andars

Xilinx has a pretty clear disincentive to optimizing Vivado for minimum area. If it does a bad job optimizing, the user is forced to switch to a larger FPGA -> more $ for Xilinx.

> more CPU vendors more risk averse

Seeing how Itanium went, their risk aversion seems pretty wise.

Itanium seems to satisfy most of your desires with its VLIW/EPIC architecture, which exposes much more to the compiler.

Higher frequency on power lines would be less efficient, not more. High voltage DC is the most efficient, but is most expensive because you can no longer use transformer for step-up and step-down.

TI definitely did not have a monopoly on ICs for those 25 years. Those first patents were on Kilby's process which was essentially abandoned. Fairchild and Intel did a tremendous amount of work getting ICs to be more useful over that timespan (see self-aligned gates).

Frankly, I don't care if the page is titled "FUBAR license XYZ123" since I was asked to agree to the EULA at the third link alone. The third link is presented and you agree when you try to download the GeForce or Titan driver from the GeForce website (geforce.com). I understand that nvidia.com presents the second link, but GeForce.com presents the third link. I said as much in my edit. Please read the entirety of my post before correcting it.

Thanks for pointing out that the original Japanese article did have a link.

My apologies, I misunderstood. The CUDA toolkit license online doesn't appear to have this clause at this time.

CUDA requires the kernel driver. All three licenses matter.

No. Tesla and Quadro are definitely exempt from this potential update to the EULA, and you can have a lot of GeForce cards outside of a datacenter.

It probably depends a lot. There is the SXM vs. PCIe distinction plus discounts for bulk and all kinds of negotiations.

But you can already not buy a datacenter's worth of GeForce; they have limitations on the quantity per customer. That seems like a far more effective measure than this clause.

Sometimes. The GeForce GTX 690 used the same chip as the Quadro K5000 (the GK104). More recent generations typically have different silicon (or bins with defective regions) in the pro and gaming lines. For instance, high end pascal gaming cards use GP104, but Tesla P100 uses the GP100.

For more details, see https://en.m.wikipedia.org/wiki/List_of_Nvidia_graphics_proc... (Notably the "code name" columns).

Just curious, what cards are you comparing when referring to the 10x multiple?

I just tried for Titan V on Linux through geforce.com and got a EULA without this clause. Downloading through nvidia.com probably gets the no data center EULA.

Of course it matters which version we are looking at. It was not initially obvious to me that this clause was added, not removed between 2009 and now. See my edit: Two different driver download pages link to different EULAs.

I went to https://www.geforce.com/drivers and searched for 1060 on Linux.

Then clicked through to https://www.geforce.com/drivers/results/126577. This page has the standard "*By clicking the "Agree & Download" button, you are confirming that you have read and agree to be bound by the License For Customer Use of NVIDIA Software..." That sentence links to the EULA that I linked to.

EDIT: Alright, it matters which site you download the driver from. See the edit to my original comment.

So, my summary:

Unsourced Japanese news outlet published an article claiming this clause was added to the EULA. (Edit: located here: https://wirelesswire.jp/2017/12/62708/ )

This version of the EULA, located at: http://www.nvidia.com/content/DriverDownload-March2009/licen..., has the no-data center clause. Note the "2009".

The version linked from the actual driver download page (at https://www.geforce.com/drivers/license ), has no such clause.

I think I'll postpone my outrage until the clause appears on the EULA that I actually have to agree to when I download GeForce drivers.

Alright, edit: For me, downloading drivers through https://www.geforce.com/drivers gets me the second EULA that I linked to. However, downloading drivers through http://www.nvidia.com/Download/index.aspx?lang=en-us gets me a EULA with this data center "limitation". This seems to me to be pretty problematic and an ineffective update.

Anyone else bothered by the 2009 in the url?

Edit: Some driver download pages currently link to this page.

Probably memory bound.

Here is the code for Realtalk: https://pbs.twimg.com/media/DJc2esBVwAAnj4X?format=jpg

Source: https://twitter.com/redblobgames/status/907253802931126272

or perhaps:

- Are concerned about potential damaging baseless accusations/rumors/gossip from third parties that you don't trust?

Necessary disclaimer, because internet discussions are great:

Do not interpret this as an endorsement of Pence's policy. I am not interested in arguing about the character of Pence. I am simply pointing out that there are other valid reasons that someone could adopt such a policy.

In that sentence, I am fairly confident that by "them", 'czep is referring to S and others like him. In the next sentence, 'czep is suggesting that the other members of S's group voluntarily resign to "ruin" S.

My understanding of the article:

When people use Facebook in ways that reproduce interactions through technology that existed before Facebook, the results are positive (e.g. texts between close friends, IMs, emails, etc). When people use the core features that social media introduced, the results are negative (e.g. status updates to your 1000+ "friends", "likes", etc).

Is this an accurate summary of the findings discussed in the article?

It seems like you are confusing vector algebra with linear algebra. The cross product is not a part of the theory of vector spaces and linear transformations.

Geometric algebra is pretty cool, but it is a potential replacement for vector algebra, not linear algebra. It is defined on a vector space, which means that the theory of linear algebra (concepts like linear transformation, bases, etc.) applies.

As a sidenote, reducing Maxwell's equations in the 4-equation Heaviside form to a single equation is not necessarily a boon. Each of the four has a clear physical meaning in the vector calculus formulation, which may be lost when moving to a single equation.

RISC-V is an instruction set architecture that is freely available for use. That is, it specifies the instructions and instruction encodings used by a CPU, like x86 and ARM. Unlike x86 and ARM, you don't need to license the ISA to create cores implementing it.

RISC-V cores are not from a specific manufacturer.

The Friedberg, Insel, Spence book [1] passes your test. Linear transformations first, then matrices as a representation of transformations with respect to a specified basis. Depending on your tastes, you may find it too dry, but imo it is a clean presentation.

[1]: https://www.amazon.com/Linear-Algebra-4th-Stephen-Friedberg/...

> Unlocked FP16 ... outside of the Tesla series

Titan V has HMMA, so it most likely also has full hardware FP16. It seems to have a binned V100 chip with 3 functioning HBM2 stacks.

It is strange that you expect to be able to buy the server grade GPUs for GeForce series prices and are outraged when reality doesn't meet your expectations.

It strikes me as pretty awesome that the V100 is now available for only $3k in the Titan V, rather than ~$10k for the Tesla V100. I reckon you need to readjust your expectations.

Larger (higher capacity) HBM2 stacks do not yet exist and the manufacturing process is pretty complicated/expensive. It's worth noting that all the stacks end up on the same piece of silicon as the GPU itself. Don't assume that they are holding back tech due to lack of competition.

650 GB/s is pretty impressive.

Could you clarify the "4 orders of magnitude" comment?

~10 atoms is ballpark 2nm. If you are referring to a length dimension with your comment, four orders of magnitude larger would have some dimension on the order of 20um, corresponding to a 1970s process.