How & Why NVIDIA Calculates RTX-OPS For GeForce RTX!

When it comes to performance, everyone knows what GigaPixels/s, GigaTexels/s, GB/s or even TFLOPS mean. But what the heck is RTX-OPS? And how did the GeForce RTX 2080 Ti get 78 Tera RTX-OPS of performance?

In this article, we are going to share with you exactly what RTX-OPS means, and how NVIDIA arrived at the magical 78 Tera RTX-OPS figure for the flagship GeForce RTX 2080 Ti graphics card. You will also learn why RTX-OPS cannot be compared with any other performance metrics out there…

 

NVIDIA RTX-OPS

When NVIDIA first announced the GeForce RTX specifications, one performance metric that stood out was RTX-OPS per second. It was easy enough to understand the other new metric – Giga Rays per second, but what the heck is RTX-OPS?

Is that a measure of the overall performance of the GeForce RTX card? Or just the new Tensor and Ray Tracing cores? What exactly does it mean to have 78 Tera RTX-OPS per second of performance?

 

Jeff Yen Explains RTX-OPS Calculations

In this short video, NVIDIA Director of Technical Marketing for APAC, Jeff Yen, explains how they calculate RTX-OPS, and the rationale behind creating it.

 

Why NVIDIA Created RTX-OPS

The NVIDIA Turing architecture is quite different from NVIDIA Pascal. Here are some improvements that prevent a direct performance comparison between the two GPU families.

  1. The new Turing SM (Streaming Multiprocessor) can execute integer and floating-point instructions simultaneously, using independent execution units.
  2. Turing-based GPUs have new Turing Tensor Cores that are optimised for inferencing operations, with support for much faster but less precise INT8 and INT4 modes.
  3. Turing also enables real-time ray tracing capabilities for the first time, with dedicated Turing RT Cores.

The performance of these separate components are measured differently. The GeForce RTX 2080 Ti (US | UK | Malaysia), for example, boasts :

  • Floating-point performance of up to 28.5 TFLOPS (FP16) or 14.2 TFLOPS (FP32).
  • Tensor core performance of up to 113.8 TFLOPS.
  • Ray tracing performance of up to 10 Giga Rays per second or 100 tera-ops per second.

So where does 78 Tera RTX-OPS come in?

[adrotate group=”1″]

In a game that uses hybrid rendering – combining traditional shader operations with real-time ray tracing and AI inferencing, the different “engines” in the Turing GPU can work simultaneously. Hence, their depiction of the Turing frame :

This is based on NVIDIA’s analysis of a typical hybrid rendering workload with DLSS enabled.

  • Deep neural network processing of DLSS using the Tensor cores takes about 20% of the frame time.
  • FP32 shading using the CUDA cores take about 80% of the frame time.
  • Ray tracing takes about half the FP32 shading time, but are performed by separate RT cores.
  • The new integer execution units (independent of the floating-point units) take up about 28% of the frame time.

In a hybrid rendering situation, the performance of the separate engines will not tell us what the actual performance will be. Hence, NVIDIA invented RTX-OPS – a workload-based estimate of the Turing GPU’s overall performance.

 

This Is How NVIDIA Calculates RTX-OPS (In Detail)

Using their analysis of the typical hybrid rendering workload, NVIDIA derived a formula that they feel will give us a better representation of the Turing GPU’s overall performance.

RTX-OPS is basically calculated using this formula :

Peak FP32 performance (in TFLOPS) x 80%
+
Peak INT32 performance (in TFLOPS) x 28%
+
Peak Ray Tracing performance (in tera-ops per second) x 40%
+
Tensor core performance (in TFLOPS) x 20%

When you plug in the individual performance figures for the GeForce RTX 2080 Ti (rounded up), you will get :

(14 x 80%) + (14 x 28%) + (100 x 40%) + (114 x 20%)
= 78 Tera RTX-OPS

So that, ladies and gentlemen, is how NVIDIA calculates RTX-OPS! Now you see why it cannot be used to compare the performance of GeForce RTX cards with the previous-generation GeForce GTX cards, or any other graphics cards in the market.

What do you think of their new performance metric, and the formula that created it? Is this a useful metric for you? Tell us in the comments!

 

Where To Purchase The GeForce RTX?

Here are some GeForce RTX 2080 Ti and GeForce RTX 2080 purchase links in Malaysia :

  • GIGABYTE GeForce RTX 2080 Ti WindForce OC 11G : RM 5,699
  • GIGABYTE GeForce RTX 2080 Ti Gaming OC 11G : RM 5,799
  • ZOTAC GAMING GeForce RTX 2080 Ti Triple Fan : RM 5,379
  • ZOTAC GAMING GeForce RTX 2080 Ti AMP : RM 5,639
  • GIGABYTE GeForce RTX 2080 WindForce OC 8G : RM 3,899
  • GIGABYTE GeForce RTX 2080 Gaming OC 8G : RM 3,999
  • ZOTAC GAMING GeForce RTX 2080 Blower : RM 3,699
  • ZOTAC GAMING GeForce RTX 2080 Twin Fan : RM 3,719
  • ZOTAC GAMING GeForce RTX 2080 AMP : RM 3,859

Here are some GeForce RTX 2080 Ti and GeForce RTX 2080 purchase links in the US :

  • EVGA GeForce RTX 2080 Ti FTW3 DT Gaming 11GB : $819.99
  • NVIDIA GeForce RTX 2080 Founders Edition : $954.00
  • ASUS GeForce RTX 2080 08G Dual Fan OC Edition : $948.99
  • EVGA GeForce RTX 2080 XC Ultra Gaming 8G : $949.00
  • GIGABYTE GeForce RTX 2080 Gaming OC 8GB : $919.99
  • GIGABYTE GeForce RTX 2080 WindForce OC 8GB : $789.99
  • ZOTAC GAMING GeForce RTX 2080 AMP 8GB : $839.99

Here are some GeForce RTX 2080 Ti and GeForce RTX 2080 purchase links in the US :

  • KFA2 GeForce RTX 2080 Ti OC Black : £1,703.85
  • ASUS GeForce RTX 2080 Dual O8G : £845.87
  • GIGABYTE GeForce RTX 2080 GAMING OC 8G : £984.61
  • GIGABYTE GeForce RTX 2080 WindForce OC 8G : £983.71
  • PNY GeForce RTX 2080 XLR8 OC Twin Fan : £916.53
  • ZOTAC Gaming GeForce RTX 2080 AMP 8GB : £645.51

 

Recommended Reading

[adrotate group=”2″]

Go Back To > Computer Hardware | Gaming | Home

 

Support Tech ARP!

If you like our work, you can help support our work by visiting our sponsors, participating in the Tech ARP Forums, or even donating to our fund. Any help you can render is greatly appreciated!


Leave a ReplyCancel reply