Is it harder to design GPU than CPU? I don’t know. But if we can design first class CPU, what is preventing us from designing first class GPU? Talent? Management? Lack of corporate support? At least we should know where the problem is, right?
19 replies (most recent on top)
Trade knowledge and industry experience. Fabrication processes are very delicate and require long periods of time to perfect. Sometimes many years. Best thing Intel could do is steal some industry experts from competitors by paying them a very generous wage to leave their lives of comfort. Change your compensation strategy and help those 20% of the people driving the business forward while the other 80% rest and vest. Two people with same title and band do not produce the same output. Companies like Intel are terrible at lifting talent up.
@2jjs+1mSsObtZ. whatever dude. U don't sound happy.
Intel graphics kernel driver software sucks but still ppl get promoted. hiding bugs from top management and fix later as bargaining for promotions or come to take H1 visa
So happy joined online retail company. Out of this GPU mess
@1hlv - What's worse is the sub-standard engineers have been promoted many times and are now middle managers, directors, and VPs running the show.
There are sub-standard engineers at Intel...and their spouses and their brother-inlaws, their siblings etc have all been hired at Intel
GPU driver team just gave up...trying to virtualize PVC. There goes all hope of getting into or attempting to or pretending to even get closer to anu GPU markets.
Why can’t Nvidia make a good CPU?
Same reasons.
Decades of continuous refinement can’t be replicated in a few years.
First class CPU?
They both have 20+ years of advantage in the field, including technologies, patents, know-how etc. that come with it. It's really impressive that our GPUs are already so close on the first gen. It looks like we are on the right trajectory to catch up in a few years around Druid, it was just not realistic to do it on timeline as quick as our management promised - traditional policy of overhype and underdeliver. I just hope the project doesn't get axed beforehand.
You’re going to dismiss 30 years of nvidia domination?
Iterating and learning every generation?
100s of millions of lines of code in drivers developed over decades.
The smartest graphics engineers on the planet?
Leveraging their sw investment to pivot to AI?
People at Intel are fvking idi?ts if they think they can just take on NVidia after a few years. They should have bought that talent and IP 15 years ago but didn’t because every Intel exes past and present are fvking clueless about actual technology development.
well Intel can't design a first class cpu either
Raj@ architected too grand of a vision that the sh1t engineering design teams couldn't cobble up effectively lmao. This was Intel's 3rd attempt at GPU that fell flat on it's face.
Horse vs a car time to get into 0-60 million!!
Horse cannot and dies
Look at the linux driver commits by developer and compare that to the number of people "working" on the driver. There are maybe 10 active developers while the other couple of hundred "developers" do what???
@oi is so ignorant. Just blurted out the word CUDA without understanding much about it.
Larabee shows the capability
SW ecosystem & driver maturity
Sell spin out the GPU business. Then focus on making chips for Nvidia. Then, maybe, perhaps, Intel can survive.
One word: CUDA
Nobody can compete with Nvidia.