Veedrac

55 ms·

661.
by Veedrac 6y ago
Yes, 2 W/core. The difference is that at this power level, the A series chip will be a lot closer to peak performance than the Intel chip. > A 4800HS at it's 4w/core all core load is also still clocking higher than an A12X
662.
by Veedrac 6y ago
Power-performance is a very nonlinear curve, so it doesn't make sense to compare single-core TDPs with all-core TDPs. The iPhone XS is 2500 MHz when one core, but drops to 2380 MHz when both primary cores are in use, a 5% performance d
663.
by Veedrac 6y ago
The official results from spec.org have a bunch of cheating, eg. exploiting undefined behaviour to run a benchmark improperly. AnandTech uses a consistent compiler (Clang, not ICC) without settings to exploit this, hence the divergence. And
664.
by Veedrac 6y ago
FWIW you've linked Geekbench 4; Geekbench 5 is https://www.geekbench.com/doc/geekbench5-cpu-workloads.pdf I agree these benchmarks have issues, but there's a consistent trend (eg. AnandTech's SPEC result
665.
by Veedrac 6y ago
Compilation should be significantly faster on an A14 based MacBook, since an unthrottled A13 is already within 10% a 9900K for GCC.
666.
by Veedrac 6y ago
Sorry, I meant to say “I've taken some measurements of the A13's microarchitecture (mostly, its ROB size) with Andrei Frumusanu from AnandTech ”.
667.
by Veedrac 6y ago
You'd almost certainly need to redesign the hardware for that; the A13 already pushes its efficiency curve during boosts.
668.
by Veedrac 6y ago
No, I'm plotting the number of transistors per unit area.
669.
by Veedrac 6y ago
See SPECint2006 in [1] (ctrl-f “absolute performance”). SPECfp is maybe more like a 60% IPC lead, but less relevant to typical users. I've taken some measurements of the A13's microarchitecture (mostly, its ROB size), and Apple ha
670.
by Veedrac 6y ago
Here's my transistor scaling chart. TSMC 5nm should be coming this year, with 1.8x scaling. So my numbers say we haven't stalled yet. https://docs.google.com/spreadsheets/d/1NNOqbJfcISFyMd0EsSrh...
671.
by Veedrac 6y ago
Apple already has an ~80% IPC lead over Intel. So, me.
672.
by Veedrac 6y ago
You're mixing up terms here. Dennard scaling is the claim that power density stays constant with node shrinks. This has failed, but transistor scaling, as in the Moore's Law progression of transistors getting smaller in size, has
673.
by Veedrac 6y ago
This is demonstrably false; AlphaGo made significant new discoveries, for example.
674.
by Veedrac 6y ago
https://youtu.be/hx7BXih7zx8?t=513
675.
by Veedrac 6y ago
> I'm not sure how anyone who's watched the exponential growth of a brand new domain can pick a point today to and say that things aren't as good as we expected. They do worse than this; one of the articles in the section
676.
by Veedrac 6y ago
Having seen the realtime-rendered Horizon Forbidden West trailer, justifying Chrome's ‘smooth’ animations this way is like stepping into a parallel world. I can't even scroll Twitter without stutter on my hefty gaming laptop, my b
677.
by Veedrac 6y ago
They're not quite append only, there's a ‘zone random write area’ that you can update in-place that exposes the write buffer in front of a zone. To a lesser degree, it also supports parallel appends, which naturally aren't pu
678.
by Veedrac 6y ago
ZNS does support random writes to the uncommitted blocks at the head of a zone.
679.
by Veedrac 6y ago
Also buy a (half) decent mic! They're much cheaper than you might expect.
680.
by Veedrac 6y ago
The main abstraction costs are excess DRAM and overprovisioning, and those don't come from wear levelling but the block abstraction layer (aka. pretending to support random writes). So where there used to be a buzz around ‘open-channel
681.
by Veedrac 6y ago
I doubt we should read too much into the difference between 1.8x and 2x. If there's a long-term trend, it'll take more than one data point to see it. TSMC 3nm does sound less ambitious, but it's also supposedly coming early,
682.
by Veedrac 6y ago
The top results are nonsensical.
683.
by Veedrac 6y ago
And yet the UE5 demo can run at a stable 30fps on a console. Computers aren't slow any more. Even laptops can render a menu bar.
684.
by Veedrac 6y ago
> Unfortunately, Feynman’s “room at the bottom” is no longer plentiful. There's plenty of room at the bottom. Another million times the transistors is hardly implausible; a thousand times is practically a given. > Although other
685.
by Veedrac 6y ago
http://ftp.altlinux.org/pub/people/mike/elbrus/docs/elbrus_p...
686.
by Veedrac 6y ago
Stereoscopic vision ≠ depth perception. If you don't have any visual depth perception, you're probably blind.
687.
by Veedrac 6y ago
A petaflop/s-day is a nice unit because it's roughly what you get from running the fastest AI accelerators for a day. The V100 they used is 0.13 petaflop/s FP16, for example, and the recently announced A100 ups that to 0.3.
688.
by Veedrac 6y ago
It's weird that this has anything to do with the ebook. My e-reader gives those options (and many more) for plain epubs.
689.
by Veedrac 6y ago
You don't need to solve every problem to make AI worth the investment, just a large enough handful of billion-dollar problems.
690.
by Veedrac 6y ago
Magic has the unfortunate downside of preventing people comprehending what's happening. This isn't an issue exclusive to str, and having it silently autocorrect for only that one case will teach people the wrong thing.

More ›