NVIDIA GeForce GTX 980 Video Card - Performance Overview

Linus Tech Tips ·Linus Tech Tips ·2015-05-07 · 2,008 words · ~10 min read
Floatplane YouTube

Transcript

JSON SRT VTT 154
0:00 Another day, another new graphics card.
0:03 Welcome to the Lionus Tech Tips performance overview of the GTX 980 from
0:08 NVIDIA, their most advanced GPU
0:18 yet. DKL's new Rip Jaws 4 series is
0:21 their ultimate DDR4 memory with speeds of up to
0:24 3,333 MHz available. Click now to learn
0:27 more. As usual, we'll start with the unboxing and physical overview. The
0:31 Media Sample GTX 980s come packaged in this sweet box that has the card upright
0:37 looking all majestic with its magnesium alloy fan housing, aluminum frame, and
0:43 trivalent chromium plating with H.
0:47 What's that? Sorry, you've seen this user
0:50 configurable green LED lit GeForce logo before. NVIDIA hasn't really changed
0:55 their cooler design since the 690. Ah, yes. No, I can see how you might think
1:00 that, but you haven't seen the back yet. Check it out. Now, we get a full metal
1:06 back plate with a classy looking design and surface. Now, the issue with back
1:11 plates, and presumably the reason NVIDIA has never done a reference card with one
1:14 before, is that it restricts the intake of a card installed in the next slot
1:19 over in tightly packed SLI configurations. So NVIDIA made this
1:24 strategic chunk of the back plate removable for users with lots of cards
1:29 right next to each other in SLI. Not that you need to run SLI just to hook up
1:35 enough monitors anymore. As we move over to the IO, yeah, that's some pretty good
1:39 stuff right there. Three DisplayPort ports, so that's up to three G-Sync
1:44 displays off of a single card. An HDMI
1:47 2.0 port. So yeah, that's 4K 60 Hz over
1:50 HDMI. and a dual link DVI port can all
1:54 be found here. Any four of which can be used at one time on this beast of a
1:59 graphics card. Which leads us to the next notable thing physically. This is a
2:04 flagship 80 class card, but like the GTX
2:07 680, it only has two six pin connectors.
2:12 A hint at the incredible power efficiency of the Maxwell architecture
2:16 at the core of the GM 204 chip that powers this card. The GTX 980 not only
2:22 hung around at the top of our performance charts, but did it while
2:26 consuming less power than GK 110based
2:30 cards like the 780 and 780 Ti and while
2:33 running extremely quietly. So, let's talk about Maxwell for a bit here.
2:37 Without shrinking the manufacturing process, so they're still on the same
2:41 28nmter node that they used to build the Kepler based GTX 680, NVIDIA has
2:46 effectively reduced TDP by 30 W,
2:49 increased the number of CUDA cores by 33%, increased base clocks and memory
2:54 speed by around 12% and doubled the number of render output units, as well
2:59 as adding some other cool hardware bits to enable some really cool technology.
3:04 Pretty amazing, actually. So, first up in the new technology list is their
3:09 improved third generation delta color compression. A technique that allows
3:13 data coming out of the core to be compressed by a factor of 2:1, 4:1, or
3:18 even 8:1 before being committed to memory by storing the value of a change
3:23 versus the previous frame rather than the value itself. This combined with
3:27 improved caching allows the GTX 980, in spite of its mere 256-bit memory
3:33 interface, to achieve up to around 30%
3:36 better effective memory performance compared to the GTX 680. In this scene,
3:42 the pink highlighted screen elements are taking advantage of this technology.
3:46 Next up is a feature that's all about getting the most out of older games
3:50 where you have lots of GPU horsepower to spare. DSR or dynamic super
3:55 resolution is mostly just another name for a feature that has existed forever
4:00 but is now easier to use. Super sampling
4:04 anti-aliasing. The way it works is it renders the scene at a higher
4:08 resolution. So you can choose 2x, 3x, or 4x in the control panel. Then
4:12 intelligently downscales it with a filter whose sharpness or softness can
4:16 be adjusted to let you use your extra GPU horsepower that otherwise was just
4:22 sitting there to yield potentially a much more pleasing gaming experience.
4:27 It's not the same as actually running on a 4K display, but it's much better than
4:32 running native 1080p. The effect is especially noticeable on the grass in
4:37 Dark Souls II here in this footage that I took at the NVIDIA event. The next
4:42 feature is about getting more performance when we don't have enough,
4:46 and it's called MFAA or frame
4:49 anti-aliasing. That's short for multi-frame sampled anti-aliasing. Grow
4:54 up. The way this works is instead of taking a large number of samples with a
4:59 static pattern to smooth the edges of on-screen objects or textures, MFAA
5:05 takes fewer samples per frame, half as many from what we've been told, but uses
5:10 new hardware in the GTX 980 and 970 to
5:13 alternate the positions with each frame. Then it applies a clever filter. So the
5:18 performance impact is about equivalent to two times MSAA, but the resulting
5:24 image looks pretty much identical to four times MSAA. This feature isn't
5:29 enabled in the driver yet, so expect an update once we get a chance to play
5:33 around with it some more, but the initial results are impressive. Check
5:37 out the effect on the transparent textures like these wires in this
5:41 Battlefield 4 demo that NVIDIA was showing off at the launch event. Another
5:45 thing we won't be able to test for a while is the improvements that NVIDIA
5:48 has been making to the VR experience on their cards. The most notable of which
5:52 is called auto asynchronous warp. A
5:55 cheat, I guess, that allows the head tracking of a VR headset like the Oculus
