The AMD Execution Thread [2007 - 2017]

[3dilettante]

One of my friends threw interesting scenario in the air, would it be in the realm of possibilities that RTG will be spun off as a separate company, from which AMD owns majority and Intel minority, for example 51/49? It could give RTG enormous possibilities on the budget side, possibly access to better manufacturing process, which would benefit both AMD and Intel?

One of the things that comes up in a joint venture like this would be that when the JV starts asking for resources such as capital calls, or the accounts go over owner contributions over time, is how sustainable the current split is.
AMD faced this with its Globalfoundries spinoff and split ownership with ATIC.
As long as the specter of tripping some of the last covenants of the prior cross-licensing agreements with Intel remained, AMD had to maintain at least some semblance of control.
However, capital calls for GF inevitably diluted AMD's share since it couldn't contribute on the level of Abu Dhabi's investment fund. AMD made it clear they would be looking for ways to divest the remainder of their holding, and once an agreement with Intel was made the final sliver was sold off.

A joint venture has an AMD with historically limited investment as the controlling party, and then an Intel whose partnership provides the enormous budget and manufacturing contributions?
How long would AMD's inferior investment keep it in control, or why would Intel agree to give all those things if while AMD coasts by calling all the shots?

We lost around 50-60% of one of our R&D tech product team to another tech company collaborating with us when the senior lead engineer jumped to them and then targetted others involved in the development, they went from having nothing to 3 years later a product that competed moderately well to another 2 years after that being a serious competitive headache; this was an incredibly complex technology to develop from scratch - cannot say much more than this otherwise would start to become too specific.
They managed this without breaking their IP/trade secret related obligations as well.

The key is who Raja will target, which may also include specific engineers within Nvidia and other companies as well.
Working on a new fresh project with high levels of resourcing and under someone like Raja will appeal to quite a few; look how Jim Keller prefers to operate and move around for the project interest.
Of course anyone joining will need to weigh up the history of Intel-the current CEO and commitment to such a project/division like some in this thread mentioned.
And will Xeon Phi (and associated team) in the shorter term also end up under new division/Raja or at least overlap/work with them.

One related event was AMD's loss of most of its Jaguar team, with a similar sequence. The other party was mostly Samsung, although as a competitive headache I'm not sure it's threatened AMD directly yet.
It did mean that AMD's decision to focus its efforts on one CPU design was not just a decision based on outside factors, as the choice to do otherwise was effectively removed.

Anyway, I think, that Eric Demers should never leave AMD. I'm also convinced, that neither Demers nor Koduri are responsible for current situation. What we see is a result of decreased budget for AMD's GPU division, result of Read's administration.

Could you clarify the Demers statement? Did you mean he should have never left? I recall he went to Qualcomm in 2012.
One additional wrinkle to the blame game for Read is that there were at some rumors concerning the cutting of R&D started back with interim CEO Seifert, though Read obviously had the choice of changing things if he didn't agree.

 

[no-X]

3dilettante: Yes, I meant it that way. As for the costs, Read decided to lower expenses by $120 mil. per year at the end of 2011 (~10 % of workforce left AMD). Later in 2012 AMD announced another employee reduction. It resulted in prolonged development cycles and decrease from ~4 new GPUs per generation to 2-3 per generation.

 

[DrYesterday]

You asked me for an example *shrug*.
Going by what I found after you asked for an example.

I asked for an example where (your words) "court order to cease said product/services/concept against the new company to begin with..." I do not see any such example.

 

[CSI PC]

I asked for an example where (your words) "court order to cease said product/services/concept against the new company to begin with..." I do not see any such example.

Feels like you want to argue semantics.
In general you have cease injunctions with IP infringement more recently from such as AMD is bringing against some manufacturers they started early in year, or like how Apple did against Samsung.
If a company believes an ex-engineer has used trade-secrets then usually this is a request for information 1st, followed promptly by injuction request (both partial and full), followed by court and possibly full injuction given along separately with compensation, and outside of this you have the engineer/people involved sacked and potential for legal prosecution case or at a minimum the headache of a federal investigation, and then there will be the company's internal audit getting in the way as well.
So yeah it does start with court order injunction.... well could be argued discussion between companies for request of information to try and validate infringement/concerns that then is pushed more heavily by a court judge.

You ignored the whole context of my post, which also included a trade secret theft example which ended up with the engineer being sacked but before that the judge ruled he must be removed from a role working on the project he was employed by at Uber (this was a partial injuction the judge did accept), the original company suing for massive compensation, and also the Judge referring the case it to the legal federal prosecutors to investigate if there should be arrests/prosecution.
The only reason it was not worst for Uber is that like I said Waymo were undermined by their own internal emails that came to light saying the theft/use was a secondary and not important data, they overplayed their hand to the court and the judge was not happy; but it is still going to court for compensation and potential full injunction against Uber - doubt Waymo will get the full injunction though due to email admission.
https://forum.beyond3d.com/posts/2010189/

Like I said these days you are not going to usually see a blatant ex-employee use their previous companies IP/trade secrets for all of those reasons, I even provided a link showing the type of legal obligations several here are under including myself and all of us are under no illusions we would be shafted and the fallout depending how great the infringement is also a headache for the company one moved to.
No-one going to work for one of the top western tech companies is going to push it too far because of these factors, these days anyway or unless your in a company based out of China/Russia
Even a partial injunction instantly applied is a nightmare as it can remove a lot of the team from the R&D project if they all came from the same original company or stall the R&D, let alone what follows.

