Raja Koduri was director of advanced technology development at ATI from 2001 to 2009. Many of us older folk know of him because of that. He joined ATI to work on the Radeon R300,aka,the 9700 series and the follow up X800 series. This is the only time ATI/AMD managed to match/beat Nvidia in sales share in the last 20 years.He was also involved with the Radeon R700,aka, as the HD4000 series,and worked on the HD5000 series. He was responsible at Apple for the integration of retina displays,etc.
When he rejoined AMD in 2013,the company was at the verge of bankruptcy. They massively cut R and D,unlike Nvidia who increased R and D. Lisa Su admitted 70% of their R and D budget went towards Zen. At the same time the previous CEOs had locked AMD into a horrible agreement with Global Foundries,and they were paying loads in penalties because their CPU sales were going south and they couldn't use the agreed volumes.
This included pushing AMD GPUs onto an inferior GF 14NM process node,whilst Nvidia used a much better TSMC 16NM for Pascal. GF 14NM was horrible for larger die products.
Polaris was developed under him on a shoestring budget and propped up AMD consumer graphics for years. Vega was released firstly as compute card called Radeon Instinct.
Because of all the moaning AMD fans,they went re-released these cards to consumers,and had to make a loss on them. That is one of the biggest mistakes AMD made - they should have just not bothered,as it was a loss leader card.
Vega still lives on in cDNA and cDNA2. Vega OTH,did pretty well in AMD IGPs for years,beating off Intel for most of that time,despite AMD being behind in process node technology until Zen2. But since most of you would not even glance at an IGP,you wouldn't realise that.
RDNA1 was started when he was there,as it takes 3~5 years to develop a new uarch. RDNA2 in 2020 was an evolution of RDNA1,which appeared the next year. Dev kits for the PS5 already existed in 2019.
So at this point,some of this hate towards him seems more emotional,not based on the fact AMD simply had no money for years. Whether he was there,or anyone else was there,Nvidia simply spent more on R and D. The fact that AMD got even near is remarkable.
Regarding Intel - what about the rest of the company? Look at the massive amount of infighting in the company? They let go of 1000s of older,more experience workers:
People over 40 were two-and-a-half times more likely to lose their jobs in this spring's layoffs than Intel employees under 40.
www.oregonlive.com
They keep cancelling and changing CPU roadmaps,process node roadmaps,etc.They have screwed up on the process node rollout for the last 5+ years. Their CPU roadmap for servers is in a mess. Atom failed miserably.
They had years to work on the software side of things for their integrated graphics,but never bothered. Instead spent billions on all sorts of random purchases.
AMD was in a far worse situation,but it took Rory Read and Lisa Su to steady the ship. The problems at Intel stem from a lack of steady leadership at the very top for the last 5~8 years,and a lack of realistic goals.
If AMD with decades of experience,console contracts,strong links with devs,etc finds it incredibly hard to compete with Nvidia,how do people expect Intel with it's first one or two designs,with a history of not even caring about software,expect to match/beat Nvidia,let alone AMD?
It's not only the hardware but the whole software stack for 100s if not 1000s of games. If it was so easy,then look at the numerous Chinese dGPUs,some made on 7NM which have horrible performance due to poor drivers and lack of gaming support.
Apple has taken years to move to its own GPU uarch(they started moving towards it in 2014),and even then look at gaming performance? This is with a company that controls their whole ecosystem.
I remember back in the late 90s and early 2000s,when we had more dGPU companies out there,how software could make or break things.
Many here also have clearly forgotten the AMD WSA fiasco. They were locked into having to use inferior GF process nodes or pay huge financial penalties.Both Polaris and Vega were moved onto GF 14NM,which was noticeably worse than TSMC 16NM. Then on top of that 70% of AMD R and D was moved to develop the Zen CPU uarch.
This is why you saw only two Polaris dGPUs. The one behind the RX460(Polaris 11/21) was probably partially funded by Apple:
Spoiler: Don't expect to do much gaming on your shiny new MacBook Pro.
arstechnica.com
We never saw a Polaris design past Polaris 20/Polaris 30.
They literally released Vega 10 as the Radeon Instinct MI25 months before Vega 64,which was massively pushed past it's sweetspot. It was clearly never really intended as a gaming card,because they were loss leaders. Yet,the same Vega uarch is in the current Ryzen 5000 APUs. The current CDNA1 and CDNA2 uarchs are direct descendents of Vega.
RDNA3 might be the first design he had not much involvement with.RDNA1/RNDA2 were generally on better process nodes than Nvidia too. Now Nvidia is on a better process node than AMD,and we can see Nvidia pushing ahead more.
