Jim Keller leaves AMD

[frozentundra123456]

Are we discussing the right use cases?

http://www.newegg.com/Product/Produ...STMATCH&Description=A10-7850K&N=-1&isNodeId=1 lists related PCs as HTPC at ~$300-$400.

Dont think those include a windows license, so add a hundred bucks extra.

 

[DrMrLordX]

I think Bristol Ridge for mobile being released early would be a good thing.

Yes, we know.

So why are we discussing this in a thread about Jim Keller? Can we stop now?

 

[USER8000]

Factor in DDR3 2400 RAM for the A10-7850K and now its $2 more expensive (going by Newegg prices) than Athlon x 4 860K (using DDR3 1600) + GT 730 GDDR5.

And remember this is for Kaveri which is clocked at only 720 Mhz compared to Godavari A10-7870K which is 866 Mhz.

Or you could get an A8 7600 for £65 to £70,as opposed to the X4 860K and GT730 GDDR5 which will cost around £105 to £115.

 

[cbn]

I think Bristol Ridge for mobile being released early would be a good thing.

Yes, we know.

So why are we discussing this in a thread about Jim Keller?

You can read the following post and get the idea on how it started:

http://forums.anandtech.com/showpost.php?p=37995017&postcount=399

But here is a summary in a nutshell:

1. AMD focuses on mobile with APUs (leaving only the "undesirable for mobile" harvested APU parts for AM4 desktop)

2. AMD brings back low end desktop dGPUs and competes with Nvidia again (And yes DrMrLordX, Nvidia is launching new low end dGPUs. See recent GT 710 and rumor of GT 930. Low end desktop dGPUs won't be going anyway as not everybody has brand new high end iGPU for their desktop)

3. With the return of affordably priced low end AMD dGPUs Radeon becomes more ubiquitous (instead of being exclusive)...this should help Linux development for AMD.

4. With Linux development improving for AMD, the K12 APU should have a better graphics driver and stands a higher chance of launching (or being successful when it launches).

With that mentioned, seeing K12 APU successfully launch is not my motive for Bristol Ridge being focused on mobile. I believe that 4C + 512sp and 4C + 384sp (basically most of the Bristol Ridge dies) focused on mobile is good for AMD even if K12 never launches as an APU.

 

[senseamp]

I hope Tesla does some chips. Good for hardware engineers in the Valley

 

[cbn]

These are 28nm parts on a fully mature process. They are not supply constrained. (In fact, AMD's WSA gives them an incentive to produce and sell parts even at low margins - better than paying the penalties and getting nothing at all in return.) They are not yield constrained.

Speaking of 28nm as a mature process and "not yield constrained" AMD needs to do justice by the dual core desktop APUs.

At this point there should be no reason for the newly releaed A6-7470K having only 256sp:

http://www.anandtech.com/show/10009/amd-launches-excavator-on-desktop-the-65w-athlon-x4-845-for-70

Even the Intel Pentium G4500 has GT2 graphics now (which should be equivalent to the A6-7470K's iGPU), but with a much stronger CPU. Even against the lower Pentium G4400, A6-7470K just doesn't have the CPU to compete.

AMD should have at least provisioned the new dual core with 384sp (and even better would have been 448sp or 512sp).

P.S. Regarding dual core performance with a big iGPU, I've tested my Athlon x 4 860K with one module disabled using GT 730 GDDR5 (which is 7.7% faster in Firestrike than A10-7870K iGPU with dual channel DDR3 2133) and it doesn't suffer stuttering problems in the games I have used. So no reason why AMD should be crippling parts.

 

[Flash831]

At this point there should be no reason for the newly releaed A6-7470K having only 256sp:

A combination of marketing and logistical reasons probably.

If AMD puts the A6 in a bad light, more people will be willing to go for the A8 or A10. What AMD doesn't need is a cheap part which is cannibalising on their more expensive products.

And by keeping their product line slim, it might make the test & assembly process easier.

 

[DrMrLordX]

You can read the following post and get the idea on how it started:

I know how it got started, but it's still majorly off-topic, and getting dull and repetitive. Again, please stop. We've heard your analysis of Bristol Ridge already. We don't need to keep bumping an old Jim Keller thread discussing it . . . again.

 

[cbn]

A combination of marketing and logistical reasons probably.

If AMD puts the A6 in a bad light, more people will be willing to go for the A8 or A10. What AMD doesn't need is a cheap part which is cannibalising on their more expensive products.

A6 is a dual core. It doesn't matter how large the iGPU is....it is still a dual core.

In fact, there was a time when even Intel had quad cores with GT1 and dual cores with GT2. So AMD can certainly go do the same thing.

P.S. Even in a relatively non taxing game like CS:GO, my Athlon x 4 860K/GT 730 GDDR5 does much better with both modules enabled and rather than just one module enabled. So I don't people will go by just iGPU size.

And by keeping their product line slim, it might make the test & assembly process easier.

And I think that is a very good reason for AMD to put the good Bristol Ridge dies (eg, 4C + 384sp and 4C + 512sp) purely on mobile.

Then simplify AM4 desktop with an Athlon x 4 SKU, A6 dual core with large iGPU (ie, 384sp, 448sp, 512sp) and maybe an A4 APU for the most severely harvested parts.

 

[cbn]

I know how it got started, but it's still majorly off-topic, and getting dull and repetitive. Again, please stop. We've heard your analysis of Bristol Ridge already. We don't need to keep bumping an old Jim Keller thread discussing it . . . again.

DrLordMrX, if you don't like what is being written then report it to a moderator. And if they don't lock the thread, then just ignore what is being written (not every one understands the APU story and how it ties in).

And as far as AMD's obvious problems go, I won't be ignoring them.

 

[DrMrLordX]

DrLordMrX, if you don't like what is being written then report it to a moderator.

You got it. Too bad we can't discuss Keller at Tesla since that would actually be on topic . . .

And as far as AMD's obvious problems go, I won't be ignoring them.

Okey dokey!

 

[Burpo]

http://electrek.co/2016/01/28/tesla-jim-keller-apple-processor-architect-2/

 

[DrMrLordX]

Dude he looks like John Denver.

Seriously though, I am interested in why Tesla picked Keller to lead the Autopilot program. Right now they outsource for chips for use in their cars, right? It seems like maybe they want to stick to low-power CPU solutions, but the Autopilot program is going to demand more raw CPU horsepower than what they're using right now.

If Keller isn't going to design something for them in-house, then what exactly do they expect him to do? Evaluate other designs to determine which will be the best fit for their intended application?

 

[Markfw]

Well, this thread is flamebit, and 6 months old, so I am going to kill it.