-
Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.
Are TFLOPS a valid number to compare systems?
- Thread starter lunanto
- Start date
OP
OP
No, they're a terrible metric for performance in games. Taken to the extreme a 1 TF GPU could outperform a 100 TF GPU.
A GPU is much more complicated than just clocks & CUs.
A GPU is much more complicated than just clocks & CUs.
Absolutely, when comparing hardware of the same generation.
According to Digital Foundry, let's say the new consoles officially come in at 10 tflops. That might not sound that impressive next to the X's 6tflops, but those 10tflops on the new machine are more capable than a comparable flop on the X. So in the other words, the difference will be more than the 4tflops might have you believe.
Now, when it comes to the next Xbox and PS5, that's a completely valid comparison point, because they're using comparable chips.
According to Digital Foundry, let's say the new consoles officially come in at 10 tflops. That might not sound that impressive next to the X's 6tflops, but those 10tflops on the new machine are more capable than a comparable flop on the X. So in the other words, the difference will be more than the 4tflops might have you believe.
Now, when it comes to the next Xbox and PS5, that's a completely valid comparison point, because they're using comparable chips.
When based on the same architecture and when talking strictly about shaders and compute performance, maybe, but generally no. Things like polygon counts, texture quality, physics, antialiasing, pixel fill rate, lighting to an extent, etc. have little to do with floating point operations.
Still no.
As an example, Nvidia TFLOPs don't translate to AMD TFLOPs (and vice versa).
It's a stupid fucking metric, almost as bad as when companies used "bits".
I answered under the assumption the OP was asking in regards to the next consoles, which are about to be announced very soon. In that case they'll be using chips from the same maker, and thus the comparison would be completely valid.
If performance per watt aligns closely, as it does within the same GPU family, why not. Even if that isn't the case, you'll have to keep the offset in mind . If architecture X with 11 TF reaches a minimum of 53 fps and an average of 87 fps in a game and architecture Y with 9 TF reaches a minimum 61 fps and an average of 98 fps in the same game you can compare them.
You just cannot compare them on paper, without real world testing.
You just cannot compare them on paper, without real world testing.
Even in that case, it's a stupid metric because TFLOPs can only be a valid point of comparison when everything else is identical
It doesn't give a complete picture but it's a decent enough ballpark. You just have to understand what it is beyond the back-of-the-box tagline. There's so much more to computing power though.
Only if the GPUs belong to the same architecture and there are no VRAM (Capacity, bandwitdth) constraints.
Yeah, ever since I started seeing people talk about TFLOPS this and TFLOPS that I just had my brain sent back to the old *-bit days and have no desire for that to continue being a thing. We had 1, maybe 2 good generations of not having this crap and now it's back for some reason.
if you isolate just FLOPS it's an indication but not really accurate representation
for comparing PS4 to XB1 is more relevant since they are so similar, but even then...
performance is limited by lots of different things, not just the floting point capability of the GPU, and even that is more of a theoretical number that is likely not realistic when running games.
for comparing PS4 to XB1 is more relevant since they are so similar, but even then...
performance is limited by lots of different things, not just the floting point capability of the GPU, and even that is more of a theoretical number that is likely not realistic when running games.
I obviously can't speak for everyone, because numbers seem to matter most to many. For me? Not one bit. Features is what matters most to me. For example, this generation the share button has become one of my favorite features. Same with suspend / resume.
All those little extras is how I do my comparisons. I honestly don't care if a game runs / looks a little better on one versus the other.
All those little extras is how I do my comparisons. I honestly don't care if a game runs / looks a little better on one versus the other.
No
A 1060 has less Tflops than the One X.
But overall more capable delivering higher resolution with higher fps.
A 1060 has less Tflops than the One X.
But overall more capable delivering higher resolution with higher fps.
Just when you compare GPUs of the same brand and the same generation/architecture.
Well it's obviously the way to measure graphical power, so it's very valid in that respect.
And of course these next consoles will likely have nearly identical CPU, and probably be rather comparable in most ways.
"flop" is short for "floating point operation per second" (i.e., calculations with approximate decimal values) so it's just the same thing as clock rate comparisons for CPUs which, there as here, doesn't really get into the way that different chips are engineered in ways that make some operations more efficient than others.
It's ARGUABLY a more applicable statistic than pure clock rate because it's not just measuring CPU cycles but the number of operations being done, but it still doesn't do a good job of reflecting real-world use cases for hardware.
Yes and no. Obviously this is referring to the PS5 and Scarlett when you're asking this question, so in this case the answer is yes.
But I also see people trying to compare the X and Lockhart in teraflops which is pointless since they're on two different architectures. It's as equally useless when trying to compare AMD to Nvidia because, again, different architectures.
But I also see people trying to compare the X and Lockhart in teraflops which is pointless since they're on two different architectures. It's as equally useless when trying to compare AMD to Nvidia because, again, different architectures.
It's a hard number that literally cannot be directly compared between manufacturers.
I agree with this and those saying no.
The shift from Pentium 1 thru 4 to Dual Core to Core 2 Duo made me realize numbers aren't everything.
Going from dual core to quad core also.
Not directly 1:1 no but a 500gflop system will never outperform a 2 tflop system no matter the architecture. The raw computational power is a completely valid metric
And things still work out when you aren't comparing systems from decades apart?