Thoughts on the RTX 4000 series

Suffocating the PC gaming market

Nvidia and AMD having fun

Nvidia announced its RTX 4000 series video cards on September 21. They showed 3 products; the 4090, 4080 (16 GB), and 4080 (12 GB). The 4090 launches on October 12 at $1600 USD while the other two are planned for a November launch. This announcement caused quite the controversy among hardware followers and people looking forward to upgrading.

I was affected by this as well, because I’m considering upgrading my PC with this year’s new hardware. I plan to buy AMD because I prefer their Linux drivers. But after seeing what Nvidia had to show, I’m getting worried about what AMD is going to put out.

Price

This was by far the biggest problem. The 4090 will cost $1600 USD, the 4080 16 GB is $1200, and the 4080 12 GB is $900. The 4090’s cost relative to the 3090 isn’t terrible, and people buying this tier of card do not care about price or have genuine use cases. I think the 3090 is a ridiculously priced product, but that’s not important for this article.

The real problems are prices of the 4080s. As the 4080 12 GB is a special outlier, I’ll talk about the 16 GB model. The 3080’s MSRP was $700 at launch, so this is an insane $500 markup. In percentages, that’s nearly a 70% increase; absolutely mind boggling. No amount of inflation, innovation, supply chain costs, or silicon costs can justify this jump.

Why would the 4080 be this much more expensive than the 3080? It’s not as if its performance is double that of the 3080. In fact, Nvidia’s presentation was focused on trying to hide the true rasterization performance increase; they heavily focused on DLSS 3 and its frame generation feature to artificially inflate performance gains.

Why so expensive?

There’s two main theories on why Nvidia priced the 4000 series so high.

Two theories
  1. Nvidia and investors are still high on crypto profits

    Those that tried finding GPUs during the crypto boom will know. Nvidia could sell any product they made, which is why the lower end 3000 series was so disappointing. Why make a competent product and price it competitively, if you know it’s going to sell anyways for much higher?

    Nvidia got comfortable with the inflated prices and profits, and this shows in their earnings reports. Once the mining bubble popped, their ‘gaming’ earnings and forecasts took nosedives.

  2. There is an abundance of 3000 series cards to clear

    Nvidia didn’t see the crypto bubble collapsing but had an excess of 3000 series cards. Now, miners are dumping used, cheaper cards and the 4000 series is out. These factors would negate reasons to buy a new 3000 series card; you can either get the new generation or buy cheap from a miner. This situation is nearly identical to Nvidia’s 2000 series launch.

    Nvidia can’t do anything about the miners, so they decide to manipulate 4000 series cards. To let the remaining 3000 series sell, Nvidia decides to overprice the new cards in an attempt to make it look undesirable. This would push buyers to the 3000 series instead. When the old stock depletes, Nvidia might lower 4000 series prices and roll out lower end models. Whether someone buys a 3000 series or 4000 series, Nvidia wins.

Of course, there may be some genuine reasons contributing to the price hikes. Manufacturing cost, R&D, architecture cost, and new technologies Nvidia develops may be costly. However, the price is ultimately judged by the buyer, not Nvidia.

After the 4000 series was announced, lots of threads, posts, videos all pointed to similar thoughts: the prices are terrible, don’t buy the 4000 series, Nvidia is horrible, wait for AMD.

Nvidia’s story doesn’t just end here though.

The RTX “4080” 12 GB

This was another big issue that people spotted: the 12 GB 4080 is very different from the 16 GB model. It’s more accurate to say the 12 GB model is really a 4070 given the name of 4080 and its accompanying price tag.

Long explanation Each Nvidia card generation has a different design (architecture). These architectures are named after scientists. Nvidia's recent architectures have been: Pascal (GTX 1000), Turing (RTX 2000), Ampere (RTX 3000), and now Lovelace (RTX 4000).

Each silicon die uses the architecture name and a number for identification. For instance, the largest Ampere die was called GA100. The next largest was GA102, then GA103, and so on. The largest is XX100, then the number increases for smaller dies.

