Jim Keller joins Samsung

[Madpacket]

How did you ever hear about Keller compared to all the ones we dont hear about?

Nobody here cant really tell why Keller should be better. Besides he was one of the people working on K7, K8, A4 and A5.

Exactly, that's the point. You don't hear of other amazing engineering chip designers because they don't have the track record Keller has. The dude's a legend and has been a pioneer in the field since the DEC days. He's going on almost 20 of microarchitecture engineering and is still innovating. Samsung is lucky to get him. Probably exactly what they need to pull and stay ahead of Qualcomm. I first heard about Keller back when Apple bought the company he worked for (PA - Semi). He's done a lot more than lead the design of the K7/8 and A4/5 chips.

Just look at some of his accomplishments.

From Wikipedia:

"Jim Keller is a microprocessor engineer most well known for his work at AMD and Apple. He was the lead architect of AMD K8 microarchitecture (including the original Athlon 64) and was involved in designing the Athlon (K7) and Apple A4/A5 processors. He was also the coauthor of the specifications for the x86-64 instruction set and HyperTransport interconnect. From 2012 to 2015, he returned to AMD to work on the AMD K12 and Zen microarchitectures, particularly on the low-power/embedded

https://en.wikipedia.org/wiki/Jim_Keller_(engineer)

 

[AtenRa]

I bet you that not a single person here waxing lyrical about the greatness that Jim Keller is even knew the man existed before AMD put out a PR touting that it had hired him back in 2012.

By the same logic, nobody knew people left AMD the last two-three years or earlier.

EDIT: to be clear, I'm not trying to "put down" Keller -- his career history points to somebody who is extremely good at what he does. I'm just trying to say that the people saying "OMG RIP ANYBODY COMPETING WITH KELLER" grossly underestimate the talent pool within the very large and vibrant semiconductor industry.

Agreed

 

[ShintaiDK]

Exactly, that's the point. You don't hear of other amazing engineering chip designers because they don't have the track record Keller has. The dude's a legend and has been a pioneer in the field since the DEC days. He's going on almost 20 of microarchitecture engineering and is still innovating. Samsung is lucky to get him. Probably exactly what they need to pull and stay ahead of Qualcomm. I first heard about Keller back when Apple bought the company he worked for (PA - Semi). He's done a lot more than lead the design of the K7/8 and A4/5 chips.

So due to the sole reason Keller got into the media. We can conclude he is the best in the world? People being in all other MPUs are just average nobody?

You dont see the problem here do you? Can you even name lead designers on other MPUs? Conroe for example? A9? Bobcat? Others?

 

[Fjodor2001]

Besides he was one of the people working on K7, K8, A4 and A5.

Are you sure about that? Previously you have said that AMD just bought the K8 design from an external company:

K8 was a bought design from the outside.

 

[desura]

This makes me wonder about what happened at AMD in regards to Zen.

It would seem to me that if he did a good job with Zen then he would simply take a break for a little bit. Him landing up at Samsung so soon makes me think that he left a sinking ship or was poached before he had completed all of the work.

 

[ShintaiDK]

This makes me wonder about what happened at AMD in regards to Zen.

It would seem to me that if he did a good job with Zen then he would simply take a break for a little bit. Him landing up at Samsung so soon makes me think that he left a sinking ship or was poached before he had completed all of the work.

It was obvious he left unexpected with no replacement ready.

 

[desura]

It was obvious he left unexpected with no replacement ready.

I don't think that is obvious.

He doesn't seem to be the type to be very mercenary, but rather is someone who takes on projects because they are interesting. I dunno, but joining Samsung so soon instead of say, spending six months in the Caribbean, is suspicious.

 

[ShintaiDK]

I don't think that is obvious.

He doesn't seem to be the type to be very mercenary, but rather is someone who takes on projects because they are interesting. I dunno, but joining Samsung so soon instead of say, spending six months in the Caribbean, is suspicious.

If his leave from AMD was planned. Then they would have had a replacement.

And joining Samsung so fast as well leaves the impression that something was wrong at his former employer.

Else it could be excused for personal matters etc.

 

[frozentundra123456]

Who's a better chip designer than Keller?

Suddenly I hear the sound of crickets...

