hishnash

I would assume all linux GPU drivers would need to be adapted at least a little to support the larger page size (most linux AARCH64 kernel level code is writing assuming 4kb pages).

If apple were found to be in violation of Patent misuses the result would not be that apple is forced to open up the internals of how their products work but rather that patens apple has would be make null and void and others could copy them (based on the already public patent filling and nothing more).

The big issue with this is apple does not own all of the IP in a any of their hardware, much of it is a mixture of licenses from many vendors some of it owned by apple much of it just licensed.

Apples licenses for these will include very strict NDA clauses that make it impossible for apple to share the documentation they are provided but that vendor.

As described in the article by not having fixed function units apple can put more regular floating point math units. These units can be used in many, for example when doing compute tasks or when doing 3d tasks that do not make use of those very focused use cases. In the end if it is a big perf hit for those approaches devs will just use different solutions as they need to developer explicitly for metal anyway.

Why should they support vulkan? what does apple get out of that apart from less well optimised compute and shader code, using more battery and producing more heat for the same output. (the reason it would be less well optimised is Vulkan is an API designed by a group to be the best compromise of many GPU vendors.

If apple wanted to support vulkan without it being worse than Metal they would either need to add so many apple only extensions that it would be Vulkan in name only or make their GPUs be identical to AMD or Nvidia (unfortunately due to IP patents apple can't just make a copy of AMDs GPUs they need to find another IP partner and that is PowerVR).

If PowerVR had 80% of the GPU market (like Nvidia) the would have pushed Vulkan to line up with a TBDR pipeline but they do not so while you can run Vulkan on a TBDR pipeline you end up throwing away lots and lots of optimisations.

What I hope they do is use many instances of the system trained with slightly different sets of training data, otherwise, you just need to learn how to trick one version (and it will be possible) and you can trick them all.

When attacking human-managed systems they may be slow to respond, weak and unable to detect everything. But they are also weak in very different ways and will respond in different ways.

you could make indirect money by doing things like investing in some markets/betting on them to fail.

I would love to see the message log between MS Apple and Intel on this.

AMD have released patches (to reduce the near-zero risk of exploit to 0 risk) and they are not having any instability issues! so yes go with AMD

for server, Epyic is now finally available to purchase for desktop workstation Threadripper for desktop RyZen

for mobile... not many newer cpus out yet.. need to wait :(

Reading over googles project0 page it reads as if they told AMD about the issues on 2017-06-01 why would they do this if it were meltdown only?

also look at the exploit numbering:

Variant 1: bounds check bypass (CVE-2017-5753) Variant 2: branch target injection (CVE-2017-5715) Variant 3: rogue data cache load (CVE-2017-5754)

according to https://cve.mitre.org/cve/identifiers/ this is sequence based so `Variant 2` was recorded to CVE before v1 and v3.

I get it may take a long time (that is fine even if the patches took a few more days), what I don't get is that they released it to production (server) envs seemingly without testing. Surely even rudimentary testing (deploying on a few 1000 different server platforms for a few hours at least should be something that Intel does for all microcode updates, after all they are rather more important than js Node packages as you point out)

Im really wanding they had more than 6months to do these patches and they did not bother testing on a good number of systems. Its not like MS + Intel dont have enough money to buy a few 1000 testing machines and get some testers on it.

i do see a lot of people believing the Alexas system is much better but int he end alaxa is just build up on regex.

that is ok-ish for simple languages like English were the words themselves dont change.

but in many other languages, the words in sentences change depending on their relationship to each other.

eg simple information in English like "the cat is sitting on the bench" is encoding with the order of the words whereas in other languages the information the `cat` is `sitting` on the bench rather than the bench `sitting on the cat` is encoded in how the word `cat` and `sitting` and `bench` (and maybe `on`) change. This change can be rather complex based on lots and lots of factors such as gender of the `cat` and `bench` the type of verb eg `sitting` and the preposition `on` (some languages don't even bother with this since it's added to the verb).

apple's approach is much more upfront work since they do all this natural language logic in every language they support. But for developers who use Sirikit is is much better.

but a lot of this brain drain has gone to private security companies that then sell vulnerabilities to national bodies. (Like the one in Israel that sold a load of 0day exploits for the iPhone to the CIA)

intel want to be able to operate all around the world and not be forbidden form non US markets if it is seen that they directly pass an exploit to the US gov so that US gov can use it to exploit into other nations systems then those other nations will ban Intel from sales. (even US allies would ban intel)

It's not about moral its about sales, if Intel were to start to tell the US gov directly about security loopholes in products they were still selling to other nations those nations would blacklist intel for life.

Just as the US would blacklist intel if they told another nation directly first. (this is even true if they told allied nations) we have seen that just being allies does not stop the US spying (and I'm sure does not stop the opposite as well just turns out the others might be better at keeping things hush hush)

Yes through you may have left out the UK (with GCHQ) who do most of the spying for the NSA within the US, so as to ensure the NSA doesn't break any rules as they are permitted to take data from another nations department but have restrictions on local citizen data ;) they intern share data they gather on the in UK.

