Raytracing on Meteor Lake's iGPU
chipsandcheese.com
chipsandcheese.com
- Why are relatively expensive and large GPUs like in the RTX 3070 commonly called “midrange” online? The meaning of “midrange” seems to be creeping upward.
Because the 30xx generation of graphics cards includes the 3050, 3060, 3070, 3080 and 3090. The 3070 is right in the middle of the range. Hence, midrange.
High-end gpus right now are the 4090 and 4080.
Within this world "midrange" has been creeping upward not via big shifts in these fundamental characteristics but via:
1. Prices increasing steadily across the board, due to shortages and market power 2. Power budgets (and corresponding board/cooler sizes) increasing across the board
The fundamentals (memory bus width -- still 256-bit; die size and performance relative to the top of the line) remain "mid-range" in exactly this same sense.
Nvidia has shifted its portfolio to the left, so they can charge more for a smaller chip. I suspect this is mostly due to increasing manufacturing costs, not (just) pricing power.
This is also a reason why they are banking on AI upscalers to drive improvements in the future.
The fact they can do this speaks to the lack of competition in the GPU market at the midrange and up. Compare this to the CPU market, where we now have Intel and AMD giving it everything to leapfrog each other every 6-9 months.
I don't begrudge nvidia wanting to spend a generation consolidating their market position - that's their right - just as it's mine to look at a GPU that performs roughly the same as one I bought 7 years ago at almost the same price and say "no thanks".
Afaik the new gen still manages to improve over the old one, albeit modestly, do you know of an example where is that not the case?
Specifically, at 4K, the 3060 outperforms the 4060 in Cyberpunk. Most of the charts do show a gain, but I'd describe it as marginal rather than modest.
Out of curiosity, which GPUs are you referring to, and how are you measuring or comparing perf?
I'm comparing perf by looking at sites likes userbenchmark and reading/watching reviews of newer cards. I'm running into games I can't play at a level that I'm happy with, but based on reviews I don't think a 4060 or 4060Ti would make a meaningful difference.
I’m not sure how you’re doing the price comparison. 1080ti launched at $699 (with inflation that’s $870 today). The 4060ti launched at $399. Why does that seem the same to you?
We’re in a ray tracing thread and for ray tracing, there’s no contest, the 4060ti wins there by a long way.
UserBench says today the 4060ti is 17% faster for 20% lower price today, which is about 33% better perf per dollar… which interestingly tends to match their user score and sentiment.
https://gpu.userbenchmark.com/Compare/Nvidia-RTX-4060-Ti-vs-...
Also worth noting that some old games are bottlenecked on the software, not on the GPU, so it’s not surprising to find a game that was around for the 1080 or before and doesn’t get any better with more modern GPUs. Comparing the 1080 to the 4080, fp32 perf went up ~5x, and due to Amdahl’s law, game benchmarks tend to see more like ~3x in overall system perf.
I don't make purchasing decisions based on what's 'fair', I look at what capabilities I have, what I lack, and what my budget is.
> Why does that seem the same to you?
Exchange rate. But you make a valid point that I should also factor in inflation.
> We’re in a ray tracing thread and for ray tracing, there’s no contest, the 4060ti wins there by a long way.
It wins in the sense that it can do ray tracing while the 1080Ti cannot, but based on my understanding, the xx60-tier cards aren't viable for raytracing at 60fps without DLSS anyway, and I'm still not convinced I want to consider DLSS performance to be the number I care about in my purchasing decisions.
> UserBench says today the 4060ti is 17% faster for 20% lower price today,
The most important question I look at is "what games can I run at what resolution" and I don't think there's a single game I can get acceptable performance at 4K on a 4060Ti or even a 4070, that I can't get acceptable 4K performance on a 1080Ti. There are probably a couple that I could run at 1080p that I have to run at 720p currently. I'm not using a variable/high refresh rate monitor and i'm not playing competitive shooters.
> I don’t think there’s a single game I can get acceptable performance at 4K on a 4060Ti or even a 4070, that I can’t get acceptable 4K performance on a 1080Ti.
This is a very different claim than you made above. Yes, all the GPUs since the 1080 can get “acceptable” 4K performance on most games that will run on the 1080 (I assume acceptable means >= 30fps), so that’s not surprising. This is ignoring the game quality settings. The benchmarks you pointed at demonstrate that if the settings have a 4060ti running just barely at 30fps, then the 1080ti will not be able to hit 30fps with the same settings, because the 1080ti is 17% slower.
As sibling pointed out, the $699 1080 Ti would be $870 today in U.S. dollars.
The 4060 Ti is also a remarkably bad product (for the price). If you swap in the 4070, you're looking at a launch MSRP of $599 which is less than the 1080 Ti even before inflation, and about 50% faster (and over 30% faster than the 4060 Ti.)
Still not mind-blowing given the nearly 7 years between launches. There's a big delta going up to the 4080, both in price and performance. On UserBenchmark it's "135%" faster than a 1080 Ti, but at a mind-boggling $1200.
That has been the case since before I touched my first computer. The first mainframe I touched had 16 megabytes of memory, and my first desktop had 48 kilobytes. Both were midrange in their respective categories (although the mainframe was in the top in memory, it was only average in processing power - the Apple II+ was better than a VIC-20 and performed a little better than a C-64, but was slower than most of the "professional" personal computers).
My work laptop is midrange today and is many, many orders of magnitude more powerful than the supercomputers of that period.
To put things into perspective, my phone runs Unix on a RISC CPU closely coupled with an array processor.
That's Moore's law. It has slowed down a bit, in part from the difficulty of doubling density every couple years, but also from the fact even a 5th generation Intel i3 is vastly more capable than what the average user needs. In GPUs it's a little bit different, with games requiring increasingly ludicrous amounts of compute power, pushing the "good enough" range upwards every year.
If you really like to play Pac-Man, an 8-bit CPU and a simple CRTC should suffice.
Still, the visual spectacle of what some of these products can create is a fantastic experience in itself. It can enhance immersion in a drastic way, and especially for story heavy games, that can act as a direct multiplier on an already good story and structure. Bad games will still never go past being just a tech demo.
So it's probably useful if you've got a 4090, but beyond that more try it once, go "Huh, that's neat", and then disable it.
But there's always a chicken and egg problem for every new feature - it may make sense to push support even if it's not particularly useful right now, as once it's ubiquitous there may be more use cases for lower total performance cards - "light" ray tracing may be a better solution than some rasterization tricks for things like GI, or even things like AI line of sight or other non-specifically-graphics tasks.
But as gamedevs have to support the non-RT path anyway, it probably doesn't make sense to develop two separate paths. So it's relegated to "optional" visual features only.
RT is pretty nice and more and more games are featuring it, problem is at least on consoles they cannot go crazy with it so some games only do reflections for example or just shadows or just global illumination.
Sony is supposedly releasing their Pro model of the PS5 later this year with improved support for RT.
For a mobile GPU, like Meteor Lake? I wouldn't even bother.