Intel Linux Graphics Driver Prepares Task & Mesh Shaders, Vulkan Dynamic Rendering

[qarium]

So let me get this straight: You compare a huge die and huge sku from 4 and a half years ago which had super expensive HBM memory, on an older process node, to a current mid-tier die and sku with 128bit ram bus, and think just because you got 8% more performance for like 6-7% less price and somewhat less power consumption, that is progress?
Not even AMD shills would say that....

nice try... "super expensive HBM memory" LOL and 128mb L3 Infinity cache in 7nm TSMC is cheap ??? LOL

"on an older process node" this is always the case if you compare products from 2017 to 2021 products.

"and think just because you got 8% more performance for like 6-7% less price and somewhat less power consumption, that is progress?"

yes exactly this is the case. you get 8% more performance and 6-7& lower price and less power consumtion and you get AV1 codex support of the GPU and you get raytracing support in hardware...

yes this is called progess... yes i know you call it: not good enough

but it is what it is. it is still progess even if you call it not good enough

[TemplarGR]

nice try... "super expensive HBM memory" LOL and 128mb L3 Infinity cache in 7nm TSMC is cheap ??? LOL

"on an older process node" this is always the case if you compare products from 2017 to 2021 products.

"and think just because you got 8% more performance for like 6-7% less price and somewhat less power consumption, that is progress?"

yes exactly this is the case. you get 8% more performance and 6-7& lower price and less power consumtion and you get AV1 codex support of the GPU and you get raytracing support in hardware...

yes this is called progess... yes i know you call it: not good enough

but it is what it is. it is still progess even if you call it not good enough

Yes, HBM memory in 2017 was more expensive than the infinity cache in TSMC.

And 4+ years are a huge time period in terms of technology. Remember that in cpus, there was a huge evolution during the same time period, and prices didn't increase but decreased for the performance. CPUs are created from the same transistors as gpus, last time i checked. Perhaps cpus are made from sand and gpus are made from gold, right?

Yes, this is a disgusting duopoly, a typical price fixing cartel, and USA and EU do nothing to fine those companies for what they do. There is no market oversight anymore.

[MadeUpName]

Tell me about these Chinese GPUs, please...

Open Source drivers? It has Imagination Technologies IP, that makes me worrier about this.

Cost?

Distribution?

Availability?

Benchmarks?

Michael Get one and review it, please.

Something unexpected yet cool is that the cards have many APIs enabled. Specifically mentioned are OpenGL, OpenGL ES, OpenCL, Vulkan, and DirectX -- though we don’t know which version.

https://www.pcgamer.com/new-chinese-gpus-with-up-to-10-tflops-announced/

China had to do this due to US threats. But now that China had made the decision to build them you can expect that in 5 - 10 years they will be on the leading edge and eating market share from US companies outside of the US. US get rich quick schemes do more damage to US companies than Chinese. .Don't expect to see these outside of China for a few years.. But if they can fill some of the demand from China it will help people outside of China.

If you want Michael to test it maybe you should buy one and send it to him.

[anth]

Don't get your hopes up in current market, Intel dGPUs are said to not have throttled mining performance; just like AMD.

If a GPU was intended to have good graphics performance but bad mining performance by design rather than by driver limitations the way to do it would be to take advantage of both needing high memory bandwidth but the memory locations required for graphics being predictable while those for mining are intentionally not. That'd lead to a GPU with a narrow memory bus but large cache, ie like RDNA2, though I think AMD's goal was a step towards GPU chiplets and it is just coincidence that these are bad for mining.

The other way to make a GPU bad for mining is to keep the memory below the 4.7GB which etherium currently requires, which will happen for some of the smaller new generation GPUs due out soon.

[jaxa]

now you say intel is the competition we want and we need for the "entry level and budget level gpus"s... well i think you do not follow the news as close as i do...

because there is another new competitor in the market Innosilicon Fantasy One(Fenghua No. 1) ,,, and i think they will be cheaper than intel...

https://wccftech.com/chinese-gpu-mak...ions-5-tflops/

so your theory AMD/Nvidia is doing anything only because of intel... maybe they do it because there comes even more competition than just intel ?

Those Innosilicon GPUs will not make much of a difference outside of China, although it is something to keep an eye on.

AMD/Nvidia will be adjusting their budget GPU plans based on what Intel does with their 128 EU Alchemist GPU. Nvidia in particular might be giving the RTX 3050 8 GB of VRAM to counter the 4-6 GB models from AMD and Intel.

Intel ARC A380 vs. Nvidia RTX 3050 vs. AMD RX 6500 XT and RX 6400. It would be nice to see a sub-$200 price war but some of those will be closer to $250 because they can get away with it.

[smitty3268]

Nope, Ethereum network hash rate reaches ever new heights. That's where very most GPUs go to.

Ethereum is supposed to switch away from GPU mining to a new proof-of-stake model that won't require GPUs sometime next year.

I expect GPU prices to get a lot better when that happens. Maybe in Q3.

[qarium]

Yes, HBM memory in 2017 was more expensive than the infinity cache in TSMC.
And 4+ years are a huge time period in terms of technology. Remember that in cpus, there was a huge evolution during the same time period, and prices didn't increase but decreased for the performance. CPUs are created from the same transistors as gpus, last time i checked. Perhaps cpus are made from sand and gpus are made from gold, right?
Yes, this is a disgusting duopoly, a typical price fixing cartel, and USA and EU do nothing to fine those companies for what they do. There is no market oversight anymore.

you could do the vega64 perfromance in 2021 much cheaper if you do not add new features...
vega64 has only 264 decode the 6600X has AV1 decode...
also raytracing hardware support...
also infinity cache

all this increase the transistor count without giving you more performance. if you compare it to the vega64 in non-raytracing performance.

"And 4+ years are a huge time period in terms of technology."

it is up to you to proof that intel can do it better with more features with lower price...

you claimed there is no progress but i proofed to you there is process.

"Yes, HBM memory in 2017 was more expensive than the infinity cache in TSMC."

no one outside of AMD knows this for sure. in 2017 they plain and simple could not do it better because 14nm was the best node they have infinity cache was impossible on 14nm... now on 7nm they can do infinity cache. HBM was the only way to get any decent performance.

also remember the ability to do something cheaper with the same performance is also a technologe advantage you have or not have.