Teraflops are NOT a Measurement of Gaming Performance

[Pemalite]

JRPGfan said:

If theres no drastic bottlenecks to them, and their the same architecture in core design, you CAN absolutely use Teraflops as related to gameing performance.

No you can't.
Because...
It doesn't account for geometry performance...
It doesn't account for 8-bit integer performance...
It doesn't account for 16-bit integer performance...
It doesn't account for 24-bit integer performance...
It doesn't account for 32-bit integer performance...
It doesn't account for 64-bit integer performance...
It doesn't account for 16-bit floating point performance...
It doesn't account for 64-bit foating point performance...
It doesn't account for 8-bit floating point performance...
It doesn't account for fillrate performance...
It doesn't account for Ray Tracing performance...

It doesn't account for a whole lot of a GPU.

If you somehow disagree with that... And still think Teraflops is relevant... Then YOU need to explain why. And I mean "why" with appropriate technical detail... Otherwise your position on this is incorrect and thus illogical. (Then again, I have lost count how many times we have been over this anyway. But I digress.)

JRPGfan said:

Also Architecture to Architecture, you can.
If you take 2 cards that are both GNC, you can compaire them.
Such as the case with the Playstation 4 + Xbox One X.

No you absolutely can't. It will never be Apples to Apples that way.

Even when we use the identical architecture, there is significant deviation.

I.E. With Graphics Core Next AMD doubled the amount of geometry processors in Graphics Core Next 1.2/Hawaii, Hawaii and Tonga also have a geometry front end that was twice as wide as Tahiti's.

What this means is that in heavy geometry scenarios such as games leveraging a significant amount of Tessellation or Polygons, it's going to be 70% faster at the same "Teraflops".

The other optimization that happened is on the opposite side of the rendering pipeline in the Render Output Pipelines... Namely Delta Colour Compression, which meant AMD could reduce the amount of bandwidth by 27% but still hit the same levels of performance... So in an Apples to Apples world, GCN 1.2 and 1.3 parts would have more bandwidth.

The Playstation 4 Pro and Xbox One X have some significant deviations in hardware, which will always make them directly non-comparable... The Xbox One X has the crossbar memory controller, it also has the command processor which offloads some CPU tasks.

SvennoJ said:

Ah yeah the memory speed war. Adding ESRAM speeds to DDR3 speed at the start of last gen to make it look closer to PS4.

However if everything else is the same it is a use-able measure. If rumors are true

Except...
We don't know if everything is the same.

And even in the Playstation 4 Pro vs Xbox One X's case... The Xbox One X is non comparable, despite the fact it's GCN... It's still using GCN that is substantially more modern and thus more capable... And also had some semi-custom touches to bolster performance.

SvennoJ said:

Series X is 3x Lockhart and 1.3x PS5
PS5 is 2.3x Lockhart (about the same difference as ps4 pro to base ps4)

Of course XBox One had other benefits compared to ps4, as well as the ps4 pro had to ps4.
It will be a weird start of a new gen if all these rumors are true.

XBox One X was 4.2x XBox One with other improvements as well. And that had games running in native 4K (or very close) on XBox One X while the XBox One was struggling below 1080p with other things turned down as well. So maybe 3x difference is enough for 1080p vs 4K. PS4 pro couldn't get past 1440p vs 1080p ps4 with 2.3x increase (plus boosted cpu)

It's a shame no matter how you look at it. The pro machines were just used for higher resolution, which Lockhart seems to pass on to the new generation with its existence.

Changes in resolution changes bottlenecks in the underlying hardware.

Developers were sacrificing visual fidelity chasing the 4k dream.

JRPGfan said:

Playstation 5 + Xbox Series X, are BOTH confirmed to be useing RDNA 2.

Its about 50% more effecient in terms of Performance/watt, than a RDNA1 card is, according to AMD.

This means, that even if Playstation 5, uses less power on the GPU than a current 5700XT card does, it could beat it in performance.

I say wait before we count our chickens. We don't know how such a GPU would perform in say... Ray Tracing which is the big next-gen feature.

JRPGfan said:

Huh?

They will be Ryzen 2, 8core.
Also their rumored to run like 3,2Ghz....

Thats not "low end"

Ryzen is currently beating Intel at Mhz vs Mhz right?
This means that if you take a equal cpu from intel, thats 8cores, and also clocked at 3,2ghz it would be slower.

Not everyone buys a Intel -K part, and overclocks to 5ghz.

This is "quite" a beefy CPU imo, esp compaired to the slow jaguar cores that were inside the current Playstation 4 / Xbox One X.

It's mid range.
Ryzen 8-cores will start to push in the low-end probably with Zen 4 at 5nm... AMD is already throwing 8 Zen cores into notebooks now.

sales2099 said:

Because now everybody is talking about how teraflops don’t necessarily matter instead of how Xbox is weaker like in 2013. I say that’s a PR victory in itself and addresses the people that talk the loudest on the internet. 

Not everybody. A few in this thread are still clinging to the hopes and dreams that Teraflops do matter... Examples above for instance.

victor83fernandes said:

With the launch of the ps4 and xbox One the talk was resolution, 1080p vs 900, or 900vs 760, teraflops were very rarely talked at all.

They were talked about on this forum rather heavily.

Last edited by Pemalite - on 09 March 2020