The Philosophy of ARC GPU from Intel.

[FuuMasta]

So I was thinking about the best possible route that Intel could take to get a massive market share in the GPU market.

 

The truth is that Intel will probably never enter the top of the line GPU market with performance on the level of the 4080/4090 or their equivalent every gen anyway BUT...

Intel could chose to become the reasonably priced GPU manufacturer that absolutely FOCUS on efficiency and performance per watt. (FPS per Watt and GFlops per Watt)

 

We have to get real at some point: GPU are more powerful every generation but they also use more power.

RTX 2070 --------- 175W -------- 6,497 GFlops ---------------- 37.13 GFlops/W

RTX 2080 --------- 215W -------- 8,920 GFlops ---------------- 41.49 GFlops/W

RTX 3070 --------- 220W -------- 17, 664 GFlops ------------- 80.29 GFlops/W

RTX 3080 --------- 320W -------- 25,068 GFlops -------------- 78.34 GFlops/W

RTX 3090 --------- 350W -------- 29,284 GFlops -------------- 83.67 GFlops/W

RTX 4080 --------- 320W -------- 42,998 GFlops -------------- 134.37 GFlops/W

RTX 4090 --------- 450W -------- 73,073 GFlops -------------- 162.38 GFlops/W

 

This is unsustainable: Electric cost and scarcity is a problem that grow every year and we can now see with the 40 Series how bad this path is hardware-wise.

So yes, I think that Intel should take that path and I hope that if they have not took that decision yet that this thread will become popular enough to catch their attention.

 

Sadly, at the moment they seem to be behind what I was hoping for:

A770 16GB ------ 225W -------- 17,203 GFlops --------------- 76.46 GFlops/W

 

Somehow, I hope that this idea gain in popularity and this is the way that Intel choose to go.

[IkeaGnome]

Besides the 3080 your chart has Nvidia consistently getting more efficient with newer architectures and higher up on the sku list within the same architecture though. Looks like Nvidia is doing what you’re wanting Intel to do?

Thats the natural progression of things. Intel will have to do that. Same power same performance as last gen won’t sell. Same performance more power won’t sell. More performance same power and more performance with slightly more power will sell. 
 

If power is a big worry, look at getting a more power hungry card and under volting. 

[FuuMasta]

4 minutes ago, IkeaGnome said:

Looks like Nvidia is doing what you’re wanting Intel to do?

2 Things:

-I'm calling for Nvidia to stop the rise of power usage, for the sake of the environment.

-I'm blaming Intel about not being able to beat the efficiency of the 3080, a +2 years old GPU.

[porina]

5 minutes ago, FuuMasta said:

-I'm calling for Nvidia to stop the rise of power usage, for the sake of the environment.

You don't have to buy the top model. As long as efficiency continues to increase, you can pick lower models and still see improvements gen on gen.

 

5 minutes ago, FuuMasta said:

-I'm blaming Intel about not being able to beat the efficiency of the 3080, a +2 years old GPU.

FLOPS might be useful as an indicator of compute performance, but it definitely isn't good as a gaming performance indicator cross architectures/generations.

 

I'd also be more interested in architecture efficiency than product efficiency, since products can be chosen to sit at any point on the efficiency curve, it is more interesting to compare the curves.

 

The hardware was "done" about a year ago. The extra wait was mostly to get software in a better state. Give them a couple of gens and I think they'll have a good go at knocking AMD off 2nd spot. This takes time.

[IkeaGnome]

16 minutes ago, FuuMasta said:

-I'm calling for Nvidia to stop the rise of power usage, for the sake of the environment.

Blame gamers and everyone else who wants more performance every generation. Many generations of screaming for more performance while not giving time for efficiency to catch up to those performance jumps is on you, me, and everyone else.

17 minutes ago, FuuMasta said:

-I'm blaming Intel about not being able to beat the efficiency of the 3080, a +2 years old GPU.

Cards have been "out" for a while now. Drivers kept them from being released. I think you were having too high of expectations for the first round of cards for them. I wouldn't expect them to be competitive until later on this decade.

[FuuMasta]

4 hours ago, IkeaGnome said:

Blame gamers and everyone else who wants more performance every generation. Many generations of screaming for more performance while not giving time for efficiency to catch up to those performance jumps is on you, me, and everyone else.

The other thing that seriouly bug me is how mobile hardware is always a shit-ton lot more efficient.

If we compare a 3080 to a 3080m, the desktop use 2.5x to 3x more power on both the GPU and the CPU than the laptop with a 3080m.

However, you only get a 30-40% increase in FPS.

How can this be acceptable or possible to begin with? I have no idea.

 

[solado]

2 minutes ago, FuuMasta said:

How can this be acceptable or possible to begin with? I have no idea.