klelatti

23 ms·

451.
SiFive Rolls Out RISC-V Cores Aimed at Generative AI and ML(allaboutcircuits.com)
151 points by klelatti 3y ago|64 comments
452.
by klelatti 3y ago
Thanks so much! It's a really interesting design.
453.
Going Vertical: Gate All Around, 3D DRAM, 3D NAND(semianalysis.com)
2 points by klelatti 3y ago|0 comments
454.
by klelatti 3y ago
Are you contesting that it was more approachable? I didn’t say the Spectrum was more interesting - it definitely wasn’t!
455.
by klelatti 3y ago
The simpler hardware of the Spectrum was much more approachable. And the basic was so slow that you really had to learn some Z80 assembly!
456.
by klelatti 3y ago
I was surprised to find that a VM was used in the Busicom calculator that was the motivation for and first use of the Intel 4004 microprocessor. Wrote about it here: https://thechipletter.substack.com/p/bytecode-and-the
457.
Nvidia's Plans to Crush Competition(semianalysis.com)
4 points by klelatti 3y ago|0 comments
458.
by klelatti 3y ago
Author here. As noted elsewhere this originally had a paywall in the middle but I’ve removed it for the time being. Hope the extra projects are of interest.
459.
by klelatti 3y ago
Your definition of ‘push’ here is make compulsory - which doesn’t make sense in the context anyway. I’ll leave others to judge if the quoted paper pushes it and move on.
460.
by klelatti 3y ago
> Yet, it does not push macro-op fusion in implementations Yes it does. It pushes it as an alternative to adding instructions to the ISA. It says we don’t need these extra instructions because architects can use macro-op fusion to remove
461.
by klelatti 3y ago
Seriously? You weren’t aware of this? The Renewed Case for the Reduced Instruction Set Computer: Avoiding ISA Bloat with Macro-Op Fusion for RISC-V https://people.eecs.berkeley.edu/~krste/papers/EECS-2016-130...
462.
by klelatti 3y ago
That’s such a strange criticism given how keen the designers of RISC-V have been to push macro-op fusion.
463.
by klelatti 3y ago
About twenty years ago I hosted a group of business visitors from Slovenia in the U.K. and was surprised to receive a beehive panel as a ‘thank you’ at the end of the meeting. It was a lovely gift and I still have the panel.
464.
by klelatti 3y ago
Thanks so much for a super helpful reply!
465.
by klelatti 3y ago
On a slight tangent, as you've worked on both Arm and RISC-V cores, could you comment on how the two ISAs compare from a design standpoint? How much more complexity does designing for an ISA like ARM64 add when compared to a more &#x27
466.
The Race for Semiconductor Supremacy(ft.com)
2 points by klelatti 3y ago|0 comments
467.
Neural Networks and the Origins of Modern Computing(thechipletter.substack.com)
2 points by klelatti 3y ago|0 comments
468.
by klelatti 3y ago
> Qualcomm sent them ~$270m last fiscal year > ARM is pushing them away > “tens of millions” in Qualcomm’s telling > ARM “winning” this case looks like losing to me. I'm sorry, this is once again a very one sided portrayal
469.
by klelatti 3y ago
> a terrible idea Well you’re not really holding off judgment then! It’s also a terrible idea to let big customers breach their contracts but as you say we haven’t seen the contracts concerned.
470.
by klelatti 3y ago
> on the idea that Qualcomm did not get the same rights as Nuvia had to use Arm’s designs The point being that they wanted the rights Nuvia had but without the obligations that Nuvia license implies - most specifically on fees. And yes t
471.
by klelatti 3y ago
Qualcomm took the first step in this case by trying to sell Nuvia derived designs under their own license agreement. Whoever wins it’s not really Arm going on the ‘attack’ if there is a reasonable case that the license has been breached.
472.
by klelatti 3y ago
> For a product developer, it eliminates the issue of being tied to the limited portfolio of cores available from a proprietary ISA. Hmmm this is the same Qualcomm that had an Arm architecture license for years and ended up using off the
473.
Apocalypse Proof: 33 Thomas Street in New York City(placesjournal.org)
136 points by klelatti 3y ago|72 comments
474.
by klelatti 3y ago
> Apple holds a rare Architecture license to the ARM ISA due to their role as a founding partner, and can make original implementations without incurring licensing dues Sorry, the architecture license wasn’t a thing when Arm was founded
475.
by klelatti 3y ago
They have an architectural license (for Arm v8/9) because they are a huge customer with deep pockets not because of a shareholding they sold a decade before that architecture was announced!
476.
by klelatti 3y ago
> renamed Acorn Electron Actually renamed Acorn Proton. Electron came later.
477.
Happy 65th Birthday, Integrated Circuit(thechipletter.substack.com)
20 points by klelatti 3y ago|1 comments
478.
China AI and Semiconductors Rise: US Sanctions Have Failed(semianalysis.com)
6 points by klelatti 3y ago|1 comments
479.
Intel announces Arm investment, talks up RISC-V(tomshardware.com)
150 points by klelatti 3y ago|94 comments
480.
TPUv5e: New Benchmark in Cost-Efficient Inference/Training for<200B Param Models(semianalysis.com)
1 points by klelatti 3y ago|0 comments

More ›