Scott Herkelman Announces His Departure from AMD

[Vayra86]

I never understand the "I will miss you all and the great times" posts coming from someone who's leaving without a plan on where to go next.


Agreed.

Its PR, basically its a round of applause for oneself disguised as a round of applause for the company and everyone else. The world is fantastic, look at me being all positive and moving up in it.

Puke buckets do not suffice. I hate it, but I do it too whenever I move on... Though it might be good fun to write such an email the exact opposite way: "Hey guys, it was an absolute disaster working here, with you, in this company, on this assignment, so I'm off. Enjoy your continued misery! Bye"

 

[Deleted member 185088]

That's a silly concern, AMD has lost that battle a very long time ago. The Radeon division needs fresh leadership.

Indeed, when Intel cheated and went unpunished, capitalism à l'américaine.

 

[stimpy88]

It must be frustrating to work for the Radeon division. So much promise, but also so much amateurism when you get past the engineers.

I can't help but wonder, now that Intel has shown its cards, so to speak, that AMD CPU's are also about to look a little light on features soon. It's been a long time since they added anything of interest to their CPU's, such as a fast AI core, or other accelerators like Apple does, for instance. It's also time to leave dual channel memory behind on the desktop, but I don't see them doing that either.

They have all this great tech and enthusiasm, but then they start playing silly games by holding performance and features back, because Intel doesn't have them, instead of getting in there first, so that the market will adopt their standards. Well, Intel will soon overtake with features, if not performance, so Zen 5 better be something more than just an IPC increase, but I won't hold my breath, because AMD. They love to piss their advantage away.

 

[rbgc]

He knows it's going to be tough if the rumors are true AMD are pulling out of the high end market.

Future (next one or two generations) is mid range graphics card with low power consumption and improved FG for QHD and soon or later for 4K too. AMD, Intel, NVIDIA, all will go this way. It is unstoppable trend already. And NVIDIA will also make some 500W+ models to keep status of performance leader.

 

[Denver]

It must be frustrating to work for the Radeon division. So much promise, but also so much amateurism when you get past the engineers.

I can't help but wonder, now that Intel has shown its cards, so to speak, that AMD CPU's are also about to look a little light on features soon. It's been a long time since they added anything of interest to their CPU's, such as a fast AI core, or other accelerators like Apple does, for instance. It's also time to leave dual channel memory behind on the desktop, but I don't see them doing that either.

They have all this great tech and enthusiasm, but then they start playing silly games by holding performance and features back, because Intel doesn't have them, instead of getting in there first, so that the market will adopt their standards. Well, Intel will soon overtake with features, if not performance, so Zen 5 better be something more than just an IPC increase, but I won't hold my breath, because AMD. They love to piss their advantage away.

Are you calling AMD engineers amateurs? Lol Based on what? Oh God, People don't even know where to point their fingers these days. With the capital they have, AMD has played smart in the GPU division, skillfully weighing risk vs. return.

You would be surprised to know that the 7900xtx has potential close to the 4090, and the reason there is still a gap between them is more a software issue than a hardware issue: https://chipsandcheese.com/2023/09/...erformance-on-nvidias-4090-and-amds-7900-xtx/

 

[Hxx]

This so far from the truth, is ridiculous. You have This so far from the truth, is ridiculous. You have no ideea how a corporation functions and why there is this structure.

The decision making person is x1000 more important, even then the most talented engineer in the company.

Anyways, the rumours are he is going to Intel, so probably he is going to bring some talented force with him.
Hopefully Intel will become much more competitive with him and especially after the scammer Raja left.

ive been working in corporate for the past 10+ years in higher managerial positions (sure never an svp .. I wish ) but you wanna run that by me again how an svp all of the sudden is critical to innovation ? They’re not and of course some folks would think that , because of the PR bs around these roles

 

[wNotyarD]

It must be frustrating to work for the Radeon division. So much promise, but also so much amateurism when you get past the engineers.

I can't help but wonder, now that Intel has shown its cards, so to speak, that AMD CPU's are also about to look a little light on features soon. It's been a long time since they added anything of interest to their CPU's, such as a fast AI core, or other accelerators like Apple does, for instance. It's also time to leave dual channel memory behind on the desktop, but I don't see them doing that either.

They have all this great tech and enthusiasm, but then they start playing silly games by holding performance and features back, because Intel doesn't have them, instead of getting in there first, so that the market will adopt their standards. Well, Intel will soon overtake with features, if not performance, so Zen 5 better be something more than just an IPC increase, but I won't hold my breath, because AMD. They love to piss their advantage away.

Is Phoenix a joke to you? I mean, its tech is still just starting (for AMD) and its availability is sketchy, but its development will only go further.

 

[catulitechup]

we never forget you

 

[Dr. Dro]

Are you calling AMD engineers amateurs? Lol Based on what? Oh God, People don't even know where to point their fingers these days. With the capital they have, AMD has played smart in the GPU division, skillfully weighing risk vs. return.

