ribit

22 ms·

181.
by ribit 4y ago
> Docker is definitely a lot slower Docker on macOS is busted. I have heard that folks have had much better success with alternative implementation such as nerdctl. M1 virtualisation on itself is very fast and has almost no overhead. &gt
182.
by ribit 4y ago
The 28-core Xeon W-3275M is similar or even slightly better than either the 5950X or 12900KS on multicore workloads, so it won't change the picture much. Overall, M1 Ultra CPU is comparable to a Zen3 32-core Threadripper if you aggrega
183.
by ribit 4y ago
C++ exception model is complex and the implementations are intricate. I do not think you one can easily dismiss the possibility of runtime overhead just because it seems like no extra work is done. It's much more subtle that that. The
184.
by ribit 5y ago
I think it’s great the kind of initiatives are becoming more widespread. Personally, I have little interest in these kind of laptops but the more choice and variety is out there the better it is for the users.
185.
by ribit 5y ago
Thanks for this! Can you elaborate a bit more and maybe give me some pointers where I can read about these things in more depth? I of course know about the MMU on traditional dGPUs but I was not aware that “integrated” GPUs also have a sepa
186.
by ribit 5y ago
There is nothing wrong with treating G13 as an IMR GPU. But might be able better performance and efficiency when doing for example image processing (or a task that maps to it conceptually) via the rendering pipeline with tile shaders. This
187.
by ribit 5y ago
Quick comment on the tile stuff: Apple GPUs basically rasterize primitives into small tiles (16x16 or 32x32 pixels), tracking primitive coverage and barycentric coordinates per pixel. Once all primitives are rasterized, the entire tile is s
188.
by ribit 5y ago
The only problem: Apple GPUs are general purpose GPUs and other GPUs have bigger memory constraints than Apple G13. It is kind of funny that they complain about the small cache on M1 Ultra when 3090 Ti has 10x less cache
189.
by ribit 5y ago
The author of those tweets doesn't seem to understand what TLB is (their terminology is off), doesn't seem to understand how GPUs work (they are explicitly designed to hide memory stalls by switching to a different kernel, which i
190.
by ribit 5y ago
Monomorphisation is a double-edged blade. Sometimes keeping the code smaller and hot is better than inlining everything, especially when your application does not exclusively own all the system resources (an assumption that many “systems pr
191.
by ribit 5y ago
I think you have all of this right. Parts of IMG IP Apple still uses is the TBDR stuff, compression and some other fixed-function things. It is unclear how much of it Apple has modified. The actual compute and scheduling engine is Apple&#x2
192.
by ribit 5y ago
Sure. But it does not affect you as developer in any form or fashion. It's just a thing the linker does. You can still distribute your binaries, disassemble them etc. as you ever would.
193.
by ribit 5y ago
By using the provided system tools that report power usage of the GPU cluster? Also, I am telling you the system diagnostics output if an actual physical machine. What are you quoting Wikipedia for? You can literally go measure these things
194.
by ribit 5y ago
Well, it seems like you made up your mind without doing any testing or educating yourself, so I am not quite sure what I can do to help you. Already your entirely nonsensical comment of “needing to sign the executable” speaks volumes. Why d
195.
by ribit 5y ago
“Official sources” in this case is testing done by me personally. I am the author of the post on RWT linked previously. I would be happy to share my benchmarking code for your scrutiny if you want. The M1 variant tested was base M1 in a 13
196.
by ribit 5y ago
G13 (in the 8-core/1024 ALU config as in M1) delivers 2.6TFLOPS with sustained power consumption of 10W. That's almost an order of magnitude better than 1030. Sure, node definitely matters, but going from 14nm to 5nm cannot explai
197.
by ribit 5y ago
1030 has been dethroned a while ago. Apple G13 delivers 260GFLOPS/W in a general-purpose GPU. I mean, their phone has more GPU FLOPS than a 1030.
198.
by ribit 5y ago
ET-minion can complete with GPUs because it is essentially a GPU. Most GPUs are in-order RISC machines with very wide SIMD ALUs and huge register banks. Whether it runs RISC-V or some proprietary ISA doesn’t really matter. How you build thi
199.
by ribit 5y ago
No, you are right. It’s just frustrating that there is almost no information on this topic. The only data I was able to find is a very short paper from 2017 that compares code density for some simple use cases. The results suggest that Aarc
200.
by ribit 5y ago
Thanks for replying, much appreciated. I admit that my interest primarily goes towards desktop application and my comment was made from that angle. Regarding conditional instructions on in-order microarchitecture: I neither have a strong op
201.
by ribit 5y ago
I have difficulty following the points the author is trying to make. - Even with instruction compression the type of code they present will take more space than, say, Aarch64. - The entire section on conditional execution doesn't make
202.
by ribit 5y ago
Ah, I see. True, Rust has a more restricted model that allows it to track lifetimes more precisely. As far as I know, Clang will omit reference counting if it can be statically determined but of course it has fewer optimization opportunitie
203.
by ribit 5y ago
What exactly do you mean by compile time GC? Various techniques like compiler-optimized GC have been around for a while. If you are looking for mainstream stuff, Apple’s automatic reference counting for Objective C comes to mind (has been s
204.
by ribit 5y ago
Anandtech has been one of my primary sources for quality tech journalism and unbiased reviews, mostly thanks to the works of Ian and Andrei. In the industry where folks report the TDP as actual power consumption these guys were true beacons
205.
by ribit 5y ago
Pretty much that. Safety gimmick aside, Rust is (mostly) what C++ would have been if designed with modern knowledge of programming languages theory and compiler practice in mind. It’s not like Rust brings that much novel things to the table
206.
by ribit 5y ago
I think Rust’s hole purpose is to be an ergonomic language that supports modern practices and generates very fast executables. Safety is a nice thing to have, but that’s just an icing on the cake for folks who need it. For me at least, a pr
207.
by ribit 5y ago
Something that is not quite clear to me yet (I did read the discussion below, thank you Hector for indulging us, very informative): isn't the end behaviour up to the drive controller? That is, how can we be sure that Linux actually doe
208.
by ribit 5y ago
That is true. Proton is a pragmatic solution. But it hardly a satisfactory solution, and I worry that it is not sustainable solution either. I do not really have a horse in the run personally, as I am not a Linux user (my platform of choice
209.
by ribit 5y ago
I was not mentioning Proton explicitly. But the core of my point is that Proton de-facto delegates Vulkan to be a second class citizen whose purpose it is to serve as a backend of implementing a Microsoft-compatible technology stack. With P
210.
by ribit 5y ago
I vehemently disagree with sentiment expressed in the article. Sure, on the surface the work done Valve does significantly improve the state of Linux gaming. But this is not Linux gaming. This is Windows gaming under Linux. We need to thi

More ›