Recent content by tunafish

I can't help but think that a new console in 2025 or 26 is a mistake. I really don't think that there is any "there" there for improving console performance on raster loads. Current gen consoles are good enough that you can make the kind of game you want to make, and have no horrible glaring...

I would be genuinely surprised. By every rumor it will be faster than a 7800xt, I think they can sell N48 as x800 and x700 without any kind of issues. I'm just hoping they have the sense to not use x900.

To be clear, AMD usually has like a month from announcement to availability.

What role do you see local AI inference having? The way I see it, all the interesting near future AI products will run on models that are way too big to fit on a smartphone or a laptop. They will run on huge inference server farms instead. This also makes more economic sense, because batching...

Not arguing for any specific performance level of N48, but I'd like to make two observations: One of the more counterintuitive findings from statistics is that you should always expect to revert to the mean. If two very tall people have kids, you should expect the kids to be shorter than them...

Makes sense for an AI chip. I wonder if nV will make a separate server GPU product line for FP64, or if they will just cede the market to AMD. It's a much smaller market than AI, but it's quite prestigious and sees a lot of software investment.

What? No. The rules, as defined in antitrust law, basically say that when you have market power (also, as defined in antitrust law), you can no longer do things like deciding who you sell to based on what benefits you, but you will instead have to start treating all customers the same. There's...

Depends, I don't know how large a market share Nintendo has and whether that is considered comparable. If a game console maker wanted to do orders from both AMD and some other source, and AMD punished them for this, they'd quite possibly get sued and lose. Again, a dominant market position is...

The difference is that Nvidia is probably dominant enough in their market that them doing it is illegal. Being a monopolist/having market power is not in itself wrong, it just means that you have to play under different rules. Rewarding brand loyalty (or punishing disloyalty, same thing) is...

Note that you cannot harvest every defective die even if every part of the die is redundant. Faults don't just mean "transistor doesn't work", there are plenty of potential faults that trash the entire chip even if the part it occurred in isn't important. The canonical example is a direct short...

Dispatching work is easy. The problem with treating multiple pieces of silicon as a single GPU is how do you manage memory coherency, when your program touches the same value in two different places, how do you make sure they see the same thing. How would you do coherency? I would expect there...

Yes, just repeating myself here, right now in inference the platform that people are using is PyTorch, with vLLM and running on NVidia hardware. That's where AMD needs to be good at to compete, and other platforms and workloads just matter a lot less. I think that MI300X will do well enough...

Tensorflow is basically obsolete, no-one uses it anymore. Today, you are using a weird in-house thing or PyTorch. Er. The standard is PyTorch, and for inference that usually means using vLLM. Most people run it on NVidia, but that's what's being used, and that's what makes sense to benchmark...

The news that they have major outside customers for 18A is much more important than their financial performance of the past quarter. Intel's current state is mostly caused by falling off the leading edge of process tech, if they can catch that train again they can come back in force.

Ellewood and Brooklin coming late 2024. Based on this slide: It seems like they refresh a bunch of things, but don't add more ram, or a more powerful apu.