As a supplement to the Low Profile video card thread that I have been trying to maintain, I recently took some time to hunt down a few cheap low-end GPUs to update my own performance testing data. There are a lot of benchmarks available for the GeForce 1050ti (the current best option for a low profile, double wide video card) and 750ti (the previous champion) but there isn't a lot of interest in performance of cards slower than those, so they generally don't get articles on sites like TR. There are a few out there, but I figured no one would mind if I added to the noise with some additional benchmark data.
I don't have the time to play with hardware the way I did ten years ago, so I don't have a thorough set of tests here with multiple resolutions and a dozen games, and I skipped trying to figure out how to benchmark most titles that didn't have a built-in benchmark tool. I did try to cover multiple APIs, however.
I decided that I would just run these tests in 1080p, regular "HD" resolution, even though we all know that these cards aren't designed to push heavily into that resolution. My intent with this first batch of tests is to show the generational leap in performance going from the Kepler / GCN 1.0 parts to Pascal / GCN 4. I'll try and do a followup later on, with just the newer cards, including 720p res and maybe a handful more tests.
Anyways, without further adieu, here's the info and scores.
The test system:
i7-3770 (non-K)
16GB DDR3-1600
240GB SSD
GeForce drivers used: 388.43
Radeom drivers used: 17.11.4
This is an old Dell Optiplex 9010 MT system. It's a regular looking PC, not one of the mini versions, but they have an SFF version of it that's usually available for quite cheap. It was the last generation Optiplex to use a standard power supply connection on the motherboard, so if you bought an MT version you could replace the power supply and pop a higher-end GPU in there. Obviously the LP systems don't have that option regardless of which generation you choose, and we're not interested in cards that require additional power for these tests anyhow. The Dells are generally solid boxes, with a decent UEFI, but it won't excite you for gaming. Which I guess is the point?
The cards:
MSI GeForce GT 730 Low Profile 2GB GDDR5 (N730K-2GD5LP/OC)
EVGA GeForce GT 740 Low Profile 2GB GDDR5 (02G-P4-3740-KR)
Visiontek Radeon HD 7750 "Eye 6" 2GB GDDR5 (900614)
MSI GeForce GT 1030 Low Profile 2GB GDDR5 (GT 1030 2G LP OC, Newegg link)
MSI Radeon RX 550 Low Profile 2GB GDDR5 (RX 550 2GT LP OC Newegg link)
All with 2GB of mem to try and isolate that for generation to generation testing. The Visiontek card is the only one of the bunch that isn't low profile; it shares clock speeds with the actual low profile, 2GB GDDR5, quad mDP HD 7750 that they used to sell but finding the LP cards for a reasonable price is not very common these days, for some reason.
The tests (1080p where it could be specified, full screen on a 1080p display):
3Dmark (multiple)
Unigine Superposition ("medium" quality)
Unigine Valley & Heaven (defaults)
gfxbench (Manhattan 3.1, Car chase)
Hitman (2016) (DX12)
Yes, lots of canned benchmarks for this test. I'd like to add another game or two for the next round, but I have no idea when I'm going to have time for that. Results reported in average FPS, which, I know, frame times, time below X ms, stutter, etc. But I don't have the setup to really get that in-depth, and honestly, we're talking about video cards that sell for $70 on a good day here.
The results:
Hitman (1080p) (avg fps):
quality settings:
Code: Select all
Graphics Settings:
RESOLUTION: 1920 x 1080
ResolutionWidth = 1920
ResolutionHeight = 1080
Refreshrate = 0
Fullscreen = 1
ExclusiveFullscreen = 1
VSync = 1
VSyncInterval = 1
Monitor = 0
Adapter = 0
Aspectratio = 0
WindowPosX = 0
WindowPosY = 0
WindowWidth = 1920
WindowHeight = 1080
Stereoscopic = 0
Stereo_Depth = 3.000000
Stereo_Strength = 0.030000
WindowMaximized = 0
FocusLoss = 0
UseGdiCursor = 0
HDR = 0
ShadowQuality = 1
ShadowResolution = 1
TextureResolution = 0
TextureFilter = 4
SSAO = 0
MirrorQuality = 0
AntiAliasing = 1
LevelOfDetail = 2
MotionBlur = 0
Bokeh = 0
SuperSampling = 1.000000
Gamma = 1.000000
QualityProfile = 4
GeForce GT 730: 12.26
GeForce GT 740: 14.02
Radeon HD 7750: 22.21
GeForce GT 1030: 23.80
Radeon RX 550: 35.85
Not surprisingly the RX 550 is hugely faster than the GT 1030 here. It's also a 50% increase over the HD 7750, which is awesome. Unfortunately, the 550 doesn't seem to be available in a low profile single slot form factor from anyone. Also interesting to see such a small different between the 730 and 740. I had guessed a while back that the extra memory bandwidth would result in better performance on the 740, but it's really not even worth mentioning.
