The Love of Money...

Nvidia's 50-series graphics card launch.

I recently decided to upgrade my office workstation. I also discovered I chose the worst possible time to do so. No, not the tariffs. It was the graphics cards. I’m not a hard-core gamer, but work and the sometimes bit of play require the horsepower on occasion. I’m a believer in going solid mid or upper mid end as I don’t need to climb the far side price-performance cliff to shave off every second, but I want something that doesn’t need to be replaced right away either, and the 2070 super I got five years ago is holding up pretty well.

Aside: these days, unless one has very specific needs or just really enjoys assembling things, the money saved on building your own workstation is more than made up for by the time spent. Even architects I’ve dealt with either buy a Dell precision or XPS or the equivalent, depending on their GPU needs and the computer role - don’t mess with Alienware, please - or they’ll go with a more boutique assembler. That said, be careful who you buy from even then.¹

So I started digging, and discovered that it’s going to be a while before it’s worth spending money on an upgraded GPU.

Since my card was released, two whole generations of NVIDIA cards have come out, and as it happens, we were on the cusp of a third, the 50 series, being released. In the past that would have meant I could get the “old” series, save a few bucks, and get a major improvement in my graphics performance, or just waited a bit for prices to settle down after launch. And the existing AMD/Radeon cards wouldn’t cut it for work outside of games. ²

Sadly, NVIDIA has demonstrated they have lost their way, both in terms of understanding all of their markets, but also technically. They no longer respect the people who put them on the map. In short, I don’t see myself buying the new generation, personally, or recommending one unless there is literally no other viable option to meet performance specs. Sadly, the latter includes a lot of Avid and Revit users.

It’s fair to ask - how have they dropped the ball? if you feel like a video recap:

It’s not really a new generation

The underlying processor is made using the same process technology. Improvements in architecture will get you some boost, but that also means that the cards won’t be significantly faster for raw rendering speeds. To be fair, there’s a whole new set of AI-related computation infrastructure, and there is a slight boost, and they managed to make them more power efficient, but…

Price

My 2070 super was under $600. Inflation is an issue and prices have crept up, but each jump of the 30 and 40 series cards resulted in a large uplift, such that the 4080 could easily beat the previous gen “90” series card, the 3090, and so on. Even with prices edging up, performance was going up more. Not so much here. A near-mythical reference/ MSRP/baseline 5070 ti is now a $750 card, and barely faster than the previous generation 4070 ti. Leaving aside supply issues and scalpers, most of the GPU sellers such as ASUS, Gigabyte, etc., are adding “features” to justify much higher price tags. Notably - the supposedly MSRP card to be sold for $750 sent to reviewers was showing up on websites at a $900 price point until reviewers complained, because they had done their reviews on the presumption that the price they were quoted was accurate.

Worse, the flagship card, the 5090, is now $2000 for the reference “founders edition” model, a 25% jump. My entire computer cost me less than that. It’s direct predecessor was $1600. The cheapest, the 5070 if you can find it at MSRP, is $550. Keep an mind that an entire PS5 Pro is $700 (and I find that ridiculous price for a console already), and the Xbox series X is $500, and I’m not getting either of those because they too have gotten sufficiently large and hot that I cannot put them where my previous Xbox is.

These prices and the underlying bloat of AAA titles are more and more driving gaming on the PC out of the budget of all but the wealthiest. Some games are requiring ray tracing to even run - and that has a huge impact on performance at any price level compared to similar graphics without ray tracing.

Performance

Again, there is a slight bump in efficiency and performance, but a number of reviewers have pointed out that compared to the previous generation, the 50-series raw render performance was less a whole new generation than a revamped “super”, or perhaps “super super” variant of the existing series.

So why was NVIDIA boasting about performance boosts?

Well, the new cards are loaded with processors optimized for AI workloads, so NVIDIA leaned into that, and decided that instead of making the card render more frames, it was going to use “AI” to generate in-between frames so that visually, the games appeared to be running at much higher frame rates, with hundreds of frames per second. Granted, even if I was younger I never was in the class of gamer to tell the difference in responsiveness between 40 frames a second and 60 frames a second, but there are plenty of people who can, and did not appreciate the disconnect between the visual smoothness and the responsiveness. All in all, the anger over these “fake frames” would likely be less if the cards performed better in general raster performance relative to the last generation, or the marketing didn’t try to sell it so hard, or the prices hadn’t gone up so much.

