so if 390X is 5% faster than a GM200 but GM200 uses a similar amount of power to a 780Ti, say 268W, and 390X uses 300W, how is this a deal breaker?

Also, after-market 980 uses about 190-200W. You guys honestly believe that GM200 will use less than 250W at peak load?

I don't think enthusiast gamers looking to buy $600+ big die 390X/GM200 even view 30-50W GPU differences. Chances are given NV's premium pricing and 6GB of VRAM vs. AMD's 4GB, there will be other WAY more important factors determining the purchasing decision than 390X's 300W TPD/power usage.

I would say even if 390X was $549 and was tied with a $699 GM200, the loyal NV buyers would still purchase NV. OTOH, if 390X CF proves smoother in practices than SLI, then I am sure 100-150W of power usage difference will be ignored by multi-GPU users.

I am more interested in seeing how cut down/close in performance 2nd best R9 390 will be vs. 2nd best GM200.

Hopefully AMD releases in early Q2 (I know this is probably unrealistic) but with Witcher 3 arriving in May, that will definitely drive some upgrades. I think Dragon Age: Inqusition drove a decent amount of sales in Q4, and much of that was probably 970s/980s along with 290s.

It's looking like there is a strong chance both GM200/390X will miss most of the major game launches in the 1st half of the year:

1. Dying Light - Jan 27, 2015
2. Way of the Samurai - Jan 2015
3. Grim Fandango Remastered - Jan 27, 2015
4. Evolve - Feb 10, 2015
5. Total War: Attila - Feb 17, 2015
5. Project CARS - March 17, 2015
6. BF Hardline - March 17, 2015
7. GTA V - March 24, 2015
8. Pillars of Eternity - March 26, 2015
9. Mortal Kombat X - April 14, 2015 (beginning of Q2)
http://www.ign.com/games/upcoming/pc?sortBy=releaseDate&sortOrder=asc&filter=top

Q1 2015 is stacked for new games, well before TW3 comes out. The cards most benefiting the releases are GTX970/980. Since AMD more or less confirmed nothing until Q2 2015, their cards will miss a lot of these major games to begin with.