NVIDIA GeForce GTX 660 and GTX 650 Launch Together on September 12

[btarunr]

According to a new 3DCenter.org report, NVIDIA is planning a joint launch of two its key mid-range GPUs, the GeForce GTX 660 and GeForce GTX 650, on September 12, 2012. The two GPUs will let NVIDIA capture key sub-$300 and sub-$200 price-points, with performance levels that could destabilize AMD's Radeon HD 7800 and HD 7700 series. Based on the GK106 silicon, the GTX 660 is said to pack 960 CUDA cores and 2 GB of GDDR5 memory across a 192-bit wide memory interface; while the GK107-based GeForce GTX 650 is said to pack 384 CUDA cores and 1 GB of GDDR5 memory across a 128-bit wide memory interface.



Source: 3DCenter.org

[ViperXTR]

384, it lost a lot of those precious CUDA cores and its still called "GTX"?

[blibba]

Originally Posted by ViperXTR

384, it lost a lot of those precious CUDA cores and its still called "GTX"?

I imagine it'll perform a lot like the old 550Ti.

There's a big gap between the GTX650 and GK106 though, especially if they want to produce any GK106 chips that aren't fully enabled, as these would have to be called GTX650Ti.

[btarunr]

Yup, that's a logical next step (GK106 with 768 cores, 128-bit GDDR5).

[Benetanegia]

Originally Posted by btarunr

Yup, that's a logical next step (GK106 with 768 cores, 128-bit GDDR5).

Is it really a 960 SP part? Or is it crippled like the GTX 460. Right now, I see the same motivations for doing as they did back then. Namely:

- Full chip would be too close to the big boys*. 660 Ti for sure, maybe even dangerously close to 670 in lower resolutions considering the huge price difference.
- Less binning and qualification "troubles": almost every functional chip qualifies.
- Posibility of "free" upgrade in the near future, aka 760 Ti.

* Aside from the odd SMX number, what appear to be the final clocks also hint at this IMO. There's no reason that the fastest GK106 SKU comes with lower clocks than GK104. Logic says the smaller chip clocks slightly better and historically this has been the case 99% of times.

[btarunr]

Originally Posted by Benetanegia

Is it really a 960 SP part? Or is it crippled like the GTX 460.

Yes, I think it's the latter. 960/192 = 5. I don't think NV would design a chip with an odd number of SMXs. So it has to be 1152 (192*6).

[Benetanegia]

Originally Posted by btarunr

Yes, I think it's the latter. 960/192 = 5. I don't think NV would design a chip with an odd number of SMXs. So it has to be 1152 (192*6).

That's what I think too, but it does not exclude the posibility of a card such as you described lol. The 650 does not look as it will put a lot of pressure on its price point.

[Mistral]

Awesome, it'll be out before the hd8xxx series.

[alwayssts]

TT (yeah, taboo...apologies) shows 660 is essentially 90% the speed of a 7870 for all intents and purposes (with AA), or half-way between a 7870 and 7850. Pretty much what the specs (equal to 1120 radeon shaders, tdp between 7850/7870) imply. I suppose if you look at it from a somewhat delusional and cherry-picked perspective, it is similar to 7870.

Still think it should be a $220-230 (if 7870 250), then 200 (if 230) card. It probably will be once the launch fervor is over, 7870 gets a cut, and have their prices settle. 7870 is looking better and better as it gets cheaper...clearly Pitcairn is the better-designed budget chip. It seems AMD is staking claim to the 'these chips don't need 2x6-pin but it allows us to make them cheaper and offer better value with overclocking' philosophy the past couple generations. I think that will pay off vs 660 as the overclocking prowess gets over some tangible walls at 1080p 660 won't be able to match, if only barely.

There is an argument to be made for the design though; less rops (which are not needed on 7800 but part of the setup engine structure which is groups of 16) should somewhat help offset the greater shader units in Pitcairn, so it should clock reasonably well with similar power (both will use ~150w overclocked even though 7870 has another power connector). It seems at the end of the day AMD just makes a more compact design to do the same task...probably down to the slower, be it wider, memory controller which may use less power and/or space. It seems very interesting 1280sp (overclocked on 28nm) is essentially right beyond the scope of what 192-bit would be capable of even with fast/overclocked ram. Some engineer at AMD had his thinking cap on when maximizing perf/mm with this chip...now that we have something to compare it to I can see why they were so excited about it.

650 just makes me lol.

About a 650ti...kinda wonder if they will wait until Oland (or whatever the 8700 is called) to launch that. I gotta believe that will be somewhat similar (perhaps 896sp, 16rop, probably 192-bit/5gbps) and clock-for-clock a pretty even match-up to gk106 with an SMX disabled. I would think with those designs (and likely 130w tdp) the AMD part would be less logic and therefore able to be volted higher and max out it's clock potential (overlapping into the territory but not beating 660's potential), while the 650ti would be another ~1075 max chip that basically kills 8750...but we shall see.

[Harlequin_uk]

as i said in the other thread - the GTX 650 is simply a rename of the GT 640 DDR5 OEM thats allready available