• Happy holidays, folks! Thanks to each and every one of you for being part of the Tom's Hardware community!

Will The Nvidia GeForce GTX 960 Launch At CES In January?

Status
Not open for further replies.
Honestly, I think 3GB is probably better for a mid range GPU, though there will be premium 4GB variants if it is 2GB, and RAM is expensive. $250 is a bit rich for my blood, I think $200 - 230 would be an ok price until the 950 + ti comes out most likely much later.
 
If GM206 is like GM204 the GTX960 is half the SMM of GTX980 but worse a fourth cut down of GTX970 SMM. At That much of a cut down I wouldn't pay more than $150. It better be at least 192-bit with 3GB if Nvidia expects to get $200. With a 128bit memory its only 8 SMM with 1024 which will suck. This could be the specs of the 950 or 950ti.
 
I suspect OEMs are getting 'steamed' with their card inventories as is.

By adding another card to the mix which likely cannibalizes the most popular card in the product stack, things are going to get much more steamy ...

 


GM204-block-diagram-1024_w_600.jpg
Looking at the tomshardware review of the 980 the 960 could very well be 8 SMM's, and the memory interface seems to fit well with 2GB if so. I don't really see nVidia making a radically different die with 10 SMM blocks, and somehow carrying a bigger memory interface, I would expect the 950 ti to have 6 SMM, and the 950 to have for or 5, with less texure units (comparing to the 750 series)
 
192 bit!!??? Why is NVidia making all the bus bandwidths small on Maxwell GPUS?
Nvidia is using a clever texture compression technique that significantly lowers the amount of bandwidth needed. Pretty neat stuff!

I'm really excited about this card, it's definitely going to upset the R9 280s and maybe even the 290s @ 1080p until AMD cuts prices further or comes out with some new competition.

 


Rumor has it AMD will introduce 20nm cards in Q1 2015, while Nvidia will skip 20nm and release 16nm cards later in 2015 (Q3-4?).
 


well there are already rumor talking about AMD will stay with 28nm as well for their 300 series. and instead of TSMC 16nm FF they will move to GF 14nm (most likely FF as well) for their next node.
 
2GB, 192 bit and $250.00? That sucks. You might as well get an R9 290 for less money with twice the VRAM and way more memory bandwidth. If those specs are true, this card isn't worth more than $200.00. Low power draw is moot when you're spending $50.00 more than the card is really worth.
 
I have the MSI hooked up to a 4690K what a nice card next year I will get another when the prices drop. It even runs on a 550 80plus. My R 280X wouldn't even turn on with it. Why I bought the card 2 280's would need at least 850 or higher, a good 750 should run Both 970's in SLI. Then I can do a 4k later.
 
People here are forgetting that Maxwell were specifically designed to consume less bandwidth in order to reduce tdp by using larger l2 cache....GTX 750 and 750 ti only consume around 40% of the bandwidth available on 128 bit bus width so 192 bit seems about right size for 960...VRAM however is a different story and with next gen consoles using unified RAM and VRAM the multiplats will start demanding around 3-4 GB in the near future...
 

If they have enough memory bandwidth efficiency gains to afford dropping down to a narrower memory bus, it saves power, memory subsystem complexity, ball count on the GPU, parts count on the PCB, etc.

If you can provide the same or better performance for $20-30 lower total cost, why not? If Nvidia feels comfortable doing it, it must mean the GPU had very little use for the extra memory bandwidth.
 

Due to the many delays with 20-22nm at many chip foundries while 14-16nm appear to still be mostly on-schedule, I would not be too surprised if more chip designers decided to skip 20nm.

Depending on how high demand will be for 14-16nm though, some chip designers might get forced to re-tool their design for 20nm due to backlogs.
 
Status
Not open for further replies.