Hacker News Reader

(author here) I appreciate the feedback, but I have trouble understanding where you're coming from.

"what exactly makes it ambitious?" I thought I outlined that it was much more powerful than Intel's prior (RPL) iGPUs both in the first paragraph, and in the conclusion. It competes with the powerful iGPUs AMD has been getting into handhelds.

"AMD as afterthoughts" - Can you explain how you got that impression? I opened by noting how AMD's APUs are extremely competitive if not downright dominant in handhelds (Steam Deck and ROG Ally called out as specific examples) and threatens Intel in the laptop scene too.

"4K 120 FPS" - uh no, you're not getting that on an iGPU unless it's a game from 15 years ago. I suggest checking the very wide variety of other reviewers who run game benchmarks on devices like the Steam Deck or ROG Ally. 1080P or 720P 30 FPS is a good target, and you might need medium or low graphics to get there. That's what I mean by compromised gaming. It's not the same experience as say, gaming on a desktop with a midrange discrete card.

"lesser" iGPUs imo aren't a new sweet spot, the sweet spot is just holding on to older cards that still deliver better performance than these iGPUs. For example check Steam's hardware survey (https://store.steampowered.com/hwsurvey/videocard/). There are more people with a GTX 1080 than a RTX 4080. And PC games are optimizing for stable hardware capability. The latest games are usually playable on Pascal.

> "4K 120 FPS" - uh no, you're not getting that on an iGPU

You are not going to get that from many dGPUs either. You might get that from the high end in the last or previous gen, but not from the midrange or 2-gen-old models.

It really depends on the game.

I have a 4090 (for work, I swear…). Cyberpunk is smooth, but I don’t get 4k 120fps. But I also play a lot of little indie games. Terraria? Stardew valley? Slay the spire? This stuff doesn’t need a dgpu at all. And I suspect a significant percentage of global gaming hours go into stuff like this now. Games that really push the hardware are expensive. (Well, or badly made). Either way, it’s usually bad for business.

Basically, NVidia is the bar. Ambitious to me implies at least challenging NVidia. AMD makes competitive GPUs in certain arenas (and for Linux is the main choice due to drivers), but for me no definition of "ambitious" involves even thinking about AMD.

Well, 4k 120 from an iGPU ... you're saying THAT is ambitious? There's the bar!

Historically Intel has about every 5 years started to rumble about getting serious in the discrete markets, and they make some marketing fluff, but nothing even remotely competitive outside the iGPU "meh" range ever comes out.

So if I hear Intel being "ambitious" and then read an article that basically pretends (I'm not accusing you of anything) NVidia doesn't exist, well, seems like a failed premise to me.

I'm pretty negative on Intel over the last decade, you'd think I was a spurned contractor (I'm not, never worked there). Intel is definitely in the "prove it" mode. They've so massively failed/squandered opportunity at smartphone chips, SSDs, memory, graphics, and then finally screwed the pooch in process tech and CPUs. So clearly an engineering company that was hijacked by finance MBAs are driven into the ground, and it is HARD for companies to come back from that poison, especially when they had about 30 years of near-unchallenged monopoly dominance in the marketplace.

I didn't want to imply "author sux Lol" the article was pretty in-depth and information depth, but it remains the basic premise is flawed, because the source marketing/press release by Intel is about an 80% chance of being BS or "same story, different half decade".

4K 120..."There's the bar!" - By ambitious I meant Intel's serious about getting competitive gaming performance in the handheld or thin/light laptop category. MTL's iGPU is ambitious compared to older standard Intel iGPUs like the HD 530.

"Nvidia doesn't exist" They don't exist in the iGPU market, unless you count the Nintendo Switch. The Switch doesn't run the same games that Meteor Lake and Phoenix do, and therefore I don't think it's an interesting comparison. But I do have data at https://chipsandcheese.com/2023/12/23/nintendo-switchs-igpu-... if you want to factor in Nvidia. Same with Nvidia's discrete cards or AMD's desktop RDNA 3 variant (with the larger 192 KB vector register file). Neither of those can fit in the same form factors and power envelopes that Meteor Lake and Phoenix compete in.

What Intel source marketing/press release stuff did you take issue with? I'll be honest, I didn't go over their Meteor Lake marketing/press release materials in detail. But if they did claim something crazy and didn't deliver, I can understand the disappointment.

https://wccftech.com/gpu-market-rebounds-q2-2023-amd-nvidia-...

> The integrated segment had a total of 48.82 million units shipped worldwide followed by the high-end GPU segment which saw 6.84 million GPU shipments, 2.59 million shipments in the mid-range category, and 1.81 million in the entry-level segment. Workstation GPUs also shipped 1.50 million units.

Integrated graphics may not have the profit margins of dedicated, but in sheer quantity, they dwarf add-in boards.

But how many of those 48.82 units are actually used in any meaningful capacity?

Intel is spending a great deal of money manufacturing stuff that’s then utterly wasted on any system with dedicated GPU’s or more commonly never use more than a small fraction of that 3D capacity. It just seems like a multi billion dollar waste from a company that’s so used to being a near monopoly it can’t step back from the iGPU trap.

>Historically Intel has about every 5 years started to rumble about getting serious in the discrete markets, and they make some marketing fluff, but nothing even remotely competitive outside the iGPU "meh" range ever comes out.

