Who will provide the NX GPU?

[dongo8]

JEMC said:

dongo8 said:

I am no computer expert, but as far as I have heard and read FLOPS are FLOPS. All of the other stuff doesn't matter, because FLOPS are literally processes performed per second, therefor a lot of the things listed, would ALREADY be taken into account.

The problem with Flops, and the reason why Pemalite always tries to warn people to not make that mistake, is that they are only valid for one of the tasks a GPU can do, and gaming is not that task.

Let's put an example that explains what he's talking about:

(data taken from The Tech Report)

• AMD R9 380: 3.4 TFlops
• AMD R9 390: 5.1 TFlops
• AMD RX 480: 5.8 TFlops
• Nvidia GTX 970: 3.4 TFlops
• Nvidia GTX 1060: 3.8 TFlops
• Nvidia GTX 1070: 5.7 TFlops

If we only look the TFlops, then the performance of those cards, from slower to faster would be

R9 380 = GTX 970 < GTX 1060 < R9 390 < GTX 1070 < RX 480

Well, let's look at the relative performance of those cards tested with 16 games:

(graph taken from The Tech Report review of the GTX 1060, that's why this card is the base card at 100%.)

Surprise!

The GTX 970 is not only much faster than the equal (in Flops) 380, but it's almost on par with the 390 and 480. The 1060 is around 10% faster than the 480 despite having 2 full TeraFlops less, and the 1070 simply destroys them all.

That's the reason why TFlops are not a good way to compare performance in videoconsoles, even less is they are from two different vendors and from different generations too.

TheLastStarFighter said:

Yeah, of course, I'm just speculating on the 1+1+3+3 thingy.

I'm not sure if they will do the suplimental device thing, but it does make a tonne of sense if they can do it right.  It would be mostly new for consoles, but PC's use multipe GPUs all the time, and Alienware offers a supplimental GPU for their laptops now.

So, if done right, Nintendo could offer a portable system for the bulk of their customers, the "3DS audience".  They could offer a second SKU that includes the portable and the SCU docking station for home gaming.  Let's just throw out price points of $299 and $399 to start.  The "home" SKU could include the power-boosting docking station and pro-style controller.  This console setup could be appealing to those that want the home Nintendo experience, or the "Wii U" audience.

If you have both the NX and its dock with say an 800 GFLOP GPU unit, that unit alone would be screaching performance for a portable.  Combined 2X, the home unit would be acceptable levels for current consoles.

The goal here from a company perspective, would be that you sell a tonne of the portable NX and get real good 3rd party support and have extensive 1st party support focused on the single system.  SInce a kid can then dock the system to their TV and up performance and play their favorite titles, why bother buying another home console?  This appealing setup could then lead to more third party support, which leads to more sales, which leads to the system being more in demand, and it's a continous positive cycle.

PCs have it, but they need special drivers and they don't scale perfectly. And we're talking about a secondary GPU, because if that powered dock has another Tegra to work in tandem, that's another CPU thrown in the mix...

Also, in your example, if they make two SKUs with the NX and then dock+controller, why can't they make the dock+controller a console by itself? Why does it need the NX? It's a wasted oportunity.

Makes sense, but again I think that he may have jumped on the subject way too negatively, and down Starfighter's throat way too quickly. Starfighter said essentially the same thing that I did, (it's all theoretical until put into practice, of course) which is that FLOPS are FLOPS. Meaning if it has the power to process at say 4 TFLOPS than that is the power it has to process at. I wasn't trying to put it into a full test machine and see what the ACTUAL output was, everyone knows that will vary depending on where those are allocated and what it is paired with. The point is he jumped on Starfighter because he made an analytical statement without taking into account the other components and the possible things these FLOPS could be attributed to. Well that's not what we were doing, we were simply stating that at X number of FLOPS, it SHOULD be able to perform at a high level.

I do appreciate the side by side comparisons and the attempted explanation of what he was getting at though. I already had thought about that though, and it makes sense, my statement remains the same, FLOPS are FLOPS. The allocation is going to be different for sure, but according to NVidia and the outlets that I have read, the PASCAL architecture is pretty special and should keep them at a pretty high number due to its efficiency. It's a wait and see game, unfortunately no one is working with absolutes. Thank you again though! : )