[Tweaktown] Jim Keller leaving Tesla to join Intel

[coercitiv]

It's almost as if we found out about Keller just in time to compensate for the official 10nm bad news.

 

[moonbogg]

Keller is competing with himself, LMAO

He's arming both sides. He is the great tech arms trader of the 21st century. His skills are for hire, but knowledge of the consequences he does not desire.

 

[NostaSeronx]

Will he end up working on this?

Cove based cores are from the Bulldozer legacy. Sunnycove(10nm) -> WLcove(7nm?) -> Oceancove(5nm?) are the cores.

There is also a project of utilizing Softmachine's VISC for everything on the die. GPU is getting it first with HW-accelerated(Fixed Function) Jitter translation. It is heavily based on the Intel IR(Intermediate language/intermediate representation) in their compilers.
-> https://software.intel.com/en-us/articles/code-generation-options-for-intelr-graphics-technology
By default, the compiler will generate the Virtual ISA instructions. <-- IR/IL

 

[Nothingness]

He's arming both sides. He is the great tech arms trader of the 21st century.

You meant ARMing and ARMs?

 

[CHADBOGA]

So I guess in 3-4 years Intel will finally have a competitive architecture to Ryzen?

Is having superior IPC and clock rate capability not already more than competitive?

 

[Abwx]

Is having superior IPC and clock rate capability not already more than competitive?

They have slighty higher IPC because they use clustered execution ports but intrinsicaly Zen has higher IPC capabilities than Intel s current uarches, that s what is necessary to compete when your opponent is the one who decide of the new instructions implementations..

 

[Zucker2k]

Is having superior IPC and clock rate capability not already more than competitive?

He is clearly trolling.

 

[Abwx]

He is clearly trolling.

He does not, if Intel were sure that their uarch is better they wouldnt have hired Keller, they have supposedly enough talented people...

 

[Nothingness]

He does not, if Intel were sure that their uarch is better they wouldnt have hired Keller, they have supposedly enough talented people...

What makes you think Keller will work on a CPU project and if so on an x86 CPU project?

 

[tamz_msc]

What makes you think Keller will work on a CPU project and if so on an x86 CPU project?

Given how BK was talking about Intel's plans with their own '7nm'(full-blown EMIB for starters), it's safe to say that Keller will be involved in CPU one way or the other.

 

[Abwx]

What makes you think Keller will work on a CPU project and if so on an x86 CPU project?

Because it s what he s good at, and X86 or anything other is of no importance, we know that AMD s Zen can be easily ported to ARM if necessary, and surely to any other ISA.

 

[Nothingness]

As far as I know, Keller was doing SoC design at Apple. And not CPU at Tesla.

He might very well work at SoC level again at Intel. Re-read Intel announcement and note some words such as "heterogeneous", "GPUs, accelerators", on top of CPU. And this:

Intel today announced that Jim Keller will join Intel as a senior vice president. He will lead the company’s silicon engineering, which encompasses system-on-chip (SoC) development and integration.

 

[Zucker2k]

Core is already the dominant uarch - core for core. Obviously, Intel, just like any dominant company in any sector must plan long term if they want to remain in that position. Core has been around for ages now and it's still dominant. Yes, you'll find code in which Ryzen does better, but both the ipc and MHz are on the superior uarch's side, and that is Core!

 

[TheELF]

They have slighty higher IPC because they use clustered execution ports but intrinsicaly Zen has higher IPC capabilities than Intel s current uarches, that s what is necessary to compete when your opponent is the one who decide of the new instructions implementations..

Wait IPC still means same amount of cores at the same clocks right?
The only CPUs that fit are the
2200g 4c at 3,5Ghz with a max boost of 3.7 and the
i3-8100 4c at 3.6Ghz locked
46% to 67% doesn't really look slight.
https://www.pcgamer.com/amd-ryzen-7-2700x-is-better-than-its-predecessors-in-every-way/

 

[IRobot23]

I was also wondering about that...

2200G is on i3 8100 lvl
2400G is above i7 4770K (About MT 2400G will score near i5 8400).

i7 8700 runs higher clocks on all core and has 6C while i7 7900X has 10C. So basically both at same clocks and optimal non-realistic scaling would be 166% performance of i7 8700. They got 56% boost despite huge difference in clockspeeds.

R5 1600X could be at max 50% faster than 2400G... 60%.

 

[tamz_msc]

Wait IPC still means same amount of cores at the same clocks right?
The only CPUs that fit are the
2200g 4c at 3,5Ghz with a max boost of 3.7 and the
i3-8100 4c at 3.6Ghz locked
46% to 67% doesn't really look slight.
https://www.pcgamer.com/amd-ryzen-7-2700x-is-better-than-its-predecessors-in-every-way/

IPC tests are best performed using 1T at identical frequency, like the Stilt does. PCGamer's test suite is horrible. y-cruncher? Might as well try Linpack. Civ 6 AI test clubbed together under multi-threaded geometric mean? They're embarrassing themselves. It's because of crap like this that people like you exist whose sole job is to add fuel to the fire using cherry-picked "benchmarks" with horrible methodology.

