Intel: Meteor Lake Chiplet SoC Up and Running | Hacker News Reader

Intel: Meteor Lake Chiplet SoC Up and Running

anandtech.com

39 pointsrbanffy21 comments

>will be built on the Intel 4 process, which was formerly known as Intel’s 7nm process

This marketing will never fail to make me cringe.

Honestly I think converging on the TSMC/Samsung definition of "4nm" is probably a good thing overall for reducing confusion. Now that we're not using planar transistors any more the whole thing is a bit arbitrary but at least we can be standardized.

>> Honestly I think converging on the TSMC/Samsung definition of "4nm" is probably a good thing overall for reducing confusion.

Yeah, but if I'm not mistaken this actually IS the TSMC 4nm process, and Intel is still a couple years away from mass production of anything EUV. It is getting hard to keep up with the marketing BS. Anyone got confirmation on this being fabbed at TSMC?

Intel 4 is Intel. The Intel Alchemist GPU dies are fabbed by TSMC and Intel has clearly labeled those as being TSMC N6.

So is Intel 4 what was originally going to be their 7nm EUV? That would be good news.

Yes, this new naming scheme was announced with the roadmap back in July.

I left Intel in late 2020. Worked a bit on Meteor Lake in the pre-silicon design phase starting around '18 or so. I never really got over the lag between the most intense period of my day to day on a particular product and launch.

Such a shame that I, as an end-user is not the target audience for that shiny and promising "Flexible Tile Architechture".

I'd like 4 CPU tiles, ECC memory controller, no 3D GPU (frame-buffer, blitter and DP interface only) and no Management Engine, please...

Put in an order for $30 million of them and I'm sure they'll get you a SKU! (I'm not actually sure if $30 million is the magic number; probably depends on if there's another SKU just like it already or not)

Amazon and the other big players likely do have private SKUs for this sort of thing.

Otherwise, the 16 thread, ECC memory, no management engine, no GPU is a bit like the brown diesel manual transmission station wagon...

Apart from the "no management engine" thing, GP's configuration looks remarkably similar to some of Intel's existing Xeon SKUs. Not exactly your typical sedan, but a slightly customized pickup truck maybe.

I'd say it sounds exactly like what you'd want for a typical headless machine, like a server: no GPU, since it's headless, and no management engine, because it's a liability and can't be used for actual management anyway.

Or just wait for them to hit Ebay

> no 3D GPU

Do you like Windows? Because Direct2D, the 2D-renderer of Windows for Desktops / 2D applications, is built on top of DirectX.

That means you need a 3D GPU to run Windows 7, let alone Win8, Win10, or Win11.

------

Using 3D GPU commands, in particular a 2D rectangle as the basis of texture-mapping (with textures being uploaded to "GPU-memory"), is the standard methodology today.

Old-school 2D API (GDI) based off of bit-blitting just isn't how computers work anymore. INT 10H services (https://en.wikipedia.org/wiki/INT_10H) are long, long, long dead.

IIRC, GDI is provided on modern Windows in terms of Direct2D emulation layer (ie: GDI is built on top of Direct2D, which is built on top of DirectX 3D commands).

-------

This makes sense, because "pixel fill" / pixel-shaders are far more flexible than the ol' Bitblt. Even if you were to only be doing 2D code, modern GPU effects (ex: alpha-transparency, fades, etc. etc.) are best built out of pixel-effects and pixel-shaders.

I admit to being ignorant to Linux graphics drivers, but I'd expect the same thing. Modern hardware is built out of the 3D API / paradigm. "Running code" on a remote GPU/SIMD-cluster is far more efficient than having those effects calculated on the CPU (ex: alpha-transparency, anti-aliasing, painter's algorithm, z-depth, etc. etc.)

@dang: I posted this two days ago: https://news.ycombinator.com/item?id=31230687

How is this now at 33 points? I've observed this a few times; I'll post something and it won't get much traction, then someone else will post it and it shoots up.

There's a 14 hour difference in the time of day they were posted. My guess is this one caught the morning crowd and perhaps had less competition for eyeballs than something on a Monday?

There seems to be tiers of users. Some can post a duplicate link mere hours after you and get a boost and some cant post duplicate links even days later.

I see first comment there:

> Made for "Lower Power" a.k.a BGA trash. Sadly that's where the money is, because so many people simply do not care to buy PC for longevity they all want the castrated performance for the little convenience it has and Apple consumers being even worst accepting Soldered SSDs, RAM and everything irreparable garbage.

Why people are still so misinformed? Desoldering and soldering BGA is fairly easy and equipment to do that is accessible (you can get a rework station for less than £100 today).

Good luck reworking a CPU on a 10+ layer laptop board.

It doesn't matter how many layers board has. Desoldering / soldering process is the same be it 2 layer or 10 layer.

Boards with more layers tend to dissipate heat better.

It makes reworking much harder.

> and everything irreparable garbage

I know right. Back when we used to have vacuum tubes for computing, you just needed some basic glass blowing and electronic skills and you could repair any computer. Ever since we moved to integrated circuits and transistors, repairability has gone way down.

Reply on news.ycombinator.com