 

[CSI PC]

OK here is one back to 2013 doing another quick search:
Brocade Awarded Permanent Injunction in Intellectual Property Theft Case against A10 Networks
Court Orders A10 to Stop Selling and Shipping Products Infringing on Brocade Patents

On Aug. 6, 2012, a jury in the case of Brocade v. A10 Networks found for Brocade on four claims of patent infringement involving technologies for Global Server Load Balancing and High Availability and found that the A10 AX Series line of load balancers infringe on these specific patents. In addition, the jury found A10 liable for:

• Direct copying of proprietary Brocade code used in Brocade’s ServerIron products for use in A10 products.
• Misappropriation of four trade secrets involving techniques used in the Brocade ServerIron products that A10 applied for use in its AX series.
• Unfair competition based on interference with the contract of an engineer while he was employed at Foundry Networks, which Brocade acquired in late 2008

Also involved financial damages.

 

[DrYesterday]

Feels like you want to argue semantics.

Sorry, not my intentions. And thanks for for the Brocade reference.

 

[A1xLLcqAgt0qc2RyMz0y]

'Kernel memory leaking' Intel processor design flaw forces Linux, Windows redesign

https://www.theregister.co.uk/2018/01/02/intel_cpu_design_flaw

This sucks for all those who are running Intel processors - myself included.

But this will help AMD sell more processors as they are not affected.

In an email to the Linux kernel mailing list over Christmas, AMD said it is not affected. The wording of that message, though, rather gives the game away as to what the underlying cockup is:

AMD processors are not subject to the types of attacks that the kernel page table isolation feature protects against. The AMD microarchitecture does not allow memory references, including speculative references, that access higher privileged data when running in a lesser privileged mode when that access would result in a page fault.

A key word here is "speculative." Modern processors, like Intel's, perform speculative execution. In order to keep their internal pipelines primed with instructions to obey, the CPU cores try their best to guess what code is going to be run next, fetch it, and execute it.

 

[Kaotik]

But this will help AMD sell more processors as they are not affected.

Yet it still might affect them even though they don't have the bug - apparently Linux-devs are ignoring AMD's request for vendor check.

Older commits to the Linux kernel git, which should feature the line "if (c->x86_vendor != X86_VENDOR_AMD)" (condition that the processor should be flagged "X86_BUG_CPU_INSECURE" only if it's not an AMD processor), have been replaced with the line "/* Assume for now that ALL x86 CPUs are insecure */" with no further accepted commits in the past 10 days.

(though I did hear that some RC-version of the patch did have the AMD-line included, so there's still hope - but as it stands PTI is forced on AMD too and hurts their performance)

 

[3dilettante]

They could be playing it safe. Long-term if AMD is able to make the case this doesn't apply after the full disclosure, then there should be enough pressure to get something through.

On the other hand, there could be ancillary effects or long-term plans to isolating the kernel space that the OS devs think makes it worthwhile even for AMD, or the check as written assumes too much about what any future AMD chip may or may not do.

How readily can that check be spoofed?

 

[Malo]

Can we limit the CPU flaw discussion to the dedicated thread?

 

[A1xLLcqAgt0qc2RyMz0y]

Can we limit the CPU flaw discussion to the dedicated thread?

Why can this not be in the amd-execution thread as it can directly impact AMD's bottom line if the end result is more enterprise processor sales because of it.

And with this news:

Linux Will End Up Disabling x86 PTI For AMD Processors - Update: Now Disabled
https://www.phoronix.com/scan.php?page=news_item&px=Linux-Tip-Git-Disable-x86-PTI

AMD will have a stronger leg up in enterprise sales.

 

[Malo]

Why can this not be in the amd-execution thread as it can directly impact AMD's bottom line if the end result is more enterprise processor sales because of it.

I was just trying to limit the discussion of the flaw attacks to one thread, especially after you posted it in like 4 different places, including a new thread lol.

And with this news:

Linux Will End Up Disabling x86 PTI For AMD Processors - Update: Now Disabled
https://www.phoronix.com/scan.php?page=news_item&px=Linux-Tip-Git-Disable-x86-PTI

AMD will have a stronger leg up in enterprise sales.

Yep, definitely appropriate for the execution thread when discussing the impact of it for AMD financially.

I think it could have a significant impact in sales just from this change, but also from the backlash against Intel. But really the biggest decider for enterprise will come actual impact on performance for their respective application in their datacenters. If in the end performance isn't much for PCID Intel CPUs, how much will datacenters really care that the "flaw" is there after the patching?