Testing Nvidia RTX 3080

Amazing performance improvement; better price than the RTX 2080 Super

Posted: By Jon Peddie 09.15.20
Nvidia’s RTX3080 in box presentation 


PC Benchmarking got started back in the mid-to-late 1980s when PC clones became abundant. Of course, they weren’t perfect clones, they had their own variations and features and to the new PC's fidelity and ability to run programs compared to a genuine IBM PC, we used the Microsoft Flight Simulator 2.0.

Today, the landscape is even more varied. We have 4K screens, HDR color, ray tracing, realtime physics, particle effects, and AI. A modern interactive game today makes us a player in a cinematic experience. The games give hours, if not months, of entertainment, the equipment lasts for years, the value per entertainment dollar is extraordinary. And every year, sometimes twice in a year, the envelope gets pushed further toward the Holodeck and ultimate immersion. Benchmarks now tell us how much progress has been made from product generation to product generation, and they tell us what kind of experience we might get from the new products compared to the old. 

This year, we got Nvidia’s latest and greatest effort—the RTX 3080 based on the company’s new Ampere GPU. The AIB is beautiful to look at and comes with 10 GB of super high-speed Micron GDDR6+ memory, unique to Nvidia.

With Ampere, Nvidia doubled the cores in the RT section (Source: Nvidia)



RTX 3080

The new processor consists of an Nvidia Ampere GPU (GA102-200-K1-A1) fabricated in 8 nm, at Samsung, has 28 billion transistors, and is 628 mm² in size. The GPU has a base clock of 1.44 GHz and can surge to 1.7 GHz.

Nvidia expanded the Ampere with a new datapath that has twice as many FP32 operations (then the Tesla GPU) and delivers 128 FMA per clock. There are 8704 shaders plus 272 Tensor cores, and 68 RT cores. The chip has 272 texture mapping units (TMU), 96 ROPS, and 68 streaming multiprocessors (SMs), which are the part of the  GPU that runs the CUDA kernels. That is the same amount of SMs as the Nvidia's 2080 TI, but Nvidia says that the 2X FP32 redesign increases the CUDA cores by 100X.  With an L1 cache of 128 kb per SM, and an L2 cache of 5 MB, the GPU can deliver 164.2 Gpixels/s, and hit 29.7 FP32 TFLOPS. Its texture rate is 465 GTexels/s.
With an L1 cache of 128 kb per SM, and an L2 cache of 5 MB, the GPU can deliver 164.2 Gpixels/s, and hit 29.7 FP32 TFLOPS. Its texture rate is 465 GTexels/sec.

There is also a new L1/texture system with twice the bandwidth and 33% more total capacity with twice the cache partition size. Ampere has new accelerator cores. The second-generation RT Core can compute with twice triangle intersection rates, and the third Generation Tensor Core can do twice the math for sparse matrices.

The company also added new hardware in the GPU for motion-blur hardware acceleration.

The dual-slot AIB has 10 GB of Micron’s new GDDR6X with a 320-bit interface, and can sustain a data rate of 760 GB/s running with a clock of 1.19 GHz (19 Gbps effectively).

The AIB draws 320 W, is 11.2-inches long, weighs almost three-pounds, and has a PCI-Express 4.0 x16 interface and three DisplayPort 1.4 and one HDMI 2.1 connectors.

AIB CUDA Cores Memory Memory Interface Boost Clock Power Price 
RTX 3080 8704 10GB GDDR6X 320-bit 1.71GHz 320w $699
RTX 2080 Ti 4352 11GB GDDR6 352-bit 1.54GHz 250w $1,700
RTX 2080 Super 3072 8GB GDDR6 256-bit 1.82GHz 250w $789
Comparison of Nvidia AIBs


What we ran

We opted for packaged, easy to install, and run tests. We ran specific synthetic ray tracing benchmarks: 3Dmark Port Royal, Blender’s Cycles benchmark, Bright Memory’s Infinite RTX benchmark, Crytek’s Neon Noir benchmark, and Surgical Scalpel’s Boundary ray tracing benchmark. And we ran Port Royal, as well as Time Spy and Fire Strike. For games, we ran only those that had been enabled for ray tracing and had built-in benchmarks in order to get consistency and repeatability in the tests. We chose: Metro Exodus, Shadow of the Tomb Raider, and Wolfenstein Youngblood.

We ran all the tests on the RTX 2080 Super to get a baseline (using the new driver 456.16).

The unboxing of 3080 can be seen here:

Test results

Our test system consisted of an Intel Core i9-9900k, Windows 10 x64 Version 1909, an EVGA Dark Z390 FTW motherboard, 32 GB Corsair RAM, and 1.2 TB of NVMe.

We ran tests with DLSS on and off, or rendering set at Quality or Balanced. Then we averaged the raw scores, and we averaged the fps scores to arrive at the performance value of the Pmark calculation.

Pmark fps with DLSS on and off


The benefit of DLSS clearly shows in both the 2080 Super and the new RTX 3080.

Pmark scores


The values used in the calculations are shown in the following table.

  RTX 2080 Super DLSS on RTX 2080 Super DLSS off RTX 3080 DLSS on RTX 3080 DLSS off
Average FPS 44.9 15.2 69.1 24.1
Average Score 4080 0 6431 0
Power (W) 250 250 320 320
Price ($) $789.00 $789.00 $699.00 $699.00
K (FPS) 100000 100000 100000 100000
K (Score) 1000 1000 1000 1000
Values used in Pmark calculation


The new RTX 3080 makes up for its greater power consumption at a lower price, which makes the new AIB really powerful and with good price-performance.

What we didn’t run

Nvidia points out that the RTX 3080 is also targeted at content creator specialists, and the company lists several apps (like Adobe Premiere Pro, Autodesk Arnold, Blender Cycles, Blackmagic Design DaVinci Resolve, and D5 Render) that could be used to reveal the AIB’s superior performance. We decided not run those tests for several reasons. Primarily our interest is in gaming performance. 

We had our reasons. The Adobe tests, which use a specific file, are interesting for the Mercury Rendering Engine (MPE) stuff (which is GPU accelerated). However, as adorable as Big Bunny is, there is no RT in that old clip so all that’s being revealed is the rendering time of the 3080 over the 2080. Likewise, testing the h.264 encode/decode time of a CPU vs. the 3080 isn’t interesting at all, or unique, what’s to be learned in doing that?

What do we think?

These things really look cool. Nvidia put some real industrial design work into the shroud of the 3080, and it looks, well, it just looks cool, sleek, and unique.

Nvidia’s RTX 3080 is a well designed aesthetically pleasing AIB (Source: Babeltechnews)


When I plugged it into the EVGA Black Z390 motherboard, I noticed how cool that looked too. I remember when motherboards and graphics AIBs were just dull looking green fiberglass things with miscellaneous little black semiconductors stuck on them and wires dangling off or around them. Look at the images on Mark Poppin’s site (above), especially the last one. The trim bespeaks the dual push-pull fan design.