The NVIDIA GeForce RTX 2060 6GB Founders Edition Review: Not Quite Mainstream
by Nate Oh on January 7, 2019 9:00 AM ESTCompute & Synthetics
Shifting gears, we'll look at the compute and synthetic aspects of the RTX 2060 (6GB). As a cut-down configuration of the TU106 GPU found in the RTX 2070, we should expect a similar progression in results.
Starting off with GEMM tests, the RTX 2070's tensor cores are pulled into action with half-precision matrix multiplication, though using binaries originally compiled for Volta. Because Turing is backwards compatible and in the same compute capability family as Volta (sm_75 compared to Volta's sm_70), the benchmark continues to work out-of-the-box, though without any particular Turing optimizations.
For Turing-based GeForce, FP32 accumulation on tensors is capped at half-speed, thus resulting in the observed halved performance. Aside from product segmentation, that higher-precision mode is primarily for deep learning training purposes, something that GeForce cards wouldn't be doing in games or consumer tasks.
Moving on, we have CompuBench 2.0, the latest iteration of Kishonti's GPU compute benchmark suite offers a wide array of different practical compute workloads, and we’ve decided to focus on level set segmentation, optical flow modeling, and N-Body physics simulations.
Moving on, we'll also look at single precision floating point performance with FAHBench, the official Folding @ Home benchmark. Folding @ Home is the popular Stanford-backed research and distributed computing initiative that has work distributed to millions of volunteer computers over the internet, each of which is responsible for a tiny slice of a protein folding simulation. FAHBench can test both single precision and double precision floating point performance, with single precision being the most useful metric for most consumer cards due to their low double precision performance.
Next is Geekbench 4's GPU compute suite. A multi-faceted test suite, Geekbench 4 runs seven different GPU sub-tests, ranging from face detection to FFTs, and then averages out their scores via their geometric mean. As a result Geekbench 4 isn't testing any one workload, but rather is an average of many different basic workloads.
We'll also take a quick look at tessellation performance.
Finally, for looking at texel and pixel fillrate, we have the Beyond3D Test Suite. This test offers a slew of additional tests – many of which use behind the scenes or in our earlier architectural analysis – but for now we’ll stick to simple pixel and texel fillrates.
Facebook
Twitter
RSS
134 Comments
View All Comments
sing_electric - Monday, January 7, 2019 - link
It's likely that Nvidia has actually done something to restrict the 2060s to 6GB - either though its agreements with board makers or by physically disabling some of the RAM channels on the chip (or both). I agree, it'd be interesting to see how it performs, since I'd suspect it'd be at a decent price/perf point compared to the 2070, but that's also exactly why we're not likely to see it happen.CiccioB - Monday, January 7, 2019 - link
You can't add memory at will. You need to take into consideration the available bus, and as this is a 192bit bus, you can install 3, 6 or 12 GB of memory unless you cope with hybrid configuration thorough heavily optimized drivers (as nvidia did with 970).nevcairiel - Monday, January 7, 2019 - link
Even if they wanted to increase it, just adding 2GB more is hard to impossible. The chip has a certain memory interface, in this case 192-bit. Thats 6x 32-bit memory controller, for 6 1GB chips. You cannot just add 2 more without getting into trouble - like the 970, which had unbalanced memory speeds, which was terrible.mkaibear - Tuesday, January 8, 2019 - link
"terrible" in this case defined as "unnoticeable to anyone not obsessed with benchmark scores"Retycint - Tuesday, January 8, 2019 - link
It was unnoticeable back then, because even the most intensive game/benchmark rarely utilized more than 3.5GB of RAM. The issue, however, comes when newer games inevitably start to consume more and more VRAM - at which point the "terrible" 0.5GB of VRAM will become painfully apparent.mkaibear - Wednesday, January 9, 2019 - link
So, you agree with my original comment which was that it was not terrible at the time? Four years from launch and it's not yet "painfully apparent"?That's not a bad lifespan for a graphics card. Or if you disagree can you tell me which games, now, have noticeable performance issues from using a 970?
FWIW my 970 has been great at 1440p for me for the last 4 years. No performance issues at all.
atragorn - Monday, January 7, 2019 - link
I am more interested in that comment " yesterday’s announcement of game bundles for RTX cards, as well as ‘G-Sync Compatibility’, where NVIDIA cards will support VESA Adaptive Sync. That driver is due on the same day of the RTX 2060 (6GB) launch, and it could mean the eventual negation of AMD’s FreeSync ecosystem advantage." will ALL nvidia cards support Freesync/Freesync2 or only the the RTX series ?A5 - Monday, January 7, 2019 - link
Important to remember that VESA ASync and FreeSync aren't exactly the same.I don't *think* it will be instant compatibility with the whole FreeSync range, but it would be nice. The G-sync hardware is too expensive for its marginal benefits - this capitulation has been a loooooong time coming.
Devo2007 - Monday, January 7, 2019 - link
Anandtech's article about this last night mentioned support will be limited to Pascal & Turing cardsRyan Smith - Monday, January 7, 2019 - link
https://www.anandtech.com/show/13797/nvidia-to-sup...