NVIDIA in the comfort zone with carrots and sticks – Three “new” graphics cards, the background, the 3. Newton’s Law and the GPP | Editorial

Currently, everything that has a connector sells like hot cakes, but as a manufacturer you have to be prepared of course and therefore also look to the left (AMD) and right (Intel). Always according to the “what happens then” principle, which we also know from physics. Does anyone remember actio = reactio? That’s right, good old Newton’s Third Law! So what does this have to do with NVIDIA? Team Green has, after various teething problems, made a good showing with the Ampere generation and is now simply letting the other two manufacturers come. Sit back and watch.

And yes, it’s a comfort zone of a different kind, because to respond to competitors’ responses in the form of new products, you just have to flip a few simple levers and, if need be, use additional physical leverage with your own board partners to keep the feeding bowl as monochromatic as possible. The GPP (GeForce Partner Program) as a friendship proof of eternal loyalty has outwardly visibly served its time, but certain residues cannot be removed so easily. But more on that in a moment.

While Twitter was recently buzzing about the total upsurge of the deep green portfolio, industry partners were already smiling mildly at the time. Theoretical business games yes, effort for nothing of course no. And so, in the meantime, the realization has probably also prevailed on Twitter that, in practice, shots are only fired when there is no other way. Saving bullits is the order of the day. Three should be enough. You think. Of course, I have also dealt with this topic in the last few weeks and tapped various sources, but in the end I could only register a shrug of the shoulders. Which brings us back to the comfort zone and what the board partners can live with quite well. Because even if something “new” is added, the old resources can still be used conveniently. 

GeForce RTX 3090 Super

Let’s start with the supposedly all-important performance crown. The GeForce RTX 3090 is fast and has significantly more memory than its direct competitor in the form of the Radeon RX 6900XT(X). The performance on the other hand, if you look at the computing power and the raster graphics, is not quite so clear (depending on the application). If AMD manages to shrink Navi 21 and maybe even conjure up a little more fire under the ass, then they simply take the GA102 full expansion of the RTX A6000 with 10,752 instead of 10,496 shaders and unleash it on the astonished clientele. Half the RAM compared to the pro card, but with 24GB of GDDR6X still faster for those who don’t need to take a full bath in memory expansion.

The effort is minimal for everyone involved, as the yield should have increased by now and no one even has to change anything. Same PCB including assembly, only the chip changes inside. The additional expense here is zero, if you exclude packaging and marketing. The board partners (AIC) do not even have to change the forecast, because everything is already planned. Thus, the rumored GeForce RTX 3090 tends to be a true replacement, completely replacing the current RTX 3090. It wouldn’t make sense to have them side by side anyway.

The timing and realization here are entirely up to AMD, and the likelihood of a boost increases to the extent that AMD manages to spruce up the RX 6900XT as a refresh. And the production changeover only takes as long as it takes to insert the new belt with the different chip, so a whopping 10 minutes should be more than enough.

GeForce RTX 3070 Super

Now Team Blue comes into play. Of course, with only 8 gigabytes you can’t do anything with NVIDIA, Master Jensen doesn’t even have to count the sugar cubes in his morning coffee. Since they peg the performance of the biggest upcoming ARC card at a GeForce RTX 3070 Ti (and Jensen’s spies have actually always been right), all you have to do is adjust the memory accordingly. Since the required 2GB modules are only available as GDDR6 without X and you don’t want to put on extra boards, you just continue to use the full version of the GA104 and only change the RAM. Here, too, only one belt is exchanged, on which the modules for SMD assembly snuggle comfortably together. Ten minutes, if that.

So again, the technical effort of AIC is exactly zero and you don’t need a pre-run either, because the GPU and memory are supplied by NVIDIA as a kit anyway. This would automatically make the RTX 3070 Ti obsolete, while the RTX 3070 could still be argued over. For the Forecast, it’s still the same boards and components as the RTX 3070 Ti, and the super card will surely be released just before Intel’s real launch. So the probability is very high, just-in-time anyway. And while we’re on the subject of Intel, another footnote may be of interest.

Intel will not be able to produce its own cards exclusively and alone, you need one or the other board partners. I don’t want to write about the pros and cons of being an exclusive board partner again, we’ve been through that before. Whereby the framework conditions for the respective partners (AIC at NVIDIA, AIB at AMD) will hardly change. However, the friendship shackles rings in the form of the GPP, which has meanwhile been cancelled again, are still in everyone’s memory and so both NVIDIA and AMD will already make sure that there will be no lateral jumps. To what extent contractual or monetary levers are applied here is decided, as always, by the carpet floor of the GPU manufacturers.

However, I wouldn’t be surprised if, in addition to the mixed producers Asus, Gigabyte and MSI, which are considered to be set, new companies would suddenly appear in the near future, whose sorting into existing company groups could be quite interesting. Giants, such as PC Partner, already have a rather elegant corporate construct that leaves all the loopholes open. While NVIDIA-exclusive brands like Zotac, Manli or inno3D are directly included, Sapphire is a completely independent company in its own company complex, which also indirectly acts as a contract manufacturer for AMD’s reference cards.

There one could install with security something similarly stored also with and/or for Intel. Just as a thought experiment…. And then there’s ASRock, who are currently AMD-exclusive, but shouldn’t really be happy when they look at their direct competitors and their combination options. Intel will have to solve all that, and Jensen will surely unpack various treats or even a sour drop or two to keep the friends nicely on track. Together with Lisa, of course, because it’s certainly more fun as a couple.

GeForce RTX 2060 12 GB

Finally, here we have the upcoming VW Beetle in Jensen’s VGA fleet. And runs and runs (and runs)…. But you’ll probably need Thor’s hammer to finally bury this thing at some point. Because a GeForce RTX 3050 Ti, as it was originally rumored, would probably have to beat Navi24 (which is still coming) gallantly. Raytracing hardly plays a role with these tiny devices anyway, then rather the memory. And because six isn’t sexy, Jensen’s gonna give you a slap on the wrist. RAM glut instead of innovation, it certainly pays off somehow.

With the old Turing card, Jensen has a real undead in the race that will be hard to bury by simple means as long as the current shortage situation continues. Not even garlic or Van Helsing will help against this zombie.