hydroreadsstuff

I like the Dark Souls model for user input - messages. https://darksouls.fandom.com/wiki/Messages Premeditated words and sentence structure. With that there is no need for moderation or anti-abuse mechanics. Not saying this is 100% applicable here. But for their use case it's a good solution.

The 4x comes from the neural accelerators (tensor core in NVIDIA jargon). It's 4x fp16 over the vector path (And 8x compared to M1 because at some point they 2x'd the fp16 vector path). Therefore LLM prefill(context processing/TTFT), diffusion models (image gen), and e.g. video and photo effects that make use of them can be up to 4x faster. At fp16 that's the same speed at the same clock as NVIDIA. But NVIDIA still has 2xfp8 and 4xnvfp4.

Batch-1 token generation, that is often quoted, does not benefit from this. It's purely RAM bandwidth-limited.

afaia emulators like Fex are within 30 to 70% of native performance. On the fringes worse or better. But overall emulation seems totally fine. Plus emulator technology in general could be used for binary optimization rather than strict mappings, opening up space for more optimization.

Some companies like to stress the efficiency or performance of Arm SoCs, but really this is a hedge against more expensive x86 hardware. AMD has increased prices of mobile SoCs radically recently. I'm looking forward to having more affordable SoC options for laptops, handhelds and desktops, perhaps from Mediatek or other lower-cost vendors.

The history of the PC is one of commoditization. A fractured multi-polar landscape is detrimental to the ecosystem/productivity and should ultimately fail.

x86 emulation is an important puzzle piece, and I'm happy Valve recognizes this and sponsors it.

That's what it seems like. Some people here disagree with you, but I can add anecdata that my employer insisted I do no coding on such a VISA.

Indeed, the memory model has a decent impact. Unfortunately it's difficult to isolate in measurement. Only Apple has support for weak memory order and TSO in the same hardware.

This means GPUs are dead for local enthusiast AI. And SoCs with big RAM are in.

Because 17B active parameters should reach enough performance on 256bit LPDDR5x.

Overallcoation has a limit. You only have so much RAM/storage. Beyond that you start swapping. I could really use a hash table (or similar structure) that degrades less with higher occupancy.

For other platforms there is already https://github.com/nviennot/core-to-core-latency

This project works around a limitation in MacOS, namely lack of thread pinning, and makes it possible to do it without Asahi Linux, and without using custom MacOS kernel extensions and disabling security features.

Did they raise 6B or are they valued at 6B and don't disclose the raised amount? Probably the latter?

If that was the intend an animation that doesn't literally crush things would have worked much better. Let it fall into a black hole and let an iPad emerge or whatever. The dramatic effect of a hydraulic press adds nothing positive.

I can tell you that I watched the recording, cringed for a few seconds and skipped it, and moved on with my life. After the outrage, revisiting my 3 seconds of feelings, I tend to agree that destroying nice things isn't a great thing to do in an ad.

Egosoft was founded in 1988 and is still doing its thing

I wish the author would have put some actual numbers on the cost of the different technologies. Arguing about tradeoffs including that information would be much more meaningful.

Why did they make the front out of glas? This weigh a lot more than plastic (or aluminium if transparency is not needed), and is probably wuite expensive, too.

It’s 4GB per address space / per process. In aggregate you can use more.

Webex is worse. Sharepoint & powerpoint integration, large scale broadcast (300+?people uses different techniques) and transcripts are helpful. Haven’t used the chat much, but it seems not good.

And because Itanium's performance was worse or at least not much better than their x86 parts.

The system is set up to avoid such shenanigans. If desired, the question is will they find a way to work around the rules. I agree with you, but then the trust in the system would be greatly damaged. They might as well have two different interest rates / not pay IOR.

Afaik the needed difference in interest payment from the fed comes from the treasury. So it’s indirectly connected to the budget. In theory though you can still raise more debt to pay the interest. But I’m not sure about the longterm consequences of this.

Having inflation above the interest rate helps decreasing the debt/gdp ratio.

You don't need to mint on Ethereum. I'm there are cheaper options.

If that were the case you could probably see an interposer. And I think the B/W would be even higher.

I hope we will see this on more devices. This is a huge boon to performance.

Might even forebode soldering RAM onto packages from here on out and forever.

Steamdeck will probably have a crazy 100gb/sec ram b/w. twice of current laptops and desktops.

How do they get 200/400GB per second RAM bandwidth? Isn't that like 4/8 channel DDR5. 4/8 times as fast as current Intel/AMD CPUs/APUs? (E.g. https://www.intel.com/content/www/us/en/products/sku/201837/... with 45.8GB/s)

Laptop/desktop have 2 channels. High-end desktop can have 4 channels. Servers have 8 channels.

How does Apple do that? I was always assuming that having that many channels is prohibitive in terms of either power consumption and/or chip size. But I guess I was wrong.

It can't be GDDR because chips with the required density don't exist, right?

Bautzener Mittelscharf or bust

KOSS as well

I would look at the smaller Chinese (not SMIC) ones. Without looking I'd guess they are already highly valued. I have no idea though how long it could take to overtake TSMC/Intel/Samsung.

As someone who worked in HPC Projects in the EU, I'd say the vast majority of the grants is wasted. E.g. on

- obscure software and programming paradigms/libraries

- one-off's

- projects that create only reports, papers and recommendations

- endless recreations of the same software over the years and decades

- plain dumb projects

- attempts at recreating other people's software (e.g. U.S.)

There is also a large amount of overlap in work/content between projects, and lots of time is appropriated for unrelated work e.g. employees working on their PhDs.

On the other hand the industry takes a lot more money and plays similar games.

Project partners do not take the right path, because it's not conducive to fulfilling a grant, getting the next grant or increasing your citation count. The incentives are wrong.

Infineon, X-Fab (formerly ZMD) are coming to my mind.

Not sure you would include foreign companies with plants in the EU. Nexperia, TI, GloFo.

https://en.wikipedia.org/wiki/List_of_semiconductor_fabricat...

Aside from that I'd expected chip designers like ARM (+licensees), Bull Atos (would you call that an OEM?) and related software companies to take part in this.