You would be surprised to know that the 7900xtx has potential close to the 4090, and the reason there is still a gap between them is more a software issue than a hardware issue: https://chipsandcheese.com/2023/09/...erformance-on-nvidias-4090-and-amds-7900-xtx/

If you re-read what he said, the only people he left out of that remark are the engineers. It's a sentiment I echo.

Now, don't take that chips and cheese article as the 7900 XTX having similar potential to a 4090, that simply isn't true and you're delusional and lying to yourself if you sincerely believe that. The article itself states that Ada is woefully underutilized due to the way this specific engine is programmed. Which is far from efficiently, I may remark. I'm still trying to find the justification for the usage of insanely large registers such as these (that conveniently only fit on Navi 31 and 32, not even 33 supports them), but I'm sure there has been no influence from AMD, they just restrained themselves on paying off just to have DLSS out of the picture on launch and not a thing else...

AMD is moral, ethical and our FRIEND, after all

 

[AusWolf]

Its PR, basically its a round of applause for oneself disguised as a round of applause for the company and everyone else. The world is fantastic, look at me being all positive and moving up in it.

Puke buckets do not suffice. I hate it, but I do it too whenever I move on... Though it might be good fun to write such an email the exact opposite way: "Hey guys, it was an absolute disaster working here, with you, in this company, on this assignment, so I'm off. Enjoy your continued misery! Bye"

When I don't have anything positive to say, I try to say nothing. But each to their own, I guess. If I won the lottery, I'd write something positive about my boss and my team too. Otherwise, I'd just say where I was going to be transparent to the people who know me.

 

[Denver]

If you re-read what he said, the only people he left out of that remark are the engineers. It's a sentiment I echo.

Now, don't take that chips and cheese article as the 7900 XTX having similar potential to a 4090, that simply isn't true and you're delusional and lying to yourself if you sincerely believe that. The article itself states that Ada is woefully underutilized due to the way this specific engine is programmed. Which is far from efficiently, I may remark. I'm still trying to find the justification for the usage of insanely large registers such as these (that conveniently only fit on Navi 31 and 32, not even 33 supports them), but I'm sure there has been no influence from AMD, they just restrained themselves on paying off just to have DLSS out of the picture on launch and not a thing else...

AMD is moral, ethical and our FRIEND, after all

"RDNA 3 incorporates sizable cached register files (192 kB) to enhance shader occupancy and enable better VOPD usage. However, expanding VOPD without constraints would necessitate substantial modifications to all other resources, potentially making CUs 4-wide. While VOPD and related hardware changes are efficient and transistor-count-friendly, they rely on the compiler's ability to identify optimal VOPD usage

You just quickly glanced at the article. In short, AMD focused on creating an area-efficient design, like I said, playing smart. But it is more difficult to extract the full potential of the architecture than ADA, it requires more work on the software side. With proper use of the dual-issue FP32, the 7900xtx is very close to the 4090, the problem is that this is in the hands of the devs.

Finally, both are companies trying to grow and make a profit, however, no one will be able to deny that over the years, Nvidia has accumulated an extensive and obscure list of dominance tactics through control, sabotage and the imposition of closed solutions as a way of maneuvering the market. period.

 

[Dr. Dro]

"RDNA 3 incorporates sizable cached register files (192 kB) to enhance shader occupancy and enable better VOPD usage. However, expanding VOPD without constraints would necessitate substantial modifications to all other resources, potentially making CUs 4-wide. While VOPD and related hardware changes are efficient and transistor-count-friendly, they rely on the compiler's ability to identify optimal VOPD usage

You just quickly glanced at the article. In short, AMD focused on creating an area-efficient design, like I said, playing smart. But it is more difficult to extract the full potential of the architecture than ADA, it requires more work on the software side. With proper use of the dual-issue FP32, the 7900xtx is very close to the 4090, the problem is that this is in the hands of the devs.

Finally, both are companies trying to grow and make a profit, however, no one will be able to deny that over the years, Nvidia has accumulated an extensive and obscure list of dominance tactics through control, sabotage and the imposition of closed solutions as a way of maneuvering the market. period.

No, I've read that when it was initially posted. Hand optimizing for one architecture always yields the best results.

The cornerstone of a balanced architecture is one that plays nice with all kinds of workloads, not one that was created specifically targeting its strengths.

 

[rv8000]

No, I've read that when it was initially posted. Hand optimizing for one architecture always yields the best results.

The cornerstone of a balanced architecture is one that plays nice with all kinds of workloads, not one that was created specifically targeting its strengths.

Not that you can compare 1:1, but the 7900XTX is at a huge disadvantage when it comes to physical hardware in the shader dept. to where is surprising it can match a 4090 in any game; there are an handful in which it does.

Its also quite clear that when it comes to 1080p, the 4090 has some form of bottleneck in the majority of games, to where its often just a hair faster than a 4080; seems to hint at architectural inefficiency, which probably accounts for their fetish to cram upscaling down consumers throats.

Theres no way to guarantee a dev or GPU vendor is going to play fair, or if they were actually trying to favor one or the other. All companies do it, have done it, and will continue to want their products to be viewed in the best possible light. To single either out and not include the other is telling to say the least.