GPUs with the same tier number (ex. 1060 3GB and 1060 6GB, 2070 and 2070 Super) tend to use the same die. This isn’t 100% consistent, but is generally accurate. For Lovelace, the 4080 16 GB uses AD103 and the 12 GB uses AD104. Both are called 4080s yet use different dies, which is something that hasn’t happened before.

In simple terms, the silicon that goes into the 16 GB is very different from the 12 GB card; there’s around a 25% difference between the two. The memory bus is also much larger on the 16 GB (256 bits) than on the 12 GB (192 bits). That’s a massive difference in hardware, so much that some people decided to call the 4080 12 GB a 4060 in disguise because of x60 cards typically having a 192 bit bus.

Overall, the 16 GB card is significantly faster than the 12 GB one. It’s large enough to decisively justify the 12 GB card as a whole tier below, hence the 4070 in disguise. This means that Nvidia is selling the 4070 for $900, when the 3070 MSRP was $500.

Even if the real 4080 (16 GB) sold for $900, that’s still a hefty $200 increase over the 3080’s MSRP. Even if there are enthusiasts willing to pay for high tier cards, that price jump is a signficant barrier. But Nvidia did one better and charge that price for a 4070.

What will AMD do?

The Nvidia story should be clear by now: insane pricing and deceptive naming. There’s more to the 4000 series announcement, but this is the biggest part of it.

So how about AMD? There are a lot of people inhaling hopium, praying that AMD will release competitive cards at significantly lower prices. This would have a few objectively good benefits.

  1. AMD could gain a lot of market share
  2. AMD’s brand loyalty can improve
  3. It would cause Nvidia to lower prices or have difficulty selling 3000 series cards

Essentially, the ideal scenario is a repeat of what early Ryzen CPUs did to Intel. This would be good for AMD and for the PC market. AMD can repeat their gains in the CPU space against Nvidia, and consumers can finally afford video cards at more sane prices. It would also pressure Nvidia to be more competitive. There is also no rumour of AMD sitting on heaps of unsold 6000 series cards. Therefore, 7000 series products shouldn’t be competing with the 6000 series.

There’s a good reason why I consider this a dream though. The conservative side of me believes a repeat of the 6000 series is more likely: AMD prices their cards slightly below Nvidia with roughly equal performance. This maintains higher margins and doesn’t trigger Nvidia to lower prices and neutralize a price war. Even if market conditions are more favourable for lower prices and better competition, I’m not confident betting on the 7000 series to dominate this generation.

I really hope I’m wrong, it would be amazing to see affordable video cards again. Ultimately, what AMD does depends on what they want to go for: actually attempt to gain market share, or conservatively follow Nvidia pricing and disappoint the market. Extrapolating AMD’s recent behaviour leads to the underwhelming route.

What about Intel?

Instead of praying that AMD will save gamers, people should really be praying for Intel to save the market. Intel has many incentives to enter the GPU space; they have nothing to lose and would get access to the lucrative datacenter market. With three competitors, current prices would be harder to maintain.

Intel also has the money to brute force their way in. They have the cash to endure some attrition and even have their own fabs, which is an extremely valuable asset. If their R&D goes well, they could potentially turn their GPUs into a massive success.

Intel is starting from complete scratch; their first generation Arc cards have terrible drivers and are basically vapourware. However, the hardware does exist, the hope is not dead. Intel will have to fight on price initially to garner trust and loyalty, which is good for the consumer. The GPU market is just too profitable in the long run for Intel to already bail, and gamers benefit from their pursuit. Intel just has to persist for a few years until their software and hardware mature, and then they’ll get the money they’re after. Only then does the consumer GPU market have a real chance of becoming healthy again.

Praying

The DIY PC market is in uncertain territory. Performance gains are good, but pricing is unsustainably increasing. Intel’s entry into the GPU market has the potential to fix many problems, and that’s what people really need. If current pricing trends continue, the PC market might start losing a lot of buyers.

I’m still debating on upgrading with this generation’s hardware, but of course I will wait for details about RDNA3 and Raptor Lake first. In the worst case I’ll go with older hardware, but I also want to make this hardware leap as significant as possible. Nvidia’s pricing is an ominous foreshadow that I hope will be overturned, A market where a mid-range GPU alone costs as much as a console is a bad one.