Fox2232:

He apparently has some operational knowledge and keeps his primarily compute aspirations. I think intel got him to make good compute GPU, not gaming one. Intel still keeps almost entire CPU server business. But they have entire GPU server business to take.

Quite offtopic but I have to say it: that's exactly what he was hired for. Anyone expecting a (good) gaming GPU from Intel is nuts, they just want to bite into the server compute and cryto markets. (No sources, just speculation but damn I would bet a paycheck to it any day of the week)

I think the biggest problem with Vega is that AMD fell into the same trap Intel did with the Pentium 4 processors. They increased the length of the instruction pipeline in order to ramp clock speeds but overestimated how far this would allow them to push the clock speed before power became a problem. This resulted in a reasonable bump in operating frequencies, but the lower IPC of the longer pipeline offset some of the clock speed gain, which ultimately left Vega performance falling short of expectations. The Vega memory interface also dropped from 512 to 256 bits, resulting in slightly lower bandwidth compared to the previous Fury HBM interface. I think it would have been interesting to see how the Polaris architecture would scale compared to Vega if AMD had released a 4096:256:64 HBM 4th gen GCN version i.e. basically a 4th gen Fury X. I suspect it would have been pretty close to the performance of the Vega 64 and it could have probably been brought to market significantly sooner than Vega. This would have allowed AMD to compete with the 1080 GTX at launch rather than leaving Nvidia with no competition at the top end for almost 1.5 years.

schmidtbag:

Does Martin Ashton drive an Aston Martin? 😛 Anyway, pretty weird to see both companies trading employees like this. Though, no offense to him but I think Raja was more valuable to AMD than Martin was to Intel.

Not sure why but your joke reminds of this quote : How many Lowes would Rob Lowe rob if Rob Lowe could rob Lowes?

For a moment, I was dead sure Hilbert was smoking some good stuff and had inadvertently typed Martin Ashton instead of Aston Martin....:D

schmidtbag:

I agree with all of this. Though, I think you may be slightly underestimating the amount of changes that went into Vega. The reason I say this is because the driver stability was pretty atrocious on release day (specifically, for Vega FE). Before someone jumps in saying "hurr durr AMD always makes crappy drivers on release day" this was no ordinary situation. If an architecture is mostly the same, the drivers ought to have been much more stable, since I imagine most of it would just be copy+paste. But, maybe I'm wrong.

I remember there was quite some list of changes for Vega. Quite a few of them should have given improved gaming performance per clock. We do not get to see them now. In the end it is even hard to argue for existence of HBCC. I am sure those changes are present in each Vega chip. And maybe one day AMD figures that some of them can be actually used. But now, it is already too late for Vega. (Still would be nice for Vega owners to get some performance boost.) Now, they are like dead weight.

schmidtbag:

I'm interested in what you feel those hints are. Knowing you, your thoughts tend to be more focused on evidence and context than most people. That doesn't mean I'm going to agree, but given when Embra said, all we can do is guess, so I see this as nothing more than a friendly debate.

1st big blow to Raja's image is above. 2nd is way he left. When you are large corporation and want to get rid of someone who you do not want to see again in office... You give him 2 choices. Involuntary termination on hour notice. (Damage to career. But potential lawsuit against company.) Or he signs paper where he agrees with termination and gets paid vacation for contracted release period. (2~3 months depending on type of contract/country.) Here it looked like they did not want to see him in office again. Maybe he created some conflict as Vega was presented and above things got revealed. 3rd is 3rd iteration of Polaris. If it comes, It will show few things: - clock bump => were there really need for Vega to get 10% clock above already released Polaris or was it just due to more mature manufacturing process and tiny bit of tweaking? - Pushing forward quite old Polaris instead of doing 1/2 Vega (32CU+1x 8GB HBM2). Polaris here is 36CU and costs 5.7B transistors. Vega at 32CUs would be like 6.6B transistors and it would be weaker.

Fox2232:

And maybe one day AMD figures that some of them can be actually used. But now, it is already too late for Vega. (Still would be nice for Vega owners to get some performance boost.) Now, they are like dead weight.

Since the new APUs use Vega, I don't think AMD is going to be abandoning them quiet so soon. I would say progress is being made for the drivers, just not as much as we'd all have wanted. The Linux drivers in particular seem to be a bit held back. That being said, once next-gen GPUs come out, I think Vega will be largely left behind (at least on Windows).

1st big blow to Raja's image is above. 2nd is way he left.

