By using this site, you agree to our Privacy Policy and our Terms of Use. Close
HollyGamer on 14 December 2019
Edit Thread
Pemalite said:
HollyGamer said:

I agree using pure teraflop metrix is false, but if you have other GPU that prove the other way around then it can be used as benchmark as well, RX 5500 is bad GPU example. You can use RX 5700 or 5700 Xt as example tho. 

You can check this video tho

No. You can't. It's inaccurate... Which my Radeon RX 5500 example proves without a doubt.
Not only is the Radeon RX 5500 a newer and more efficient architecture... It also has more flops than the Radeon RX 590. - Yet cannot decisively out benchmark it.

Ergo. Flops is useless.

Another example is the Geforce 1030... The DDR4 variant with a 250mhz core overclock has 1.076Teraflops of performance (1402Mhz core * 2 Instructions per clock * 384 Cuda cores) - So it should decisively beat the GDDR5 variant which is 0.942 Teraflops, right? Wrong.

Even though they are based on the same GPU, the GDDR5 wins hands down, almost doubling in some instances.
https://www.techspot.com/review/1658-geforce-gt-1030-abomination/

Or lets take the Radeon 5870 and 7850? The Radeon 5870 has 2.72 Teraflops or performance verses the 7850's 1.76Teraflops, you would assume that the Radeon 5870 with an almost extra 1 Teraflop of performance would be the easy winner, right? Again. Wrong.
The 7850 is the clear victor. - https://www.anandtech.com/bench/product/1062?vs=1076

Same thing with the Radeon 7750. The DDR3 vs GDDR5 variants, same amount of Gflops, the DDR3 version even has twice the amount of Ram.
Yet the GDDR5 is the clear victor.
https://www.goldfries.com/computing/gddr3-vs-gddr5-graphic-card-comparison-see-the-difference-with-the-amd-radeon-hd-7750/

So basically in dot point form...

* Newer GPU with more flops = Doesn't mean it's faster.
* Same GPU's, one with more flops. = Doesn't mean it's faster.
* Same GPU with identical flops. = Doesn't mean it's faster.

So on what planet can we even assume that Flops has any kind of relevancy or accuracy when every example blows that assumption out of the water?

Please. Do tell. I am all ears.

I think Digital Foundry has answer your theory flawlessly , and also tell me what metrix you suggest