Do you personally know every chip designer that works for Samsung, Apple, Intel, nVidia, etc? I am not saying that Keller is great or not, but to infer that he is the best chip designer simply because he is the most well known is absurd. There may be, and probably are, tens or hundreds of competent chip designers working behind the scenes that no one knows about that may be just as competent. Nobody really knows, but that is the point.

To be honest, I think he could have a big impact on a smaller, struggling company like AMD. But in a huge, already successful company like Samsung I think the impact of a single engineer will be much less.

 

[Phynaz]

I don't think that is obvious.

He doesn't seem to be the type to be very mercenary, but rather is someone who takes on projects because they are interesting. I dunno, but joining Samsung so soon instead of say, spending six months in the Caribbean, is suspicious.

Could you spend six months in the Caribbean between jobs? What makes you think he could?

 

[thesmokingman]

Keller will have an impact wherever he goes. And he likes to be a force wherever he is, then move on. He was behind the K8/K7, AMD64 projects, then moved onto Apple A4/5 which is killing it in the mobile space. He came back to AMD, to be a key figure behind Zen. Now with that done, he's off to achieve awesomeness at Samsung? It looks to me like whomever is employing Keller is in a positive position for the future. The man however seems to like to tackle new projects, break new ground and is not the type to sit idle. Btw, Samsung signing Keller looks more like a respoonse to his time at Apple.

 

[Dresdenboy]

So due to the sole reason Keller got into the media. We can conclude he is the best in the world? People being in all other MPUs are just average nobody?

He's not the best. But he was a good fit. Some personalities are better suited for specific tasks or roles than others and vice versa. And his role might not need a followup. If he was there to bring in his experience, ideas, visions, and to clean up the development processes (after seeing many others before), then he was like a high level consultant similar to Andy Glew before. If the consultant did a good job, he's not needed for too long.

You dont see the problem here do you? Can you even name lead designers on other MPUs? Conroe for example? A9? Bobcat? Others?

OK, without using the Alphabet the first is easy for me: David Ditzel (now at Intel).
A9: Dunno..
Bobcat: (names are one of my weaknesses).. not Mike Butler (BD), but Brad Burgess is one of the important guys I think. He left to Samsung IIRC.

 

[Arachnotronic]

He's not the best. But he was a good fit. Some personalities are better suited for specific tasks or roles than others and vice versa. And his role might not need a followup. If he was there to bring in his experience, ideas, visions, and to clean up the development processes (after seeing many others before), then he was like a high level consultant similar to Andy Glew before. If the consultant did a good job, he's not needed for too long.


OK, without using the Alphabet the first is easy for me: David Ditzel (now at Intel).
A9: Dunno..
Bobcat: (names are one of my weaknesses).. not Mike Butler (BD), but Brad Burgess is one of the important guys I think. He left to Samsung IIRC.

Ditzel left Intel.

 

[Dresdenboy]

Ditzel left Intel.

Oh, thanks. Seems to be common.

 

[dawheat]

Do you personally know every chip designer that works for Samsung, Apple, Intel, nVidia, etc? I am not saying that Keller is great or not, but to infer that he is the best chip designer simply because he is the most well known is absurd. There may be, and probably are, tens or hundreds of competent chip designers working behind the scenes that no one knows about that may be just as competent. Nobody really knows, but that is the point.

To be honest, I think he could have a big impact on a smaller, struggling company like AMD. But in a huge, already successful company like Samsung I think the impact of a single engineer will be much less.

That guy Samsung poached/stole/bribed from TSMC certainly seemed to be worth his weight in gold. Still takes a world class team and tech to turn it into real products so quickly, but the right person at the right team can make a huge difference.

That being said, Samsung's upcoming mobile SOC already looks pretty amazing and right up there with the A9. Really wonder what he'd be focusing on.

 

[Arachnotronic]

That being said, Samsung's upcoming mobile SOC already looks pretty amazing and right up there with the A9. Really wonder what he'd be focusing on.

Other than a few made up "leaks" on Weibo, what evidence do we have that Samsung's upcoming SoC is "pretty amazing"?

 

[ShintaiDK]

Other than a few made up "leaks" on Weibo, what evidence do we have that Samsung's upcoming SoC is "pretty amazing"?

Engineers dancing in the aisles?

 

[Andrei.]

Other than a few made up "leaks" on Weibo, what evidence do we have that Samsung's upcoming SoC is "pretty amazing"?

