HN user
I think it is well established that repeat COVID infections have an increasingly severe impact on your body.
It is also true that COVID can weaken your general immune system.
But that doesn’t mean that getting COVID won’t increase your immunity against a future COVID infection, temporarily at least, just the way the effectiveness of the vaccine is temporary.
I have a million miles on United alone, and many more on others as well: I’ve never once heard such a message either.
Same here.
OpenAI commented on the verboseness of answers as something they need to work on.
A subscription model where you know the exact terms is less exploitative than a fixed cost where it’s not even clear if the upgrade is sticky or not (as is currently the case for Tesla.)
I don’t get the “it’s not a service” argument. It’s not relevant. There’s no ongoing cost with many software licenses either. It’s just a business model.
It’s exploitative if it is not sticky and isn’t clearly communicated. But it’s not relevant for the Mercedes case where it’s very clear that it’s a subscription.
For my personal case, it still doesn’t matter, of course.
My Model Y has a $2000 in-app purchase option to increase acceleration by something like 0.5s.
It’s not a subscription, but the idea is the same.
It doesn’t bother me one bit: I bought the car knowing full well that it wasn’t included, I don’t need it (the standard acceleration is already more than I ever had before), end of story.
Car engines have been under the control of software for decades now, with different products differing by the program. The only difference here is that there’s now the option to change the program over the air.
Nobody would have complained if Mercedes had offered 2 versions without the option to upgrade.
If it works for a celebrity, it also works in case of a calamity.
Reminds me of the "Can I move? ... I'm better when I move" scene...
Being an owner and being an investor are orthogonal concepts.
It would make zero difference, except, maybe, for the PCIe bandwidth test. Because all other tests were stressed interfaces that reside within the GPU silicon itself or between the GPU and the DRAM.
Even the numbers of PCIe bandwidth test might not change much, even it's trying to test medium size memory to memory block transfers.
I’d go dither than that and claim that almost everything of importance has checksums: Google Docs, Google Sheets, Git submissions, all my important web accounts, the traffic with my bank website and so forth.
When I look at my daily home computer usage, it’s remarkable how little I calculate on my local computer that’s actually with protecting.
Why pay for something I don’t need when there’s the option to not pay for it? - me
Just yesterday, I bought an extra 64GB for my home Linux PC. I absolutely couldn’t care less about it crashing or calculating the wrong result every blue moon (in practice: never), but I did choose the RAM sticks that were $10 cheaper.
That seems like a stretch? Isn’t the more obvious explanation that laptops with a real GPU are much more expensive and that the weaker, integrated GPUs are more than good enough for the vast majority of business use?
Today’s iGPUs are fast enough comfortable run plenty of games.
I have work provided high-end POS Dell Precision engineering laptop. It has an Nvidia discrete GPU, but I don’t think I’ve ever actually needed its power, and I’d gladly trade it for a laptop without…
I was wondering about that. According to Nvidia RTX 4090 product page, it still used PCIe 4.
Moving around data is indeed a major issue for any throughout oriented device. But for a gaming GPU, PCIe BW has never been an issue in any of the benchmarks that I’ve seen. (Those benchmarks artificially reduce the number of PCIe lanes.)
In fact, the 4000 series still has PCIe 4.
Moving data around for a GPU is about feeding the shader cores by the memory system. PCIe is way too slow to make that happen. That’s why a GPU has gigabytes of local RAM.
But why would game programmers care about shader core latency??? I seriously don't understand.
We're not talking here about the latency that gamers care about, the one that's measured in milliseconds.
I've never seen any literature that complained about load/store access latency in the shader core. It's just so low level...
GP100 and GP GeForce has a different shared memory structure as well, so much so that GP100 was listed as having 30 SMs instead of 60 in some Nvidia presentations. But the base architecture (ISA, instruction delays, …) were the same.
It’s true tbat GA102 has double the FP32 units, but the way they works is very similar to the way SMs have 2x FP16 in that you need to go out of your way to benefit front them. Benchmark show this as well.
I like to think that Nvidia’s SM version nomenclature is a pretty good hint, but I guess it just boils down to personal opinion about what constitutes a base architecture.
Thanks for that.
The weird part is that this latency difference has to be due to a terrible MC design by AMD, because there's not a huge difference in latency between any of the current DRAM technologies: the interface between HBM and GDDR (and regular DDR) is different, but the underlying method of accessing the data is similar enough for the access latency to be very similar as well.
