Man, if these don't undercut the competition by a massive amount it's not going to be good at all. They're just too late.
Yep... as predicted... but when I said 'even if they match current offerings, its not enough' I was told to be wrong by about 3/4th of the readers.
'I don't care if it takes 300W'
'It can be priced 700 bucks just fine, look at what Nv is doing'
'It doesn't have to perform exactly as good as (insert any)'
Yeah yeah... guys. They need a product on-par with current competition or what will happen is what AMD has experienced the last ten years. If you're not playing in the top half end, you're not playing, and you'd better play on all metrics: Heat, Power, Noise, Performance, Support. And if you want to start leading, you also need to pile onto that with soft- and hardware design wins.
So far Raja ticks the following boxes:
- Noise/Power/Heat: depends on perf per shader unit which so far isn't looking better than competition
- Die size: relates to perf per shader unit. How much FPS per square mm. The dies we've seen are effin huge, and they're not wide because they want to run lower clocks or TDPs. They are because they need to be or there is nothing to write home about.
- Price: huge dies translate to huge price points and low wiggle room for price cuts.
- Overall performance: I suppose it can run Crysis, but who knows.
- Time to market: 2 little 2 late
- Software side: yeah. Intel has a great IGP driver after twenty years. It shows an image.
But let's keep dreaming, maybe one day we'll wake up with the sun shining and Raja six feet below. Its when he's gone that design wins happen, look at AMD right now... And a smaller node is not going to save anyone - look at the past. A node advantage is only a very short lived one and it still can't mask a shitty architecture.
What I'm missing in the whole Xe story is a strong departure from what Intel has already done. They're still just tying EUs together.