• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Poll: ** The AMD VEGA Thread **

On or off the hype train?

  • (off) Train has derailed

    Votes: 207 39.2%
  • (on) Overcrowding, standing room only

    Votes: 100 18.9%
  • (never ever got on) Chinese escalator

    Votes: 221 41.9%

  • Total voters
    528
Status
Not open for further replies.
Best not to believe in fairy tales, tbh ;)

I can't see how it makes any sense to have 2x 1080 perf in your product, but deliberately hold it back.

I can understand making a product that's not as good as you can theoretically make it, so you have another product you can release later which is better.

To to make the absolute best product you can, then deliberately reign in its performance so it's only 50% as good as it should be... that doesn't make sense. To put it this way: it's like building a Ferrari with a V8 engine (I'm not a car person btw :p), then drilling some holes in the radiator and the engine block, putting a load of contaminants in the fuel, and leaving the hand-brake on a little.

You could have made a small hatchback much cheaper, but instead you made a sports car, only to intentional sabotage its performance. It doesn't make sense. You still spent a lot of money making that sports car, when you could have saved a ton making a small hatchback.

It's not how people do things in the real world, because it costs you too much money.

Just going to throw this idea in the ring. What if it isn't intentional, what if it is simply a lack of resources/time/ability in the driver team/necessary personnel.
 
Seen this posted on hardocp, if true what is the major difference that results in amd on paper being more powerful but not in practice. I can even remember back to the radoen 8500, that was meant to be a beast of a card but fell behind the competition.


1080TI - tflops = 11.3
Vega - tflops = 12.6

980TI - tflops = 5.63
Fury X - tflops = 8.7

780TI - tflops = 5.04
290x - tflops = 5.6

680 - tflops = 3.1
HD7970 - tflops = 3.8
 
Seen this posted on hardocp, if true what is the major difference that results in amd on paper being more powerful but not in practice. I can even remember back to the radoen 8500, that was meant to be a beast of a card but fell behind the competition.


1080TI - tflops = 11.3
Vega - tflops = 12.6

980TI - tflops = 5.63
Fury X - tflops = 8.7

780TI - tflops = 5.04
290x - tflops = 5.6

680 - tflops = 3.1
HD7970 - tflops = 3.8



Efficiency, bottle-necks, load balancing, drivers. AMD try and brute force performance, Nvidia architectures try to be more efficient and intelligent to maximize resources better. For example, Nvidia have been using more advanced color compression for longer than AMD, whoch means they can use less bandwidth. AMD cards have been geometry limited in many games recently. And then there is th load balancing issues, AMD has fund it very difficult to get their PGUs runnign at 100% load, perhpas form geometry limits but also because of their scheduling hardware. This is why they have been trying to force Async compute down everyone's throats in order to try and make use of idly compute resources. Nvidia driver and hardware stack keep Nvidia GPU under better utilization, which in turn means async compute is less valuable . E.g. an AMD GPU might run at 70% utilization, adding Async compute tasks might give you 85-90%. Nvidia GPUs sit at 90% already, async compute might move that to 94% or some such. There is just less to gain for nvidia because their GPUs are already working harder.

There are also differences in the actual real-world tflop performance related to the clock speed. nvida's numbers are under-estimated because they refer to the guaranteed minimum boost clock, but basically every nvidia GPU will clock higher than that. AMD uses the max possible boost clock but the GPUs often struggle hto hit that for sustained periods.
 
Seen this posted on hardocp, if true what is the major difference that results in amd on paper being more powerful but not in practice. I can even remember back to the radoen 8500, that was meant to be a beast of a card but fell behind the competition.


1080TI - tflops = 11.3
Vega - tflops = 12.6

980TI - tflops = 5.63
Fury X - tflops = 8.7

780TI - tflops = 5.04
290x - tflops = 5.6

680 - tflops = 3.1
HD7970 - tflops = 3.8

IIRC (correct me if I'm wrong) the Nvidia TFLOPS are calculated using base clock not the actual boost clock. While it would still look bad for AMD in comparison, it is not as large a gap is these numbers suggest. As an example a stock 980Ti would sit around 1160-1200MHz on the core depending upon the model.
 