Unigine Superposition (DX12) (score / avg fps)
GT 730: 1122 / 8.4
GT 740: 1479 / 11.06
HD 7750: 1480 / 11.07
GT 1030: 2518 / 18.84
RX 550: 2390 / 17.88
Unigine Superposition (OpenGL) (score / avg fps)
GT 730: 1114 / 8.34
GT 740: 1367 / 10.23
HD 7750: 1291 / 9.66
GT 1030: 2438 / 18.24
RX 550: 1934 / 14.47
Unigine Heaven (DX11) (avg fps)
GT 730: 15.1
GT 740: 20.9
HD 7750: 21.3
GT 1030: 30.3
RX 550: 33
Unigine Valley (DX11) (avg fps)
GT 730: 16.9
GT 740: 22.4
HD 7750: 20.5
GT 1030: 36.1
RX 550: 33.1
Unigine Valley (OpenGL) (avg fps)
GT 730: 16.2
GT 740: 21.6
HD 7750: 18.9
GT 1030: 34.4
RX 550: 29.8
Unigine's tests don't get a lot of love and I'm not sure why. I'd love to see them add Vulkan to the newest one. Surprisingly the GT 1030 is trading blows with the RX 550 here. Put more weight into the Hitman results since it's an actual game, but I was surprised to see these results.
gfxbench Car Chase (avg fps)
GT 730: 29.6
GT 740: 30.16
HD 7750: 47.81
GT 1030: 59.9
RX 550: 76.74
gfxbench Manhattan 3.1 (avg fps)
GT 730: 35.49
GT 740: 51.37
HD 7750: 58.1
GT 1030: 59.85
RX 550: 83.51
I know gfxbench is typically mobile-oriented, but with some simpler tests these cards are able to stretch their legs a little more. This should give you an idea how these cards will do playing older titles.
3dmark cloud gate (graphics score / fps 1 / fps 2):
GT 730: 11652 / 48.69 / 52.8
GT 740: 13553 / 57.35 / 60.6
HD 7750: 16633 / 74.98 / 69.84
GT 1030: 20340 / 82.84 / 94.85
RX 550: 28923 / 137.56 / 115.81
3dmark sky diver (graphics score / fps 1 / fps 2):
GT 730: 5388 / 24.78 / 24.43
GT 740: 6256 / 28.99 / 28.16
HD 7750: 7650 / 34.33 / 35.56
GT 1030: 11253 / 52.16 / 50.64
RX 550: 13035 / 62.49 / 56.82
3dmark fire strike (graphics score / fps 1 / fps 2):
GT 730: 1729 / 7.92 / 7.15
GT 740: 2040 / 9.3 / 8.48
HD 7750: 2523 / 11.45 / 10.53
GT 1030: 3569 / 16.44 / 14.7
RX 550: 4278 / 19.63 / 17.68
And lastly some 3dmark results. It's a real shame the RX 550 is dual-slot. MSI had a dual slot, low profile RX 460 so I figured someone would have done a low profile, single slot RX 550 by now. Maybe we'll see one eventually, as it would be a great successor to the long-lived HD 7750 / R7 250. As it is, the GT 1030 is a very nice upgrade over the long-lived GT 730. I will be glad to see NVIDIA retire all of the old Fermi/Kepler products.
This data, and any future tests I might run, will eventually make their way into the "best low profile" thread. I have kept the GT 1030 and RX 550 card to use in future comparisons / additional benchmarks, but I've already sold off the others (bought on ebay, sold on ebay, such is the life of an old video card).