NVIDIA brings Maxwell to the high end

I bet you're too good to be invited.

 

Well, NVIDIA and AMD have the server and workstation market that has considerable use for ~225 W GPUs (Quadro, Tesla, FirePro). So for the foreseeable future the big GPUs will still be around, it's just a matter of how many gaming cards will be based on those GPUs. Note that you may be dealing with $1000 Titan-like consumer cards.

Rumors and speculation point to the "big" Maxwell chip coming out in 2015 H1, probably not in a GeForce part first.

It seems like there's some sort of split with CPUs and GPUs. At one end is the consumer/gaming GPUs that focus on low power consumption and thus you see parts such as Core M and the 750 Ti be the first of a new architecture. At the other end are the big and powerful server parts that have much higher power targets, but the chips are also more complex and they are needed less in the consumer space, so they can come later.

 

Well when the alternative is having a GPU with terrible drivers, then Nvidia are really the only option.

 

My 290 has also been a champ, though it's a single card setup and I tend to not play the latest games.

 

30 months since I got my laptop and my 7970M still has no working drivers.

 

What does "working drivers" mean?

 

AMD are retarded with their drivers (i.e. wait 2 weeks before updating) but I don't get what you mean by this,

 

I mean that regardless of OS, if I download any version of mobility drivers for my GPU, or even the regular ones, they tell me my GPU isn't supported. The only drivers that work are the preliminary ones that came on the disk. If I lost those I'd be fucked.

Doing a quick google and you'll find everyone who has this GPU has the same problem. Some found specific beta versions which worked, but AMD since took those links down. I'm essentially stuck on disk-provided 2012 drivers and can't update them.

Also worth saying that I had to return the previous ATI card I had 4970x2 because the drivers didn't work properly for that either. At the time, it just threw up errors, trying to recognise it as a regular 4970 and just generally crashing every game constantly. And none of this even touches on the abomination that is Catalyst control centre.

Seriously, have you compared CCC and Nvidia control panel. The Nvidia CP just lets you know when there's an update, and that's it. Nothing else. CCC seems like it makes it its business to interrupt you at every opportunity with various pop ups. The shit version I have on my laptop pretty much puts everything into battery save mode and runs off the integrated graphics unless I tell it otherwise manually, for every single fucking application.

Oh yeah, and a long time ago, I had a Radeon 9550. It was a long time ago but I remember having driver issues with that as well.

Seriously, my experience with ATI/AMD has never been positive. Meanwhile, up until my most recent build, I've never ever had a single problem with any of my Nvidia cards or their drivers. I can't get my 2nd GPU working at the moment (or for the past 2 years), but at the very least as far as drivers go, I've never had an Nvidia driver that worked anything less than flawlessly.

 

CCC is has always been flawless for me (besides me having to use MSI to control the fan properly.) The mobile gpu in my laptop, on the other hand, has never been anything but a pain in the ass that refused to even engage most of the time.

Also all the games I want to play are ati optimized, but that's just me.

 

The only way I could keep my E-350 "Netbook" working properly was uninstalling CCC and just keeping the driver.

 

Sometimes you just deal with a shitty OEM that couldn't give two shits about supporting their product.

Laptops aren't just boxes full of standardized parts. Nvidia & ATI are notorious for releasing a billion different SKUs (multiple SKUs per model name, even) because OEMs want to customize their increasingly integrated devices. The more integrated a device, the more the OEM is allowed to break spec.

This isn't always malicious. It's just really expensive to make sure every part is perfected implemented to spec. It's not unusual for little things like wireless radios and other integrated knicknacks to have custom drivers that can't be replaced. I'm not surprised that GPUs have also fallen prey to similar treatment.

The moral of the story is that some things don't show up on the spec sheet.

 

And if Nvidia is making things too efficient for you, there's always $230 290s out there that absolutely sweep the floor on perf/$.

At that price, the $330+ 970 and its ~10% more performance isn't looking like as much of a bargain, eh?


Hawaii
heating your home in the winter since 2013

 

So the GTX 970 is not quite what we previously thought it was: http://anandtech.com/show/8935/geforce-gtx-970-correcting-the-specs-exploring-memory-allocation

NVIDIA made an error in the specification tables. You can read the above article for the full story, but the GTX 970 actually has 56 ROPs and 1.75 MB of L2 cache instead of 64 ROPs and 2 MB of L2 cache. In addition, the 4 GB of VRAM is divided into two partitions: a 3.5 GB "fast" partition and a 0.5 GB "slow" partition.



(There's a typo: the fast segment should be (7 ports)·(7 Gbps)·(32 bits)/(8 bits/byte) = 196 GB/s.) Note that the 970 cannot read from both the "fast" and "slow" segments at the same time, so the maximum memory read bandwidth is 196 GB/s.

None of this should matter much to you performance-wise if you decided to buy the 970 mostly due to the performance benchmarks and not from numbers such as "4 GB." The GPU didn't get slower from these revelations, after all. What you think of NVIDIA itself is another question.