I would like to clarify: I totally understand that Vega tarnished his image, since the project was a failure and it was his name as the head of it. However, my point is that's not really a fair assessment of him, given the info (assuming you believe it). Pointing fingers at someone for a problem outside of their control isn't really fair. I also understand that him leaving shortly after sure didn't help his reputation, but I think it's a little more complex than how you propose:

When you are large corporation and want to get rid of someone who you do not want to see again in office... You give him 2 choices. Involuntary termination on hour notice. (Damage to career. But potential lawsuit against company.) Or he signs paper where he agrees with termination and gets paid vacation for contracted release period. (2~3 months depending on type of contract/country.) Here it looked like they did not want to see him in office again. Maybe he created some conflict as Vega was presented and above things got revealed.

I can understand where you're coming from, but there are a couple other things to consider: 1. To my understanding, Raja was pretty much immediately hired by Intel after his departure. This muddies the water a little, because whether AMD encouraged him to leave or not, it is apparent that Intel wanted him on their team before he was officially departed. Intel wouldn't hire someone they felt wasn't capable of handling a task that they have repeatedly fallen short on before. 2. As with many presidents or VPs, they often function as a singular human summary of whatever it is they represent. What this means is, regardless of how much they're actually responsible for whatever happened (good or bad), they're the ones who will take the hit to save the organization. In other words, AMD may not actually think Vega was Raja's fault, but, they're not going to take the blame for it, so it's easier to pin it on him. I say this because (from what I recall) Lisa spoke somewhat kindly of him when he left.

3rd is 3rd iteration of Polaris. If it comes, It will show few things:

I got the impression that Polaris was initially meant for lower performance tier stuff while Vega was supposed to be for high end. Due to Vega's disappointment, it is possible AMD may have adjusted to work around this. Regardless, I see how this is relevant to the lack of Vega's shortcomings, but I'm not sure I understand how this is relevant to Raja.

Gotta love people claiming VEGA was a fail. AMD was running on pennies and underfunded GPU team wasn't their main priority. This is like having a work truck (CPU team) and a Corvette (GPU team) and choosing to only fix one, which one is more logical to focus on ???? I have faith with AMD ( mostly Lisa Su ) I agree yes maybe they would of been a failure if they put 100% of all their R&D and man power on Vega and got these kind of results. I believe the GPU division got maybe 25% of the resources while the CPU division got most of the attention and resources. We will see the true potential of AMD's GPU team in the coming 1-2 years. They are making $$$$$$$$$ and not going bankrupt so I have good confidence in them. We should be thankful they're still operating. AMD is the only tech company that ran against the giant (shady INTEL) and ran on peanuts and still is alive today. This company is a true underdog.

Dimitrios1983:

Gotta love people claiming VEGA was a fail. AMD was running on pennies and underfunded GPU team wasn't their main priority. This is like having a work truck (CPU team) and a Corvette (GPU team) and choosing to only fix one, which one is more logical to focus on ???? I have faith with AMD ( mostly Lisa Su ) I agree yes maybe they would of been a failure if they put 100% of all their R&D and man power on Vega and got these kind of results. I believe the GPU division got maybe 25% of the resources while the CPU division got most of the attention and resources. We will see the true potential of AMD's GPU team in the coming 1-2 years. They are making $$$$$$$$$ and not going bankrupt so I have good confidence in them. We should be thankful they're still operating. AMD is the only tech company that ran against the giant (shady INTEL) and ran on peanuts and still is alive today. This company is a true underdog.

Take it to extreme and you'll know that you are pushing for nonsense... We did not have enough funds to make any changes, so here is exactly same GPU on exactly same manufacturing process. Please, call it success. Over promising and under delivering on technology is failure. Especially for person responsible to make realistic estimations and then deliver on those realistically promised features. If Vega was tiny bit behind schedule and all features could have been completed in few months, AMD would not force release of unfinished GCN iteration. If it was considered as complete and afterwards proved to be buggy on HW level and needed disablement of those features... But result is same. Unrealistic promise or failure to control delivery.

Fox2232:

Take it to extreme and you'll know that you are pushing for nonsense... We did not have enough funds to make any changes, so here is exactly same GPU on exactly same manufacturing process. Please, call it success. Over promising and under delivering on technology is failure. Especially for person responsible to make realistic estimations and then deliver on those realistically promised features. If Vega was tiny bit behind schedule and all features could have been completed in few months, AMD would not force release of unfinished GCN iteration. If it was considered as complete and afterwards proved to be buggy on HW level and needed disablement of those features... But result is same. Unrealistic promise or failure to control delivery.

Shrinking the GPU did manage to make it run way cooler and use less electricity. I do agree that the AMD RX 560 has all shader units "unlocked" but heard the RX480 bios hack can unlock the shaders to the same RX560. The majority of gamers with 570 & 580 cards should be enough. I understand most people on these forums love cutting edge technology and the best of the best hardware where money is not a deciding factor. It's like me commenting my Camaro LS1 being enough power in a forum where newer LS engines pump out 750HP stock and push them over 1500HP with newer Camaro's and Corvette's.