Apple Loses A7/A8 Patent Lawsuit, Could Owe University of Wisconsin Up to $862 Million

[kdarling]

Update:

According to Reuters, WARF is now asking $400 million in the damages part of the trial going on now. (Although that could triple if Apple is found to have willfully infringed.)

The reason for the lower amount is that reportedly WARF has dropped damage requests for devices sold before the trial. They didn't have to, as Apple's citations of WARF's patent proved Apple was aware of the patent long before the trial.

Note that it's not the University itself that is suing Apple. It's their separate non-profit R&D organization... the Wisconsin Alumni Research Foundation (WARF), which takes revenue from their patents and puts it back into research.

WARF, like the similar Stanford Research Institute, was founded about seventy years ago to promote research. Most people know SRI as where Apple's Siri originated as a DARPA project... similar to this WARF invention. In both cases, Apple has ultimately benefited from government R&D funding.

 

[kdarling]

Next target will be Intel, AMD, Qualcomm, Mediatek, TI, Nvidia, Samsung. Nobody can make multicore processor without the patent, if it's allowed. It shouldn't be a patent.

I don't know if it should be a patent or not, but certainly you can make multicore processors without it, since that was done for years beforehand.

So is Apple the only company using branch prediction in their CPU's?

Nope, branch prediction has been around for decades. However, the patent is not about branch prediction. (Is anyone really surprised that the internet echo chamber got it wrong?)

It specifically talks about running multiple processes in parallel... so maybe not just ordinary branch prediction, but branch prediction that also factors in which thread is running it?

You're very close. Instead of being about branch (control) speculation, it's about data speculation during parallel processing. In both cases, parallel processing is going on, but:

In parallel branch prediction (control dependence), the need is to ignore/discard the results of the instructions for the branch that was not actually taken.

In parallel data dependence, the problem is to make sure that one set of parallel processing instructions (a data consumer) does not use data that has yet to be written by another parallel process (the data producer).

As the patent puts it:

"Control speculation is well understood in the art. The points where control speculation is needed are clearly identified by conditional branch and jump instructions (we will refer to these instructions as "control transfer instructions"). Typically all control transfer instructions are speculated since: (1) control transfer instructions are relatively frequent, (2) relatively few instructions from those that appear between two consecutive control transfer instructions can be executed in parallel, (3) typically the performance of a processor that always mis-speculates on control is virtually the same as the performance of a processor that never speculates on control.

"In contrast the points where data speculation is needed are not clear since any instruction loading data from memory can be data dependent on any previous instructions that writes to memory. Consequently , predicting and tracking data dependencies, "data dependence speculation" can easily become overwhelming. Furthermore, the cost of a data mis-speculation typically cannot be neglected, that is the performance of a processor that always mis-speculates on data dependencies is far less than that of a processor that never speculates on data."

Normally this is handled by stopping the data consumer process until the data provider process has written the data.

While I will not pretend to understand it completely, basically, the WARF patent came up with methods to keep a dynamic table that allows the data consumer threads to continue processing if past experience is that the data producers will not have changed the data. Otherwise they can be delayed until the write occurs.

 

[Fenez]

Kdarling as always...very thorough, that is a guy that obviously liked doing homework.

 

[Benjamin Frost]

You do.. Me!
A cheque will be fine

It's in the post.

 

[staypuffinpc]

Aren't institutions of higher education meant to do research to propogate information in a free society? Universities aren't businesses (at least not in the manufacturing/production sense). Was the university going to be using their patents to develop products? If this proves to be successful, this might start a trend in university suing over patents/research papers...

Universities are institutions of research as much as they are of instruction. A LOT of inventions over the years occur b/c of the resources and investment committed by a university to the R&D of such ideas. As such, research universities have personnel on-hand to help faculty with licensing (and it goes both ways, as the university has to pay others for the rights to use their materials in certain regards).

At my own university, I can keep 45% of the profits of something I invent for myself, should I so choose. The university pays for the patent/trademark filings and takes care of the legal side of things. 22.5% of the money goes to my college, and the other 22.5% goes to the university. If I want to donate any portion of that 45% back to my research account, the university will match that percentage back into my research account, essentially making it so that I can use 90% of the profits on furthering that research.

 

[nsayer]

So a California jury found in favor of Apple? Color me shocked.

No argument here.

 

[I7guy]

Was this judgement overturned today?

 

[Feenician]

Edit: Just realized this was necro'd

Was this judgement overturned today?

Looks like it.

https://9to5mac.com/2018/09/28/apple-university-of-wisconsin-patent-infringement/

 

[joeblough]

seems like it was only partially overturned? in another thread part of the decision was posted which seems to indicate that the judge felt that there was still infringement in certain cases (when load/store tags are associated with only one load/store instruction rather than multiple)

 

[cmaier]

seems like it was only partially overturned? in another thread part of the decision was posted which seems to indicate that the judge felt that there was still infringement in certain cases (when load/store tags are associated with only one load/store instruction rather than multiple)

No that’s not what he said. He said there is no evidence of even intermittent infringement.