Perhaps the worst example of this is the current 5070. It was literally proclaimed as “the performance of a 4090 for $550” and compared apples to oranges by highlighting the “fake” generated frames against the 4090’s raw rendering performance. The problem was that this “performance” was entirely visual smoothness due to frame generation, and not because the card could redraw the actual scene fast enough.

Physics Engine Removed

To add insult to injury, NVIDIA removed a feature of their cards that helped in processing particle and other physics, resulting in a number of older games such as Borderlands2 actually performing worse on the latest and greatest more expensive generation. It may not have been viable for the company to support what turned out to be a little-used feature set, but having older games that should have no issues run notably and at times unplayably worse on much more expensive GPUs is not what’s called “a good look.”

QA and power design

The new generation of cards suffered from not one, but two design and manufacturing issues. First, roughly 0.5% or one in two hundred cards had an entire rendering pipeline set (ROP) non-functional, resulting in an 4% ³or worse performance drop. You heard that right, NVIDIA’s factory testing failed to catch a significant number of manufacturing defects that left the cards performing worse than the generation they replaced. Keep in mind every chip is tested.

Or they knew about it but didn’t care. Most gamers wouldn’t notice, right?

Also, power supply issues returned. Not having learned their lessons from fixing similar issues in the 40-series cards, NVIDIA didn’t design the power connectors on the board to properly accommodate the high power draw, resulting in melted cables and connectors. It doesn’t help that the cards are in some cases drawing significantly more power than they are specified to.

Supply

Graphics card launches are always subject to scalpers buying up supply and reselling them at a premium, but this is alleviated by having a solid supply. Sadly, NVIDIA didn’t have enough on hand to satisfy demand, even with the ridiculous price increases not backed up by a raise in value.

First, NVIDIA let the supply of 40-series cards flush out. This had the effect of making sure their were no cheaper “last-gen” competitors to their new product, as well as freeing up the production lines for the new generation cards because they used the same processes and lines.

Second, for one reason or another, there were not enough processors available for graphics cards at the time of launch - supply was so constrained that many reviewers have taken to calling it a paper launch, as even they had difficulty getting their hands on a retail version if they could even get a hold of one. And the launch was at the beginning of Chinese new year, so most of the factories were shut down as workers went home for the month, meaning that there would not be a supply coming in to catch up with demand any time soon.

So

In the end you have a “new” generation of cards that isn’t really, giving the appearance of improved performance without actually improving it much beyond “AI” tricks to make it looks smoother, while costing much more, and not being designed or built well - not only not working as advertised but either not working at all or damaging other computer components.

Why?

I believe that NVIDIA forgot who put them on the map, and that gamers are now a total afterthought that NVIDIA still wants to milk cash and excitement from.

Yes, it’s speculation, but it is informed speculation. AI sales to server farms are now the mainstay of NVIDIA’s bottom line. So much so that instead of spending time significantly improving 3D rendering times and performance, they improved the performance of “AI” related computations - and so the founder stands up on stage in his leather jacket and tells us how, since GPU’s helped form the backbone of AI, NVIDIA is comping around full circle and AI is now making graphics cards better. Oh, and that the 5070 has the performance of the 4090.

Given where they spent their design time, I think they had no choice but to lean on the AI features to provide interpolated frames. These AI server farms are voraciously sucking up chips and power, so the number of chips left over that don’t meet the standards of the server grade processors is lower, so that’s almost certainly a factor in the limited supply.

With supply of chips not needed for their AI business constrained, and their focus primarily on that, it’s no wonder that a significant percentage of chips ended up with problems, whether by oversight or lowered standards is unclear. It’s also little wonder that power supply issues already addressed once have resurfaced - the engineers are likely busy elsewhere on what makes the company money.

In the end, the gaming card market wasn’t enough of a focus anymore to ensure the launch wasn’t a total shitshow. NVIDIA is chasing the money, not concerned with actually making good cards for gamers. The cards gamers are getting are an afterthought of their AI business.

So yes, AI is ruining yet one more thing.

Meet Our Friends

The Last Redoubt is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

Please also join the Pilum Press Discord and/or the Autarch Discord. Pilum is the publisher of several books and short story collections including Shagduk and Thune’s Vision. Autarch is the home of the Adventurer Conquerer King RPG.

1

2. The 9000 series cards just came out as I’m writing this, do much better at video streaming, production, and image processing tasks than past generations of AMD cards if not quite up to NVIDIA yet, but the supply situation is still constrained.

3. I originally put 11%. Note - NVIDIA claims 4, others have seen higher drops in performance, but I cannot remember where I got the 11% from. That said, even 4% raster output (actual frame generation performance) drags the performance of these cards down to that of the previous gen anyway.