It is June 2023, and gamers are still waiting for NVIDIA or AMD to release a GPU that is a no-brainer for any of the different price points of the mainstream market. I accept that the RTX 4090 and the RX 7900 XTX are halo products with an accordingly high price tag. Especially the RTX 4090 is impressive in all aspects. But what about the people who do not have over 1000 Orens to burn or do not want to spend that much? The controversy and discussions about NVIDIA’s sub-1k lineup down to the RTX 4060 Ti got me thinking. These cards would be a significant or even gigantic performance upgrade for me, yet I am not interested in them. What about AMD? Seriously, what about AMD? When are they moving their butts and joining the fray?

I am aware that it is about maintaining margins and keeping all their Scrooge McDucks happy. I also know that gaming products do not sell for as much as professional workstation- or enterprise products, with AI being the new hotness.

(Let’s hope it turns out to be more useful compared to the mining energy waste.)

How much time did NVIDIA dedicate to gaming hardware at Computex 2023? But is this really worth antagonizing a vast and vocal audience that has enabled AMD and NVIDIA to get where they are? From the looks of it: Yes. When considering a younger Jensen Huang’s statements, NVIDIA’s position is especially curious.

The elephant in the room is undoubtedly graphics card prices, right? Not exclusively. While I disagree with the GPU market as it stands right now, it is not solely because new products are expensive. It is about bang for the buck. Cream-of-the-crop GPUs like the RTX 4090 and RX 7900 XTX can gladly exist where they are to serve as a show-off statement. It’s why brands like Volkswagen build a Golf GTI or Golf R, and BMW has the M series of performance vehicles. It’s for the enthusiasts. However, everything else in NVIDIA’s lineup is flawed and imbalanced in one specific way. I will mainly talk about NVIDIA, but only because their current-gen product stack is more fleshed out than AMD’s. That does not mean I think AMD is doing a good job. Where are the GPUs between the currently lowest-end RX 7600 and the flagship RX 7900 series?

So, what is this imbalance that I mentioned? If you follow the hardware news or are even an enthusiast, the latest and greatest games can put the memory systems of current mid-range GPUs through an extensive workout. Hardware Unboxed released videos demonstrating how graphics cards with 8 GB of VRAM can have serious issues today – and, therefore, in the future (16GB vs. 8GB VRAM: Radeon RX 6800 vs. GeForce RTX 3070, 2023 Revisit, Nvidia’s 16GB RTX 3070… Sort Of: 16GB A4000 vs 8GB RTX 3070).

Let me give you an analogy from the world of Kubernetes. When you deploy applications into a Kubernetes cluster, you can specify their resource requirements for CPU and RAM. Kubernetes has this concept of compressible and non-compressible resources. If the application uses more of a non-compressible resource, it is simply killed and restarted. Nothing is left of that particular resource that Kubernetes could assign to the application. Compressible resources, on the other hand, only result in throttling and slower performance in the worst case. Can you guess which is RAM and CPU, and where I am going with this? If the chip is not fast enough to run, say 4K, you lower the resolution, enable upscaling, or turn down settings heavy on compute. However, you are out of luck if the graphics card runs out of memory. The game might not necessarily crash, but its performance may be wildly unpredictable or the textures do not load correctly and everything looks flat and muddy. In short: the amount of video memory is essential.

It is just one part of the memory subsystem, however. Another part is the bus width and the resulting memory bandwidth. It is the data highway that connects the GPU to the VRAM. If you frequently drive on roads with varying lanes in one direction, you know that a “wider” highway (or Autobahn, for my fellow Germans) usually allows for more cars to pass through in less time. And finally, there is the number of PCIe lanes connecting the graphics card to the system. It is the same concept, only this time, it affects the whole graphics card’s access to the rest of the system. The best worst example of what could happen in a memory-limited scenario was the RX 6500 XT with 4 GB of memory. Hardware Unboxed released an interesting video over a year ago investigating the impact of PCIe bandwidth in such a situation.

To generalize: The combination of a low amount of video memory and a tiny memory bus rubs me the wrong way.

Consider the RTX 4070 Ti. It has a reasonably powerful chip capable of 4K gaming, but it is paired with only 12 GB of memory over a 192-bit bus. Shortly, that might not always be enough to enjoy a smooth framerate at higher resolutions. Memory-sensitive games may suffocate on the small bandwidth like a Toyota Prius would thwart a BMW M3 on a densely populated two-lane highway. It is worth noting that the RTX 3060 has the same general memory subsystem – albeit with slower VRAM. How does that make sense? As such, I consider the RTX 4070 Ti a flawed product costing a lot of money. The same logic applies to the non-Ti 4070. 600$, or 620€ in Germany for the cheapest model I would want to buy (the most affordable is 600€), is too much for a card held back by such a crucial part.

