Slashdot Mirror
NVIDIA Launches New Midrange Maxwell-Based GeForce GTX 960 Graphics Card
MojoKid writes NVIDIA is launching a new Maxwell desktop graphics card today, targeted at the sweet spot of the graphics card market ($200 or so), currently occupied by its previous gen GeForce GTX 760 and older GTX 660. The new GeForce GTX 960 features a brand new Maxwell-based GPU dubbed the GM206. NVIDIA was able to optimize the GM206's power efficiency without moving to a new process, by tweaking virtually every part of the GPU. NVIDIA's reference specifications for the GeForce GTX 960 call for a base clock of 1126MHz and a Boost clock of 1178MHz. The GPU is packing 1024 CUDA cores, 64 texture units, and 32 ROPs, which is half of what's inside their top-end GeForce GTX 980. The 2GB of GDDR5 memory on GeForce GTX 960 cards is clocked at a speedy 7GHz (effective GDDR5 data rate) over a 128-bit memory interface. The new GeForce GTX 960 is a low-power upgrade for gamers with GeForce GTX 660 class cards or older that make up a good percentage of the market now. It's usually faster than the previous generation GeForce GTX 760 card but, depending on the game title, can trail it as well, due to its narrower memory interface.
My "sweet spot" is $100, with perhaps $20 fudge factor for impatience and/or shipping. Hence, I am now running an Asus 450 GTS OC 1GB. If I were to buy a card today, which one would I buy assuming I weren't even considering throwing away money on an ATI card?
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
I think the 'range' depends on what resolution you are playing at..
For 3840x2160 - Low end
For 2560x1440 - 'Midrange'
For 1920x1080 - High end
There are other, better reviews for this card, for instance Tom's hardware, but every single hardware review story on Slashdot seems to be obliged to link to Hothardware.
So, which editor own's Hothardware shares?
is that one processor or a collection ?
Each "CUDA core" is a single processor, which can run a CUDA instance, or an OpenCL instance, or run a shading algorithm. The card has 1024 of them.
Mid-range Nvidia cards alway seem to launch a little expensive. There's always an older model from AMD that's got a better value on paper.
But I'd take this in a heartbeat over an AMD counterpart. The maxwell chips are leagues ahead of anything AMD's got. Very low power consumption and solid performance with great features. Maxwell is a huge leap over their previous offerings. Cards that are 150% faster while consuming 60% less energy than their previous generation counterparts.
"AMD has shitty drivers" is an old meme but at this point I'd say the Nvidia software stack is currently far superior to AMD's offerings. I've cards from both vendors and AMD's 'Suite' (raptr) is a pain in the ass and is playing catch-up to Nvidia's counterpart (Geforce Experience) - Yeah you don't need the suite but they make software updates easy, and can help less experienced users configure their software. They also include nice things like free game recording and streaming software. (And in Nvidia's case it's required to stream games to a Shield handheld or tablet)
A legit complaint I can see is 2GB of memory. Modern games are starting to crave lots of memory. I suspect Nvidia may be gating that feature in higher tier SKUs, or maybe we'll see 4GB cards not long after launch.
C'mon, guys, this is copy-pasted marketing fluff. Better is expected of you.
Editor, A1-AAA AmeriCaptions
From the reviews I've read it's basically it's cool, quiet, has all the latest features but in the end has almost the same performance as the 760. Where the 970 went really aggressive on pricing the 960 looks to be their "money maker" that TechPowerUp called "a cheap-to-make GPU they paired with an extremely cost-efficient PCB design that has loads of margins in it for future price wars with AMD".
Not that I think AMD is in any mood for price wars after their Q4 financials, they posted a $330 million loss, a lot of one-time charges but also a $58 million inventory write-down on their APUs. Last quarter revenue was down 13% and guidance for Q1 2015 is another 15%, they really could use some killer graphics card very, very soon. Or to put it even more harshly, their last quarter wiped out 2/3rds of their stockholder value and one more quarter like would put them in bankruptcy court.
Live today, because you never know what tomorrow brings
Isn't Maxwell that talking pig in the teevee commercials? Not the first thing I want people to think of when they hear about my new product.
Some mornings it's hardly worth chewing through the restraints to get out of bed.
nVidia's been using that for years to keep their mid range from biting into their high end. At those price points I wish I was confident enough in AMD's driver stability to buy an R9 270X. 256 bit interface + 1500 mhz core clock for $210 bucks.