I don't agree.
Turing is an evolution of Volta. In fact, in the CUDA slides of Turing, they mention explicitly that Turing shaders are binary compatible with Volta, and that's very clear from the whitepapers as well.
Ampere A100 and Ampere GeForce have the same core architecture as well.
The only differences are in HPC features (MIG, ECC), FP64, the beefiness of the tensor cores, and the lack of RTX cores on HPC units.
The jury is still out on Hopper vs Lovelace. Today's presentation definitely points to a similar difference as between A100 and Ampere GeForce.
It's more: the architectures are the same with some minor differences.
You can also see this with the SM feature levels:
Volta: SM 70, Turing SM 75
Ampere: SM 80 (A100) and SM 86 (GeForce)
Density, I can accept.
But what kind of latency are we talking about here?
CDNA has 16-wide SIMD units that retires 1 64-wide warp instruction every 4 clock cycles.
RDNA has a 32-wide SIMD unit that retires 1 32-wide warp every clock cycle. (It's uncanny how similar it to to Nvidia's Maxwell and Pascal architecture.)
Your 1/4 number makes me think that you're talking about a latency that has nothing to do with reads from memory, but with the rate at which instructions are retired? Or does it have to with the depth of the instruction pipeline? As long as there's sufficient occupancy, a latency difference of a few clock cycles shouldn't mean anything in the context of a thousand clock cycle latency for accessing DRAM?
Holy false equivalence, batman!
They're defending taking advantage of a miner's forced fire sale. Less waste, lower price for a good GPU. What's not to like?
AMD's decision to have different architectures for gaming and datacenter is still a major mystery. It's clear from Nvidia's product line that there's no reason to do so. (And, yes, Hopper and Ada are different names, but there was nothing in today's announcement that makes me believe that Ada and Hopper are a bifurcation in core architecture.)
I'd be more than happy to by a 3080 (or similar) for a bargain price, knowing that it has been run at lower voltages and power levels.
As for waste: I very much hope it will not become waste. Why would you be advocating for that?
How many giraffes is that?
With no virtual memory, no caches, and interface processors instead of direct access to external DRAM, this thing must be a programming nightmare?
Having tons of small CPUs with fast local SRAM is of course not a new idea. Back in 1998, I talked to a startup that believed it could replace standard cell ASIC design with tiny CPUs that had custom instruction sets. (I didn't believe it could: it's extremely area inefficient and way to power hungry for that kind of application. The startup went nowhere.) And the IBM Cell is indeed an obvious inspiration.
But AFAIK, the IBM Cell was hard to program. I've seen PS3 presentations where it was primarily used as a software defined GPU, because it was just too difficult to use as a general purpose processor.
Now NOT being a general purpose process is the whole point of Dojo, so maybe they can make it work. But from my limited experience with CUDA, virtual memory and direct access to DRAM is a major plus, even if the high performance compute routines make intensive use of shared memory. The fact that an interface processor is involved (how?) in managing your local SRAM must make synchronization much more complex than with CUDA, where everything is handled by the same SM that manages the calculations: your warp issues a load, it waits on a barrier, the calculations happens, sometimes in a side unit in which case you again wait on a barrier, you offload the data and wait on a barrier. And while one warp waits on a barrier, another warp can take over. It's pretty straightforward.
The Dojo model suggests that "wait on a barrier" becomes "wait on the interface processor".
If you don’t want a computer on wheels, you don’t want a German car, you want a second hand Toyota Corolla.
We had an Audi Q5 first and a Tesla second. My wife and I fight over who gets to drive the Tesla.
With a modern German car, you still get some the fancy features, but they don’t work very well (lane control comes to mind), and the electronic UI is much worse than Tesla. CarPlay is the only way out.
And let’s not even talk about the difference in driving fun.
As a counterpoint, my company has outsourced major parts of its internal IT support to the Philippines, yet my experience is always as good as how it was before they did so.
If they can do that for internal support, it should be possible for external as well?
Putting a check in the mail (or pressing some button on a web form to make it so) is one of the things that comes to mind.
A friend on mine quit his FAANG job, took a one year sabbatical, and went back on the job market. He found a remote-only job at a boutique bank in NYC, negotiated his compensation, and then told them he'd only accept the offer if it was 4 days instead of 5 at the same salary. They accepted without any pushback. He totally loves his 3-day weekends.