I think the key thing you may be missing here is Intel Arc which is Intel's first real dGPU. And now they are using that tech in their iGPUs.

I took the ambitious part to be changing how the gpu was connected to the rest of the system. Seems like a decent jump although not exactly risky.

But the conclusion sums it up well:

> Meteor Lake abandons the familiar Sandy Bridge client design that characterized Intel’s consumer CPU offerings for more than a decade. In its place, we get a flexible chiplet strategy that lets Intel pile on accelerators in response to market demands.

So they're basically following AMD's lead by going to a modular chiplet based package...

I don't think it's really about followimg AMD. It's the only sensible thing to do given the state of scaling and economics.

All interesting things Intel are doing are chiplet based. I think it's partly about showcasing their packaging tech in an attempt to differentiate their fab offering. You could argue there's nothing very special in that respect here, they've done this process and vendor mixing before elsewhere. Except this will be a high volume product line so they need good yield and capacity. But that's just my read on it, maybe it's just the cheapest way to stay relevent.

I simply meant it was ambitious compared to prior Intel iGPUs, especially stuff like Skylake GT2 where you could be playing at 720P low and still not get 30 FPS.

The chiplet strategy is kind of ambitious too because there's power overhead and they're targeting battery powered devices with Meteor Lake.

Eh, writing is hard. I never liked English class anyway

Don't let the haters get you down, I love your articles.

Oh I don't mind the discussion here at all, I'm just occasionally puzzled at things I thought I was pretty direct about.

Honestly though I don't like writing. Finding stuff out about hardware is fun. Weaving it into a coherent article is a chore.

I think most people don't appreciate how hard it is for hardware companies with established products to make non-trivial architectural changes. It's risky and it usually snowballs and becomes more risky. Especially when others in the market are already doing similar things, it downplays the change.

I love researching things too but I can't write coherently. I really enjoyed your article, thank you.

Re the power overhead. They probably get some power back by using smaller nodes on some chiplets than they'd otherwise be able to afford if doing the whole design with the same process. And they are betting on the benefits of backside-power delivery.

If you're not gaming or doing serious gpgpu stuff, the on-board GPU has been a sweet spot for well over a decade now.

> Alas the tantalizing prospect of real time raytracing seems to be backburnered under the AI hype onslaught.

Can you elaborate on this? Is that because ray tracing needs only simple vector units, while AI is driving tensor-type architectures?

Not OP, but to provide some historical perspective, RTX hardware raytracing is very firmly a gimmick and it isn't AI nonsense that's going to be the end of it. It's going to go the way of PhysX, 3D Vision, and EAX audio. Cool, but complicated and not worth the effort to game devs. Game designers have to make all the lighting twice to fully implement RT, and it's just not worth the effort to them.

Nvidia's own site[1] lists a total of 8 Full RT compatible games, half of which they themselves helped port. There are far more games that "use" it, but only in additional to traditional lighting at the same time to minimize dev costs. Based on that and past trends, I would personally predict it to be dropped after a generation or two unless they can reuse the RT cores for something else and keep it around as a vestigial feature.

[1] https://www.nvidia.com/en-us/geforce/news/nvidia-rtx-games-e...

"Full RT" means the game uses 100% raytracing for rendering (in some mode), which currently needs still far too much power to be a mainstream thing and is only added in a few games to show the prowess of the engine (IIRC a review of the Cyberpunk 2077 Full RT mode only a 4090 is really able to provide the power needed). The important entry is "yes", which shows far more entries and means there's Raytracing enhancements in addition to rasterization.

So, no, it's quite the opposite of what you stated: RT gets more important all the time, is not a gimmick and there's zero reason to assume it will be dropped in the future.

It is a gimmick in that you have to sacrifice far too much performance. An RTX 4080 will need to run at 1080+upscaling+framegen to get above 60 (!) FPS with ray tracing.

No thank you, I’ll take buttery smooth real 120 FPS at 4K. Especially because games have gotten so good at faking good lighting.

Maybe with the RTX 6xxx series it’ll be viable.

It does look fabulous though. I have a 4090 and absolutely turn RT on for cyberpunk. Even with a 4090 I use upscaling for a good frame rate. But the resulting image quality is just spectacular. That game is really beautiful.

No contest there, it looks really great! But for me, not enough to go back to “choppy” gameplay.

Personally, I'd rather vote 120fps and 4k as gimmick. If I had to choose between Raytracing or 120fps? Always Raytracing.

You could argue 4K as a gimmick if you’re sitting at TV distances, but the difference between 60 and 120 FPS is extremely jarring. Try playing at 120 and then mid-session capping it at 60.

I would hardly say it's a gimmick. Now that frameworks like epic's unreal engine and others implemented for the developer. I don't see these technologies going away. One can hope that nvidia's dominance lessons overtime.

I believe the next big thing is generative AI for NPCs as soon as the models are optimized in the hardware for the average GPU. Let's see what the next generation of Intel AMD and arm produce. Windows branding of an AI is going to make this possible. It's going to take years though for the market to be saturated with hardware capable for developers to pay attention.

You do realize that RT greatly simplifies the job artists and engineers have to do to make a scene look well lit? The only reason it's done twice currently is because GPUs aren't powerful enough yet. RT will simplify game production.