menaerus

OBD vehicle diagnostic software.

I'm interested in more details around ceph if you can share. What type of setup, what database, any fine-tuning, what type of workloads, dataset volume etc.

I'm asking because when I carried out the research on ceph, it seemed to me pretty solid and not necessarily, to my surprise, easily bottlenecked

Please see my other comment wrt Nord. What I am talking about is that flagship phones from OnePlus are suffering from the issues I described. I can't say every one of each suffers since my N=1 but the ones with the same characteristics and features I described above I am pretty sure that they do. There's a fundamental design flaw or we may call it a tradeoff.

I am not an idiot, I am not keeping my phone on direct sunlight neither do I run on "very high brightness". The phone runs hot on normal circumstances, and in summer when the temperatures are getting higher it becomes unbearable. I hope you understand now.

I had a similar issue with OnePlus 7 but not at this scale. It lasted me for 3, 3.5 years. I think this is becoming a problem more increasingly because of a beefier and beefier hardware that is put into these phones, and the heat dissipation problem hence becomes larger and larger problem which doesn't get automagically solved. I think that the best bet today is to take one with "subpar" CPU and larger battery and not so crazy supercharging capabilities

My last one was 10 pro and battery is essentially dead after 2.5 years of usage. Can't make half of a day, literally unusable, and I'm not a big phone user. Case is made of some really good material, which feels very premium, but runs son fckn hot that you can't hold it in your hands anymore, this is especially true during hot summers, and it got only worse with the last major OS update. This is a heat dissipation issue caused by the materials used, large battery, and hi performance CPU cores so I don't think my case was any special than the others.

I see that the OnePlus 15 follows the same route, and although it has good reviews, and they claim they solved the battery heat dissipation and detoriation issues with some new kind of cells, it seems that it still runs hot according to some reviews I've seen on the yt.

Before that I had OnePlus 7 and more budget friendly Nord, and they were much better than 10 Pro, although 7 shared similar type of issues as 10 Pro. Nord is a bit different because case is not premium, and the battery is not so large, and the CPU is not premium nor the supercharging as well. However, it doesn't run hot and battery after few years of usage is still able to give you a full day without the problem.

I'm pretty convinced that all their flagships with hi performance CPUs, premium case, large battery, and fast charging suffer from the same issues.

Maybe mixed CPU core architecture is an answer to that issue, which might suggest why is so prevailing in other phone manufacturers but I have not dig that deep into the topic

After two years your battery will be almost unusable so genuinely it doesn't matter.

My only issue with oneplus phones, and I owned several of them already, is that they are running incredibly hot on normal usage, and battery capacity detoriates quickly over time.

They do have a great sleek UI and great hardware, not to mention fantastic supercharging capabilities which is a life saver sometimes, but all under the big cost.

Sorry, what? Passing a pointer is a matter of wrapping the value into the CPU register. OTOH passing an offset into a shared memory is a write to main memory so several magnitudes slower.

Thread pools

I'd say it would be most likely impossible.

Obviously, transformers architecture is just one of the ingredients. Otherwise we wouldn't be seeing competing labs in this race. I also read all his interviews as a marketing material.

Nah, coroutines/async/etc often lives in various threads (ie, the workers can schedule one of them on different threads during the lifetime and they live concurrently), so you still have all the issues of threading (+ new ones since things like thread-local variables aren't reliable if an async/coroutines moves threads between calls).

In cooperative scheduling how is it possible to have two coroutines running concurrently?

Pickleball pays better than 40€/h

SIMD programmers have to be paid very well then in the China ... Jokes aside, some 2 or 3 years ago I thought that it is becoming inevitable for CPU designs to become an extended versions of their already quite capable vectorized execution engine units.

Yeah, speaking of bias ...

No, they don't force the model to use a json schema, they simply use the model to extract the data, and then they feed that OCR result into the pipeline further to evaluate the OCR results against the ground truth, and this is where JSON schema is used, and also another model (gpt4o).