 

[Abwx]

Wait IPC still means same amount of cores at the same clocks right?
The only CPUs that fit are the
2200g 4c at 3,5Ghz with a max boost of 3.7 and the
i3-8100 4c at 3.6Ghz locked
46% to 67% doesn't really look slight.
https://www.pcgamer.com/amd-ryzen-7-2700x-is-better-than-its-predecessors-in-every-way/

Compare what is comparable, not L3 cache starved 2200G vs i3...

What about comparing 6C/12T, say the 2600X and the 8700K, knowing that the latter has roughly 20% higher clocks on 6C loaded..?.

 

[coercitiv]

2600X vs. 8700 at fixed 4Ghz

 

[TheELF]

IPC tests are best performed using 1T at identical frequency, like the Stilt does. PCGamer's test suite is horrible. y-cruncher? Might as well try Linpack. Civ 6 AI test clubbed together under multi-threaded geometric mean? They're embarrassing themselves. It's because of crap like this that people like you exist whose sole job is to add fuel to the fire using cherry-picked "benchmarks" with horrible methodology.

So what is your point here? If they have the same IPC what does it matter what they run?
How can you cherry pick if there isn't something that runs slower or faster?

Compare what is comparable, not L3 cache starved 2200G vs i3...

What about comparing 6C/12T, say the 2600X and the 8700K, knowing that the latter has roughly 20% higher clocks on 6C loaded..?.

The 2400g 4c/8t (3.6 3.9Ghz turbo) has the same amount of cache,the same 4Gb for twice the amount of threads,compare the 58% it get's to the 1800x 8c/16t (3.6 4Ghz turbo) 122,cache has zero effect here.

 

[TheELF]

What about comparing 6C/12T, say the 2600X and the 8700K, knowing that the latter has roughly 20% higher clocks on 6C loaded..?.

https://www.gamersnexus.net/guides/3077-explaining-coffee-lake-turbo-8700k-8600k
6C: 4.3GHz ,a whole whopping 100Mhz more then the 2600x,get your facts straight.

 

[Abwx]

The 2400g 4c/8t (3.6 3.9Ghz turbo) has the same amount of cache,the same 4Gb for twice the amount of threads,compare the 58% it get's to the 1800x 8c/16t (3.6 4Ghz turbo) 122,cache has zero effect here.

Ryzen 1500X has higher IPC than a R5 1400 or a 2400G thanks to its 16MB cache...

Otherwise i wont bother answering you anymore, your 45% are a signal that say "troll here"....

 

[William Gaatjes]

Yes, but the tricky part is the knowledge and trade secrets involved in his work at the previous companies. While there is still a burden of proof involved, one could make the claim that knowledge/secrets from Company A were used to create the new/derivative work at Company B. A rather blatant case of this being Uber's legal scuffle with Waymo, in which they ended up settling with Waymo to the tune of $244 million over allegedly stolen code (Uber got off lightly, IMO).

I am wondering about that.
Is it not so that designers also hold patents of their own and because they have these patents they are free from litigation when using these patents and enforcing a contract that they are free to use their patents ?
I wonder if Jim Keller has a large patentfolio when it comes to cpu design.

 

[Hitman928]

I am wondering about that.
Is it not so that designers also hold patents of their own and because they have these patents they are free from litigation when using these patents and enforcing a contract that they are free to use their patents ?
I wonder if Jim Keller has a large patentfolio when it comes to cpu design.

Almost universally in my experience, when you create something that you patent while working at a company, the company owns the patent. Your name gets put on the patent which is great for your resume/reputation, but the company owns the IP. Could someone as valuable as Keller work out a deal where he retains ownership of any IP he comes up with? Possibly, but I highly, highly doubt it. The only time I've seen this happen is when a small to maybe medium sized business hires someone with a lot of experience as an independent contractor for a short time to help them with a specific project (because they can't afford to hire them as a full time employee).

 

[majord]

Is having superior IPC and clock rate capability not already more than competitive?

Neither is the only metric to determine what is a better overall architecture, and the later (freq) is only partially determined by the architecture , the other factors are phy layout and Process performance.

So with that in mind, looking at the products themselves, From a performance/watt perspective (particularly throughput/watt), Zen+ is easily a match if not better than CFL. Core size is also smaller so perf/mm2 it is ahead. Frequency is only marginally behind now also.

Common knowledge/consensus is that Intel's 14nm+/++ is still the superior process vs GF 12nmLP , both density and performance - perhaps significantly so, yet the end result does not reflect this. So either accept the Zen architecture, on many metrics is actually superior - or be brave, and try to argue Intel's process sucks. You pick!

 

[tamz_msc]

So what is your point here? If they have the same IPC what does it matter what they run?
How can you cherry pick if there isn't something that runs slower or faster?

PCGamers' tests are the last place I'd look at to compare IPC. Unless explicitly set at identical frequencies, there is no point in guessing at IPC from benchmarks like PCGamers like you were doing. Also, the use of geometric mean skews results which are purely due to lack of things like AVX2 and AVX512 on Ryzen against AMD's favor.