Hacker News new | ask | show | jobs
by Farfignoggen 644 days ago
The Majority of the gaming market's TAM is in mainstream GPUs and not Flagship GPUs! And all of Nvidia's Large Monolithic Graphics oriented GPU die Tape-Outs are for the Professional Graphics Workstation market first and foremost with some limited number of those Large GPU die samples not making the binning grade for Pro Workstation market usage and so that gets binned down for Flagship Consumer gaming!

And both AMD and Intel lack much of any larger Professional Graphics Workstation Market presence and the Pro Markups to justify that kind of investment for giant monolithic GPU Die Tape-Outs for Pro Graphics Workstations!

Take Nvidia's GP102 tape-out from the past as an example and that Giant tape-out had more Quadro Branded SKUs and only one consumer branded binning, the GTX 1080Ti! And it's the same for Nvidia's later generations where that's for Quadro/A-series branded Pro Graphics workstation GPUs where Nvidia "__102" Tape-Outs are mostly Quadro/A series(The A series branding has supplanted Quadro branding for Nvidia Pro Graphics Workstation GPUs).

1 comments

I don't know about that. It wasn't long ago that gaming dominated Nvidia's revenue. Professional non-datacenter GPUs were never really on top. No doubt in my mind the 1080ti, as it was basically a legendary gaming GPU, sold many multiple more units than all those Quadro SKUs combined. So I would say the truth is the opposite, the majority of the GP102s were binned as 1080ti with the exceptional few as Quadros.

Example: https://cdn.wccftech.com/wp-content/uploads/2019/11/NVIDIA-Q...

GP102 had more Quadro/Pro Branded SKUs/Binnings and one consumer SKU for the runt die samples that did not make the binning grade to get branded Quadro and so that became the 1080Ti. And the more than one 1080Ti variant was because that was VRAM capacity and memory clocks related for market segmentation reasons!

And then Nvidia created Volta for the Data Center and AI as that had the first generation Tensor cores in a GPU.

TU102 was the same and for Quadro at the high end with some consumer binning as well and even the top end TU104 was initially reserved for Quadro until the Turing generation cards were getting ready to be replaced by the Ampere cards. And the Top end TU104 binning eventually was released for gaming usage(under the Super or Ti Branding).

Nvidia's Pro Graphics Workstation market domination has given Nvidia the funding to create those Giant Monolithic Tape-Outs and Billions for the mask-sets for that every generation. And maybe the gaming revenues were large relative to non gaming at one time but not any longer and most all of Nvidia's later acquisitions of other companies were for the Data Center Market and not consumer/gaming!

Gamers have some sort of collective Myopia with regards to Nvidia's focus on gaming only and gamers! And Nvidia's/Tech Press's marketing focus helped establish the appearance of Nvidia as a gaming only company. But look at Jensen's Keynotes over the last many years and even at consumer/gaming focused events and Jensen's Keynotes were/are mostly AI/Enterprise and cloud services focused, much to the chagrin of gamers!

Do you have any references for this? Given the basic hacking on GPUs I was doing during that era, my memory leads me to believe the binning was pretty much in the opposite direction than as you describe.

I’m not a major gamer, so I don’t believe I have any myopia on this topic in that manner. As far as I can recall though, the GPUs with the heaviest overclocking capacity (including memory) were the flagship gaming GPUs and not the workstation Quadro based stuff. Volume was certainly in the gaming favor though and I don’t believe it to be even close. SKU count is more or less irrelevant.

My memory is certainly fallible and I was not as knee deep on the nvidia side during that era so I could very well be wrong. This goes against everything I remember from the firmware and overclocking side though. I don’t know why nvidia would have started locking down firmware so hard to keep the “pro” features locked into the workstation SKUs if it was an actual hardware binning situation vs. artificial crippling. This was right around the time that they started to really get into the datacenter space so it could be simple coincidence.

Go check how many second hand 1080 TIs are on the market vs all GP102 Quadros and Teslas. I think the myopia is in the other direction. Counting SKUs is not evidence of unit sales.
Even today the 4090, a card which everyone complains about on price, is likely to become just as legendary a gaming card in the 2020s as the 1080ti was in the 2010s, with 3-5 million sold at $1600+. Is that a lot compared to their H100 revenue? Not at all. Is it a lot compared to almost any other tech product? Yes absolutely. Even adjusted for inflation the 4090 will outsell the 1080ti. Nvidia isn’t going anywhere in the gaming market.