 

[stimpy88]

Are you calling AMD engineers amateurs? Lol Based on what? Oh God, People don't even know where to point their fingers these days. With the capital they have, AMD has played smart in the GPU division, skillfully weighing risk vs. return.

You would be surprised to know that the 7900xtx has potential close to the 4090, and the reason there is still a gap between them is more a software issue than a hardware issue: https://chipsandcheese.com/2023/09/...erformance-on-nvidias-4090-and-amds-7900-xtx/

Learn to read.

 

[Prima.Vera]

ive been working in corporate for the past 10+ years in higher managerial positions (sure never an svp .. I wish ) but you wanna run that by me again how an svp all of the sudden is critical to innovation ? They’re not and of course some folks would think that , because of the PR bs around these roles

Meant no disrespect. I only meant that if he is leaving, is possible he will bring a lot of other talented staff from AMD, including engineers.
Everybody is a mercenary in this shitty world.

 

[P4-630]

John Rayfield, Mastermind Behind Intel’s Next-Gen Meteor Lake VPU, Joins AMD

John Rayfield, Mastermind Behind Intel's Next-Gen Meteor Lake VPU, Joins AMD

Intel's Vice President and General Manager of Client AI, John Rayfield, has switched teams, as he embarks on his new journey at AMD.

wccftech.com

 

[AusWolf]

John Rayfield, Mastermind Behind Intel’s Next-Gen Meteor Lake VPU, Joins AMD

John Rayfield, Mastermind Behind Intel's Next-Gen Meteor Lake VPU, Joins AMD

Intel's Vice President and General Manager of Client AI, John Rayfield, has switched teams, as he embarks on his new journey at AMD.

wccftech.com

AMD is trying to get a foothold in the AI game, perhaps?

 

[wNotyarD]

AMD is trying to get a foothold in the AI game, perhaps?

I mean, isn't that the reason they purchased Xilinx? There's already XDNA in some APU's.

 

[stimpy88]

AMD is trying to get a foothold in the AI game, perhaps?

But isn't "AI" all about int16 and int8? Or do you mean the software side?

 

[AusWolf]

But isn't "AI" all about int16 and int8? Or do you mean the software side?

I mean, they just hired the guy behind Intel's client AI developments. They surely didn't hire him only to smile on a poster.

 

[stimpy88]

I mean, they just hired the guy behind Intel's client AI developments. They surely didn't hire him only to smile on a poster.

Good luck to AMD, I hope they carry on and help keep the cost of A.I. entry down. Personally, I only see A.I. being performant enough on GPU's (outside of custom A.I. hardware), but ideally it would be great for my $700 CPU to do it without needing a $1000 GPU as well.

 

[AusWolf]

Good luck to AMD, I hope they carry on and help keep the cost of A.I. entry down. Personally, I only see A.I. being performant enough on GPU's (outside of custom A.I. hardware), but ideally it would be great for my $700 CPU to do it without needing a $1000 GPU as well.

Regardless of performance, I just don't see much point in it on a home user level. I mean, OK, DLSS can help when your framerate is low, but so can FSR (without AI). Canvas is fun for 5 minutes, but that's it. ChatGPT revolutionises the ways people can get away with being stupid. Deepfake pictures, videos and influencers keep the content sludge flowing with minimal human involvement. Give it 5-10 years, and we won't only rewrite the past, but also the present with deepfake news. What else is there? I think that's it.

Even AMD graphics cards come with AI cores now, but I don't know about any single use case for them.

 

[Denver]

Regardless of performance, I just don't see much point in it on a home user level. I mean, OK, DLSS can help when your framerate is low, but so can FSR (without AI). Canvas is fun for 5 minutes, but that's it. ChatGPT revolutionises the ways people can get away with being stupid. Deepfake pictures, videos and influencers keep the content sludge flowing with minimal human involvement. Give it 5-10 years, and we won't only rewrite the past, but also the present with deepfake news. What else is there? I think that's it.

Even AMD graphics cards come with AI cores now, but I don't know about any single use case for them.

The best use of "AI" is in complex research and simulations, for us ordinary users it is just a silly joke.

 

[stimpy88]

Regardless of performance, I just don't see much point in it on a home user level. I mean, OK, DLSS can help when your framerate is low, but so can FSR (without AI). Canvas is fun for 5 minutes, but that's it. ChatGPT revolutionises the ways people can get away with being stupid. Deepfake pictures, videos and influencers keep the content sludge flowing with minimal human involvement. Give it 5-10 years, and we won't only rewrite the past, but also the present with deepfake news. What else is there? I think that's it.

Even AMD graphics cards come with AI cores now, but I don't know about any single use case for them.

You might see the point after the A.I. industry and its offerings are "regulated" in your country. It would then be nice to be able to run one locally, if you need to for business etc, would it not?

 

[AusWolf]

You might see the point after the A.I. industry and its offerings are "regulated" in your country. It would then be nice to be able to run one locally, if you need to for business etc, would it not?

I don't have a business. I honestly have no idea what I'd ever use AI for.