Quite unlucky CPU to run the experiments. The article doesn't mention it but I hope that the measurement numbers OP got were extracted by re-running the experiment on same type of cores. Intel Core Ultra 7 255H is a mix of performance- (6x), efficient- (8x) and low-power (2x) cores.

I am not totally following this area but the link from a commenter from above suggests that it is not SOTA but on the lower end (but still good): https://getomni.ai/blog/benchmarking-open-source-models-for-...

That's extremely cheap. I wonder how the companies making a living out of this will survive. They certainly don't charge $0.004 per scan.

I don't think you're wrong. Virtual functions is a two-pointer dereference operation (vptr, vtable[vptr]), and there we can have a d-cache miss but the main cost of using virtual functions is the increased likeliness of the i-cache miss. Cost of 30-60 cycles as per article assumes an icache-hit, and since virtual call is an indirect call (jump), it also heavily depends on the branch-target predictor and its buffer. I can easily imagine that iterating over a heterogeneous collection of objects would incur much larger cost than ~50 cycles/iteration. Branch target misprediction flushes the whole pipeline (15-20 cycles) and icache miss can easily end up being a fetch from main memory (200-300 cycles)

The article in general is interesting since it gives a rough idea of cost of operations relative one to each other but since CPUs are much more complex beasts it also gives us an incomplete picture, and if you're unaware of it the chance is that you will use it derive incomplete conclusions from it - understanding performance implications of a software running on an actual hardware is much more involved than what one article can fit.

You're missing the point that when you're building the product, and not using SBC just for fun, the image you are building, and yes even the userspace, is highly customized. Running a full-blown Linux distro is a non-requirement, and often incorrect.

They do but the order of compiling the object files does not matter, at least not in C or C++ compilation model.

I never tackled the reproducible builds topic but this is also something I had in mind while having a thought about it. Building object files for translation units, apart from macros introducing variance, should be reproducible no matter how high the number of compile jobs is running. So, isn't this largely a linking stage problem?

And what happens when E-core decoder encounters the AVX-512 instruction set? Does this mean that E-cores will have to implement some sort of (downsized) vectorized execution engine too? Otherwise, I don't see how this would work. Migrating on the HW core level would be too complicated.

And do you know about Linux? After cursory search I couldn't find anything as well. In this case it seems very expensive, if not impossible without the huge penalty, to implement such a workaround in the hardware.

Heterogeneous compute aware kernel scheduler seems like the most natural approach to me.

Would it not be possible to solve the problem of "thread migration to an incompatible core" on the kernel scheduler level?

I don't think the open source is the differentiator here. I'm sure there's plenty of solid ideas. It would be interesting though if nlnet would open source the contributions they received in which case we would have a solid understanding how many of them have been AI or not

Lack of funding on the European ground, yes, I do see and NLNet seems to be yet another example of it. If my peers had the idea to develop some PoC, and they don't run the startup, how else would they get some pocket money to survive the year of development? All other funds are reserved for bigger and already established companies and/or organizations. And if there's one funding that the individual or few of them had a chance to get it is the NLNet. If there's more than NLNet then I'd like to hear about it.

Also, from the project list it reads as if they're anti-AI - only 3 "AI" projects selected out of 67. Considering the AI development trajectory through the course of couple of years, and especially in the last 1-2 years, it is highly likely that the largest output and innovation will exactly be in the AI domain in the next years to come. So IMO it seems like a lack of vision as I am pretty sure that there are strong individuals with great ideas but for which they lack financial support.

I live in Europe btw, and I am surrounded by people being layed off and none of them are the "bootcamp" developers. So, in order to support your view you'd really have to be more concrete, otherwise your response reads as short-viewed or even manipulative.

Very low interest and investment in the AI-related projects. I wonder if it is because there has been no strong contenders or is it because it is not part of the strategy. In any case it looks a bit awkward because I am sure there's many people who are building stuff in that domain.