Intel Sends Updated GPU Firmware Handling, More Meteor Lake Graphics Code For Linux 6.1

[phoronix]

Phoronix: Intel Sends Updated GPU Firmware Handling, More Meteor Lake Graphics Code For Linux 6.1

Intel engineers have submitted another batch of "i915" kernel graphics driver changes intended for Linux 6.1 that include updated GuC firmware version handling, more DG2/Alchemist Arc Graphics work, and also more Meteor Lake "MTL" enablement for that successor to Raptor Lake...

https://www.phoronix.com/news/Linux-6.1-Intel-GPU-FW-MTL-More

[microcode]

Allegedly, the product has launched.

[Barley9432]

Got an A380 and quickly found out that basically no VKD3D games work: https://gitlab.freedesktop.org/mesa/mesa/-/issues/5003

That's a great disappointment, and the guy who was supposedly working on it left Intel last year, with the other team member not speaking for a year now. How reassuring.

[kneekoo]

Yeah, that's why I won't consider an Intel GPU until they they "prove" they have proper support. I have plenty of paper holders at arm's length already, no need for another one that also costs a lot.

[ms178]

Allegedly, the product has launched.

Allegedly, ARC won't have a bright future. Its major downsizing/cancellation has been the outcome of their internal review.

[middy]

Allegedly, ARC won't have a bright future. Its major downsizing/cancellation has been the outcome of their internal review.

yup. shame it seems like arc is cancelled.

[ms178]

yup. shame it seems like arc is cancelled.

If they realized that their consumer offerings were not good enough and are not able to fix these problems soon, it is better to pull the plug early. There is no need to throw away money on sub-par products, it's better to take their time, get back to the drawing board and come back with something more competitive. As they will keep making GPU tiles for server and CPUs, I am confident that they will throw their hat back into the ring when they have something more compelling to offer. Their ambitions were way too high, and more focus might be beneficial in the long run.

One thing is clear, the past two years have shown that we need more competition in the discrete GPU space and GPUs are getting even more relevant with the demand for GPU compute growing. Intel just announced investments for their oneAPI platform, the time will come that they want to leaverage this on the desktop, too.

[qarium]

If they realized that their consumer offerings were not good enough and are not able to fix these problems soon, it is better to pull the plug early. There is no need to throw away money on sub-par products, it's better to take their time, get back to the drawing board and come back with something more competitive. As they will keep making GPU tiles for server and CPUs, I am confident that they will throw their hat back into the ring when they have something more compelling to offer. Their ambitions were way too high, and more focus might be beneficial in the long run.
One thing is clear, the past two years have shown that we need more competition in the discrete GPU space and GPUs are getting even more relevant with the demand for GPU compute growing. Intel just announced investments for their oneAPI platform, the time will come that they want to leaverage this on the desktop, too.

problem with that is it will never happen without some kind of marketshare.
you need products out in the wild in the hand of testers and customers who make responses that you can put developers on to fix it.

they should think about it in a way of apple m1/m2 is conected to linux it is better to produce and sell hardware without any driver and work on it to get some driver in the future or better support in the future than have no hardware at all.

the release of the intel a380 was their best bet because amd 6400/6500 have no AV1 decode/encode...

the question is here why do bigger cards like 8gb or 12gb or 16gb vram cards if these cards with the driver support they have right now would only result in the fact that you need to sell the hardware below the price of production and make a loss with it.

why not improve the driver of the a380... and then make a die shink to 5nm and again only produce a 6-8gb vram low-end card and again improve the drivers.

people will not spend 600€ on an 16gb intel arc gpu if they can have a amd 6800 for 630€ ...

[ms178]

problem with that is it will never happen without some kind of marketshare.
you need products out in the wild in the hand of testers and customers who make responses that you can put developers on to fix it.

I agree, in the video it was mentioned that Battlemage could be low to mid range only. That's not something enthusiasts will be waiting for but could serve parts of the markets not willing to spend 300 EUR+ on a GPU thus serving the purpose you described very well.

the release of the intel a380 was their best bet because amd 6400/6500 have no AV1 decode/encode...

I am not sure if it is their best bet, but they certainly can compete on price and feature levels at the low end. Performance is not that important there but could get more decent with better drivers. It's a bit all over the place, a more consistent showing would be better.

the question is here why do bigger cards like 8gb or 12gb or 16gb vram cards if these cards with the driver support they have right now would only result in the fact that you need to sell the hardware below the price of production and make a loss with it.

As there were rumors about critical hardware design flaws affecting the top end cards, it is very likely that these wouldn't make the huge splash Intel was hoping for. And the chip size is bigger than a 3070 on a more advanced node but delivers less performance, that tells us that this is already some kind of an engineering disaster at least in comparison with the competition.

why not improve the driver of the a380... and then make a die shink to 5nm and again only produce a 6-8gb vram low-end card and again improve the drivers.

people will not spend 600€ on an 16gb intel arc gpu if they can have a amd 6800 for 630€ ...

I suppose that is the way they want to go with Battlemage, it remains to be seen if that becomes true. We have seen in the past with the well-received RX570/RX580 and 5700XT that you can get market and mind share with a high-volume mid-range product as that is what most people buy. That would be a good start for Intel to build their GPU business, but the drivers and hardware should be in a way better state for a release.

[qarium]

As there were rumors about critical hardware design flaws affecting the top end cards, it is very likely that these wouldn't make the huge splash Intel was hoping for. And the chip size is bigger than a 3070 on a more advanced node but delivers less performance, that tells us that this is already some kind of an engineering disaster at least in comparison with the competition.

i do wonder myself why they are on 6nm TSMC why they don't do their experiments on intel node instead. would be much cheaper for them to experiment with.

what you descripe "chip size is bigger than a 3070 on a more advanced node but delivers less performance"
is the same at a380 with 150mm² die size vs radeon 6500 with 100mm² die size.

this means they do something wrong and yes "critical hardware design flaws"

really intel bought AMD GPU IP to make their GPUs and why not buy complete chip designs ? i dont unterstand why intel want to make their own gpus with their own drivers... they could have licensed a complete gpu and use the amd driver...

I agree, in the video it was mentioned that Battlemage could be low to mid range only. That's not something enthusiasts will be waiting for but could serve parts of the markets not willing to spend 300 EUR+ on a GPU thus serving the purpose you described very well.
I suppose that is the way they want to go with Battlemage, it remains to be seen if that becomes true. We have seen in the past with the well-received RX570/RX580 and 5700XT that you can get market and mind share with a high-volume mid-range product as that is what most people buy. That would be a good start for Intel to build their GPU business, but the drivers and hardware should be in a way better state for a release.

they just need to accept their are 3. class gpu company and they need to accept they need multible generations of gpu chips not only 2...
amd did need longer than 10 years to make their opensource driver possible... so intel need 5+ years in the low-end to get the experience they need to make highend gpus..

no one will spend 600-1000€ on an intel experiment but if they are lucky people spend 300€ on an intel gpu as a experiment.

right now the gpu market is like this: Nviida is the premium brand... AMD is the performance per dollar king ... and Intel is the low-budget position...

and intel will stay in this position for at minimum 5 years.