Chose what you want to believe but there's been plenty of pointers. Also a tip: Just because things get re-posted on Weibo doesn't make them "made up".

Engineers dancing in the aisles?

Couldn't have said it better. No joke.

The whole discussion reminds me of people brushing off the 14nm Exynos reports this time last year.

 

[Arachnotronic]

Chose what you want to believe but there's been plenty of pointers. Also a tip: Just because things get re-posted on Weibo doesn't make them "made up".

Those sources totally missed the mark on the A9, which significantly dents their credibility.

I don't suppose you can point us to something that's available publicly, can you?

 

[Andrei.]

Those sources totally missed the mark on the A9, which significantly dents their credibility.

I don't suppose you can point us to something that's available publicly, can you?

Rumours aren't from a single source and whom you're referring to on Weibo isn't a source at all but just a guy who aggregates/reposts stuff. I have no idea where his A9 numbers came from but I do know where the M1 stuff came from and that one has a clean track record.

In any case what we'll see in the 8890 will be the fruit of a lot of investment and many smart people that Samsung has been gathering up over the last 4-5 years, and from what I hear the SLSI guys have grins as wide as their faces right now. The question that I'm most looking forward to get an answer on is what the mobile division will do in the S7 and how they'll choose to segment the models.

 

[Arachnotronic]

Rumours aren't from a single source and whom you're referring to on Weibo isn't a source at all but just a guy who aggregates/reposts stuff. I have no idea where his A9 numbers came from but I do know where the M1 stuff came from and that one has a clean track record.

Which M1 stuff? The one that claims GB3 ST/MT of 2136/7497 or the later one that claimed 2296/6908?

The reason that I am skeptical is that the leaks that showed these purported scores were paired with scores of other unreleased chips that were frankly way off the mark. Bad numbers for the A9/A9X, Snapdragon 820, etc.

Also, the best Exynos 8890 that I see in the GB3 database is this: https://browser.primatelabs.com/geekbench3/3397836

ST/MT: 1446/4769.

I just feel like these numbers from the "leaks" come from multiplying ~1400 by (2.3/1.4) to get ~2300.

In any case what we'll see in the 8890 will be the fruit of a lot of investment and many smart people that Samsung has been gathering up over the last 4-5 years

It'll be interesting to see what they come up with.

 

[Andrei.]

Also, the best Exynos 8890 that I see in the GB3 database is this: https://browser.primatelabs.com/geekbench3/3397836

ST/MT: 1446/4769.

I just feel like these numbers from the "leaks" come from multiplying ~1400 by (2.3/1.4) to get ~2300.

And here's a set of numbers showing S820 isn't any better than a A53: http://browser.primatelabs.com/geekbench3/3836517

It doesn't mean that'll end up like that, and there's of course better scores out. Prototypes running at capped frequencies for whatever reason (likely not wanting to leak full perf). It's not hard to imagine the same happening to the A9 scores.

 

[Arachnotronic]

And here's a set of numbers showing S820 isn't any better than a A53: http://browser.primatelabs.com/geekbench3/3836517

It doesn't mean that'll end up like that, and there's of course better scores out. Prototypes running at capped frequencies for whatever reason (likely not wanting to leak full perf). It's not hard to imagine the same happening to the A9 scores.

So are you willing to make a hard claim with respect to the kind of performance we will see from the Exynos 8890 in terms of CPU, GPU, etc.?

 

[Arachnotronic]

The question that I'm most looking forward to get an answer on is what the mobile division will do in the S7 and how they'll choose to segment the models.

Well if Exynos 8890 is as great as you say, then there is no point in Samsung Mobile doing anything but using 8890 across the board as it did with 7420.

 

[Andrei.]

So are you willing to make a hard claim with respect to the kind of performance we will see from the Exynos 8890 in terms of CPU, GPU, etc.?

I'm just saying that I have yet to hear anything actually going against those high performance claims.

Well if Exynos 8890 is as great as you say, then there is no point in Samsung Mobile doing anything but using 8890 across the board as it did with 7420.

I have a lot less faith in the mobile division making sensible decisions. The 7420 situation came to be because the S810 was worse by a considerable margin and it wouldn't have been acceptable to use both because one model variant would have been clearly much better. It all depends if the S820 efficiency ends up within reach of the 8890.