The picture worsens when we go down the product stack to the RTX 4060 Ti. It costs the same amount of money as its predecessor, yet it offers no tangible performance benefit over the 3060 Ti on average over various games. There are outliers where the performance uplift is encouraging, and I assume it comes down to games that are heavier on compute power and lighter on memory bandwidth requirements. It shows me that the GPU could be capable of more. It is merely held back by the 128-bit memory bus. Remember, the 3060 Ti has a 256-bit wide bus. An additional 8 GB of memory will not change that imbalance. It will only solve the problem of the small VRAM configuration in some current and future games. And don’t get me started on the 100$ asking price for the memory. You basically still get 3060 Ti performance but for 3070 pricing. Very compelling. On top of that, the 4060 Ti only uses 8 PCIe lanes, and if you are still rocking a PCIe 3.0 system, like Intel’s 10^th gen, you may experience issues in games that push the VRAM needle past the highest number on the gauge cluster.

I would have preferred the 4060 tier to come with the 4070 tier memory subsystem – a 192-bit bus with 12 GB and 16 PCIe lanes. First, this would have solved the problem that the 3060 from two years ago has more and better connected memory. Second, given the price, it would have made for a much more compelling product. I presume it would have also improved the performance overall to a more considerable degree. Applying the same logic to the 4070 series, it should have come with a 256-bit wide memory bus and 16 GB of VRAM. These are expensive graphics cards and should be configured appropriately.

We can only hope that AMD finds its brains in one of its headquarters and does not fumble the RX 7700 and RX 7800 series releases as they did with the RX 7600. For 270$ MSRP, it is an acceptable graphics card. AMD has a chance to release compelling products. While NVIDIA does offer a better upsampling solution and power efficiency, raw rendering performance for a given price is still the first thing people look at. It is just a question of who AMD wants to cater to: the gamers or its shareholders.

I think gamers still find themselves in a bit of a pickle. During the pandemic and mining craze, everybody wanted a new GPU because they were really compelling products. People were even ready to pay ridiculous prices to overcome the limited stock. Two and a half years later, there are no more stock issues or artificially inflated prices, and nobody is interested in purchasing a graphics card. The available options are just not good enough. If I wanted to achieve an upgrade in all aspects, the first “cheapest” option would be AMD’s RX 7900 XT. It offers more VRAM, faster VRAM, an uncompromised memory bus, and a ton of performance. That would set me back 850€, a sum I could afford but not justify based on my current PC gaming habit. Nor would I want to pay so much money for a single component.

Do I worry too much about details that should not matter as long as performance is better than my current setup? Maybe. Still, these are expensive PC components, and I would like to invest in a good product, not a flawed product. AMD is playing the waiting game, and I sincerely hope they use the time to pay attention to the market reception and adjust their upcoming products accordingly.

It pains me to say this, but in the next generation, AMD should probably give up the idea of FSR being an open technology. DLSS is simply the better solution and a must-have on lower-end products. A high-quality upsampler on slower chips is way more important than on a 1000 Orens graphics card that plays 4K just fine. This way, gamers on a tighter budget can still enjoy good and smooth graphics with an overall slower GPU. AMD needs to get there, or the only way to overcome this issue is by offering much faster products than the competition at lower prices. DLSS lets me play A Plague Tale: Requiem at 1440p on an RTX 3060 and have fun. The game looks good, and the performance is acceptable, given the hardware. AMD must get to that level, too, and it might be by developing a proprietary upsampling solution that is restricted to AMD cards if that is the only way to catch up in quality.

I am not here to call everything garbage because it is in vogue to thrash big companies. I believe I made some valid points about why I do not like what the market offers. There is a saying that there is no bad product, only a bad price. That may be true. However, I also think lowering the prices would sometimes create odd products. Let me take the 4060 Ti as an example. The GPU itself appears to be capable of more. Lowering the MSRP would put it in a place where it might sometimes peak way beyond its asking price because it is an imbalanced product. How do you think a rear-wheel drive BMW M3 would perform with Toyata Prius stock wheels?

(Actually, the latest Prius generation may not be a driving lunch box anymore.)

Right now, I do not want to buy a graphics card. There is absolutely nothing compelling on the market for me. Thank you for indulging in my musings.