5:59 Rift to sample its position again right
6:03 before a frame is finished being drawn to reduce the player's perception of the
6:08 delay between physical motion and on-screen motion. And this will be
6:12 important for folks who have trouble with motion sickness in VR. Which leads
6:16 us to the final feature, their VxGI global illumination technique that we
6:20 also can't really test here because while Unreal Engine 4 is getting
6:24 support, it'll be a while before we get real games with the tech, but it looks
6:28 so freaking cool. NVIDIA has a couple of demos to show it off. One where they
6:31 render a Cornell box, so you can see the way the direct light enters at the top,
6:36 so the tops of the shapes are the brightest with realistic soft shadows and color bleeding from reflective
6:40 surfaces. and one where they recreated as closely as possible the conditions of
6:45 this iconic photo from the 1969 lunar landing to demonstrate that it's
6:50 completely realistic and wasn't taken
6:53 somewhere in Area 51 or whatever. Now, these two demos wouldn't be impressive
6:58 at all except that they're being done in real time instead of using path tracing,
7:03 a phenomenally computationally intensive global illumination technique that
7:07 involves following light rays as they bounce around the scene. like all of
7:11 them. NVIDIA is using dedicated hardware to split the scene into cubes called
7:16 voxels that act as dynamic light sources
7:19 that approximate the results of a more accurate rayraced rendering method but
7:25 in real time. All right, so as cool as
7:28 all that technology is, how does the
7:32 card actually perform? That's what you want to know. Great question, Luke. Take
7:36 it away. Before we get into things, I want to point out that overclocking all
7:39 of our cards is yet again backed by popular demand. And because of that, you
7:43 will be able to find the overclocking results and settings in the graphics
7:46 card overclocking dock, which should be linked in the description below this
7:50 video. Also, when reading the settings for the GTX 980 in the overclocking dock
7:54 below, there's no reason to worry. There is a reason why I ignore the memory
7:58 clock speed. The 980, much like the 780 Ti before it, features a fantastic
8:02 technology called autobalancing power rails, which in its most basic
8:06 description, allows the card to take extra power overhead from parts of the
8:10 card that aren't using it and move it elsewhere, allowing you to focus on one
8:14 specific thing, which in this case resulted in the GPU clock speed being
8:18 cranked through the freaking roof, which is awesome because then you can access
8:22 all of that FPS that you wouldn't otherwise be able to. Performance of the
8:25 GTX 980 is honestly a little bit surprising in a good way. To give proper
8:30 context for that statement, we have to look at the rumor mill. Personally, I like to avoid rumors, but sometimes I
8:35 don't really have a choice. For instance, on the September 12th version
8:39 of the WAN Show, my brother at Cold SE on Twitter, and I discussed how the 980
8:44 was, according to rumors, not supposed to beat out the 780 Ti. So, when I
8:49 finally got the card in hand, I was delighted to see that more often than not, it was able to beat the 780 Ti in
8:55 both 1080p and 4K benchmarks. Now, my
8:58 overclock for the 980 feels a little bit high, running at a range of 1503 to 1492
9:04 MHz, but I really have no idea as I have
9:07 no one else's overclocks to compare to yet. And to be fair, my 780 Ti overclock
9:11 is very aggressive anyways. So, I haven't exactly done the 980 any favors
9:16 in regards to the 780 Ti's clock speed.
9:19 But don't forget, all of this performance awesomeness doesn't just
9:22 stand on its own. Not only are you paying considerably less for this card
9:26 than you would pay for a 780 Ti, but you also pay less in the future. Underload,
9:31 the 980 has considerably less power draw than the 780 Ti, which is nice if you're
9:35 actually paying for your power bill. And if you're not paying for your power
9:39 bill, maybe you can use that little tidbit of information to convince your parents that it might be time for an
9:43 upgrade. Anyways, thanks for watching my segment, guys. Back to you, Linus.
9:46 Thanks, Luke. All right, guys. So, you've seen the results and you've seen
9:50 the pricing. Links as always in the video description for pricing. So, let
9:54 us know what you think. For non-overclockers, as I'm sure you'll
9:58 have noticed from the other reviews around the web, GTX 980 is a total tank.
10:04 But when overclocking the GTX 780 Ti, we
10:08 observed that this older GK 1110-based
10:11 card becomes much more competitive if you're willing to sacrifice features and
10:16 power consumption. So, let us know what you think. If you had over 500 bucks to
10:20 spend on a graphics card, which one would get your money? Also, let us know
10:25 in the comments what other coverage you'd like to see of this card. An
10:29 overclocking guide, um, SLI performance scaling. Any other ideas? let us know.
10:34 Thanks for watching, guys. Like this video if you liked it. Dislike it if you
10:37 disliked it. You can also leave a comment if you have feelings that are just more complicated than this. Also,
10:42 check out the link in the video description to support us if you want to buy a cool t-shirt, give us a monthly
10:46 contribution, or change your Amazon bookmark to one with our affiliate code.
10:49 So, when you buy your new graphics card, we'll get a small kickback as a, you
10:53 know, thank you for providing us with these videos, Lionus. We sure do think
10:56 you're awesome. Here's your affiliate code being used in Amazon. Helps us out
11:00 a lot. Thanks again for watching. And as always, don't forget to subscribe.