Hi! I make Firefox Plug-ins. Check 'em out @ https://addons.mozilla.org/en-US/firefox/addon/youtube-mp3-podcaster/
I'm glad to see ASUS understands that a lot of gamers have a small PC. Their Asus Strix GeForce GTX 960 looks like it might fit in a Cooler Master Elite 110.
Get free satoshi (Bitcoin) and Dogecoins
My problem with this new card is, as far as I can tell it barely meets the minimum requirements for some next-gen games. Isn't the min req for Assassins Creed Unity like a GeForce 680?
The specs for all new games are out of whack with the latest generation of video cards. If you buy a $200 card, you should be able to play any game released in the same year (though probably not on Ultra).
I just don't want to pull the trigger on a $350 video card, but it looks like game devs are making that the entry level for AAA games.
You are welcome on my lawn.
You can get a used 280X for cheaper that will wipe the floor with it. The dual fan models are very quiet. They use a bit of power, but not at idle... so unless you game 24/7 and live in a super pricy electric area OR have a garbage PSU, its the far better deal. The plain ol 280 is also better.
I bought a 980 over the holidays. Runs 3x 1080p excellently. If I had the extra money to spend, I would have bought g-sync monitors, but they would cost more than my PC did, so screw it.
Buck Feta. You know what to do.
That's exactly what Maxwell did vs Kepler.
No, a CUDA core is better thought of an ALU or sub-ALU, speaking in terms of cores is clear abuse because AMD or nvidia once started to use that term. As a limited analogy, the original Pentium has two integer pipelines but is not called a dual-core CPU.
"CUDA cores" are organised into units that house 128 of them here, called an "SMM". But to understand how things are dealt with from the software point of view (threads, warps) some extensive reading is needed.
There's finally a reasonable bump coming this year from cards with a new architecture and using a totally different ram system. Up to 9x higher bandwidth.
I doubt they won't bleed us with slow bumps and increments initially but none the less I suspect the first 'significant bump' in years might occur when these cards come out in the next couple of months. Look to see the 9xx series drop in price if you prefer nvidia - but unless you need an upgrade right now urgently, apply some patience and wait to see what this HBM setup can do.
Cool. Our research folks at $DAYJOB have been building GPU-computing clouds, and have found that for many workloads, the GTX 750i was extremely cost-effective (that's the predecessor to this card, and costs include the server you plug it into and electricity as well as the graphics card), compared to much higher-end computation-focused systems. But they bought their lab hardware months ago; this looks to be about 50% faster, for a slightly higher price, so that's a win.
Bill Stewart
New Fast-Compression-only CPR http://preview.tinyurl.com/dy575ks
120 Watts
The summary should have included this.
Not completely true. When a chip draws less power to perform the same functions you can the push the clock, because consuming less power means less waste heat. As you raise the clock you get more speed. There is of course a balance to be struck, but ragardless figuring out how to accomplish the same work with less power will allow you to push the performance before hitting the limit of your heat disipation capabilities.
A CUDA core is basically an active hardware thread [pool.]
For example my GTX Titan has 2,688 CUDA Cores.
This number is derived from:
Streaming Multiprocessors: 14
* 192 Cores/SM
====
2,688 Cores
In practice that means you have 2,688 threads doing "real work" at any one time.
See this SO question/answer
http://stackoverflow.com/quest...
This is incorrect. CUDA cores are at a higher level than ALUs or FPUs; they're like small, simple cpu cores. They can do integer and floating point arithmetic, and they have hardware support for thread context switching, which they can generally do in a single clock tick. There can be varying numbers of CUDA cores in a streaming multiprocessor, but CUDA thread blocks are arranged in groups of 32 ("warps") which share a scheduling unit and which execute the same instruction in lock-step on different memory addresses. When threads running on adjacent CUDA cores read and write adjacent memory addresses, memory access is very efficient ("coalescing").
CUDA cores aren't as capable or powerful as CPU cores; they don't have things like branch prediction or preemptive execution, but they are cores none the less. They achieve high performance via sheer numbers - thousands of cores on top-end GPUs - and they're very good at streaming, which consists of doing the same operation in parallel on many array elements when each operation is independent of all the others.
Chuck Norris: Socialism == a thousand years of darkness.