> Enforce federation by splitting the middle "we actually host users and their walls and their data" company along state lines and international borders, so that they are more beholden to local authorities and jurisdictions.

this would push up the costs massively. Every single state int he US requiring its own data centre. And what do you do if I have a friend in another state and share a post with them? who has the legal power to block that post?

The issue I see with adding gov level regulation to Facebook (or any other media) is it very quickly becomes a tool for oppressive regimes.

If Facebook builds in the tools so governments can control exactly what is posted on them then expect every single nation in the world to require access to these tools overnight.

Even nations we might think of as being very condition of the right of their citizens have content they deem illegal and would if they could be required by their own laws to compel Facebook to let them control this.

This, in the end, would not just be about public content but also about all private content, messages sent over FB messenger etc.

Do we want a popover to show up next time we share a hypothetical "dick pick" with a 'lover' saying 'awaiting review from your local authorities'? God forbid you are on holiday/changing planes in a religious extream region of the world and find yourself carted off to jail for 10 years for this.

IM not sure its that much of bedfellows, with this they are not sharing their IP with Intel, these GPUs are discrete chips that AMD is just selling to Intel who is then selling them alongside their CPUs on the same package.

it's not like AMDs GPU designs are going shared with Intel. Nor are they giving up control of fabrication.

The title is not quite true,

intel will ship a package that has an AMD GPU+HBM combined with a CPU but they are not the same chip, basically, Intel is buying already fabricated GPU chips from AMD and putting them together with their CPU to sell them as one unit.

also, apple do a really good job on compilers etc to make it easier to write code and systems that jump between systems quite well. Lots of people dev on windows without using the latest tools from Microsoft this is just not how it works with apple.

I think that in the end this will be down the mainboard makers to get validation so possibly will not be validated on non-server style boards. But if there is demand given epic is basicly the same thing just bigger im sure they will do it.

Oh, yer will be very nice for Machine learning I can see a lot of people getting this.

Also even though they did not mention it for lower end servers. Servers currently using a single Xeon.

> official/validated ECC support

Given all the other Zen cpus have this i think it will, after all it is just 2 ryzen CPUs with a load of PCI!

> there was also no word on official/validated ECC support

I think given all the other Zen CPUs support it i think that is a given.

> 999$ (since it is two Ryzen R7) and 1499$

I think it will be more around the 999$ but the motherboard will be quite expensive, it is after all basically a duel CPU main board. Don't expect to pick up a board for under 500$

Yes, massivly deep learning is all about learning from data. With the work I have done I have seen that currently unless you buy a 2xeon server you can't get enough data through to justify haveing 4 GPUs on one machine. So commonly its cheaper to break out to multiple machines, but this then has an impact on how you code your learning since communication between the GPUs then is much slower.

Think of training a NN to learn cats expressions and such Meme txt to go along with pictures of cats. So 1) you collect as many cat memes as you can (easy,... 40k memes of cats later) you start training... you need to normalise all the images you need to extract the meme txt and normalise this and you need to feed this through your tenso flow (or other system) to train it. This means pusshing your small data sample of 40k through the GPU and back out and possibly doing this a lot, there is no way they will all fit on your GPU and even if they do you need to get them on there in the firrst place.

I think AMD have been really clever in how the have built the Zen system, unlike Intel these higher end CPUS are basically just multiple lower end ones put together. This means they don't need different fabrication (they just build ryzen7 CPUs) and given they have such good yields on those chips know they can bring the price down. I would expect this chip to be at most 2x a ryzen 1800.

However, it is not jus the chip cost for these systems its also the mainboard if AMD can get the motherboards to be cheaper then that will make a big difference in the overall price.

So AMD have done something really clever, it is not 1 big Die but rather 2 smaller (ryzen7) dies linked.

The issue with binding is the larger and more complex your die the more failures you have. But breaking it down to 2 smaller dies that are combined they reduce this since they are already getting relly good yields out of the ryzen7 dies.

Epic is just 4 ryzen7 dies. So no this is not worse bins the 16core threadripper is 2 ryzen7 cores.

What is really nice about the full Zen line is they are supporting ECC memory. It always felt like Intel had removed this on lower end chips just to force users to buy a Xeon even if the did not need the rest of the Xeon features.

Gamers might the most vocal high-end PC users out there but they are not the people with the most money.

The threadripper is not targeted at gamers. It is targeted at people who make money from using their machines, these people can, therefore, afford to upgrade.

Consider a user who does video work. (most modern films and tv is filmed in 8k know) if this is in Raw format as it comes off the camera for a 90 min episode that is 6000 GB. Working on this data (tone mapping, cutting, etc) needs all this data to be fed in and out of the (CPU and or GPUs) so all these PCI lanes are there so you can have very fast access to the raw data and still have space for 2 or 4 full speed GPUs to do that work for you.

Forget games no game developer optimizes their games for 4 GPUs these days even Nvida have said that they dont realy want people to do more than 2GPUs in SLI.

Our customers are (I work at https://paw.cloud) For us we are happy with django for backend since we can trust it, but when we do cool stuff with django we take a few days to right up a blog article that for us is direct marketing.