Last edited:
Seen this posted on hardocp, if true what is the major difference that results in amd on paper being more powerful but not in practice. I can even remember back to the radoen 8500, that was meant to be a beast of a card but fell behind the competition.


1080TI - tflops = 11.3
Vega - tflops = 12.6

980TI - tflops = 5.63
Fury X - tflops = 8.7

780TI - tflops = 5.04
290x - tflops = 5.6

680 - tflops = 3.1
HD7970 - tflops = 3.8

Flops is a measure of compute performance and AMD GPU's are always shader heavy that's why they're generally better for compute. NVidia focus more on pixel throughput, geometry, proper tessellation which benefits games more.
 
IIRC (correct me if I'm wrong) the Nvidia TFLOPS are calculated using base clock not the actual boost clock. While it would still look bad for AMD in comparison, it is not as large a gap is these numbers suggest. As an example a stock 980Ti would sit around 1160-1200MHz on the core depending upon the model.

Yeah in this case it is calculated on its base clock - a good boosting 980ti would be clocking in at 7.8TF a decent overclocker might even hit 8TF.

Looks like the 780ti there is based on its on paper boost clock - in actual use would be around 6TF.
 
Flops is a measure of compute performance and AMD GPU's are always shader heavy that's why they're generally better for compute. NVidia focus more on pixel throughput, geometry, proper tessellation which benefits games more.


So called "proper" tessellation is pointless when it makes no visual difference, you could put side by side images of tessellation methods and past a certain point there's no visual difference, only frame rate tanking which is the point of overdoing it.
 
Seen this posted on hardocp, if true what is the major difference that results in amd on paper being more powerful but not in practice. I can even remember back to the radoen 8500, that was meant to be a beast of a card but fell behind the competition.


1080TI - tflops = 11.3
Vega - tflops = 12.6

980TI - tflops = 5.63
Fury X - tflops = 8.7

780TI - tflops = 5.04
290x - tflops = 5.6

680 - tflops = 3.1
HD7970 - tflops = 3.8

Vega needs to have a much better hardware scheduler and much improved shader efficiency (i.e less idle time) over earlier revisions of GCN, they made start with Polaris but there's a long way to go if AMD wants to bring the tflop advantage to bear on Nvidia. But raw compute performance has it's advantages, AMD are making a massive push into the pro market with it's Radeon Instinct MI25 ready for working weather patterns, AI and financial modeling. The days of Nvidia and Intel having it all there own way in that market are all but over.
 
My 1080ti overclocked to 2075Mhz is 14.87 Tflops.
my 980ti overclocked to 1550mhz is 8.7Tflops.
most 980ti can do 1450 easily enough. This puts the 980ti over 8Tflops no problem. and most 1080ti can do 2ghz no problem. So thats around 14.5 Tflops so looking at it this way nvidia leave a lot in the tank and under rate their performance on paper.
That puts it into perspective more.

also any 1070 overclocked to 2.2ghz (Not many hit this) is only 8.4Tflops so my 980ti is faster than any 1070 like i was saying repeatedly but people kept telling me 1070 is faster unless nvidia purposely gimp my performance.
 
Last edited:
also any 1070 overclocked to 2.2ghz (Not many hit this) is only 8.4Tflops so my 980ti is faster than any 1070 like i was saying repeatedly but people kept telling me 1070 is faster unless nvidia purposely gimp my performance.

It doesn't directly equate to ingame performance - architecture, etc. will have some impact a Kepler and Maxwell card with the same on paper TFlops will often see the Kepler card struggle more in shader heavy games before the Maxwell cards while if a game uses a lot of primitive objects the Maxwell cards will often struggle before the Kepler ones, etc.

I had a borrowed 980ti shortly before my 1070 and while I didn't go nuts on overclocking it as it wasn't mine the comparisons at least in what I play was interesting - the 980ti was often managing good max framerates and similar averages but some scenes where the performance drops out a bit it would fall down quite a bit compared to the 1070.
 
Status
Not open for further replies.
Back
Top Bottom