Recent content by bit_user

The article I linked states that the ATX 3.0 certification includes the following tests: 200% of the PSU’s max-rated capacity for 100us 180% load for 1ms 160% load for 10ms It's either conformant or it's not. Look for Cybenetics ATX 3.0 certification, to know whether a PSU model is...

I agree with most of your post, but it's a common fallacy that a high-wattage rating correlates with PSU quality or that a good PSU can't safely handle running at 100% of its rated power output. Good power supply reviews will often test how a PSU behaves under its maximum rated load and any...

Given how hallucination-prone LLMs have so far been, I think it's especially important to ensure the information they provide is accurate and verifiable. This is very concerning. Sounds good, but the obvious catch is who defines "bias". That's why I think a focus on verifiable accuracy is...

Those weren't old, just not perhaps originally planned for the desktop. However, they were RDNA2, just like the rest of the RX 6000 lineup. So, not an example of mixing generations, much less re-releasing an existing model number with newer generation silicon! Okay, but AD106 was RTX 4000...

The bizarre thing is how they're using ADA silicon in a RTX 3000-series part that only ever previously used Ampere. I'm pretty sure they never did that with the GTX 1030. Why are they doing it?? Are they trying to avoid price erosion of the remaining RTX 4000 cards on the market? Are they...

Oh, just desktop graphics doesn't take much from a GPU. I have a little AMD Polaris RX 550 from 2017 running a 1440p monitor @ 144 Hz. Even running Google Earth at max quality is nice and smooth on it! At work, I'm still using a GTX 1050 Ti on a 4k monitor @ 60 Hz. Again, it's more than...

Two words: memory wall. It traditionally refers to the growing speed disparity between DRAM and CPU cores, but we're also seeing a dramatic slowing in DRAM density scaling and especially GB/$. So, I don't know that we can entirely afford to say "damn the bloat". At least, not like we used to...

LPDDR5 had to be soldered down, until we got LPCAMM. With regular SODIMMs, the signal integrity wasn't good enough. SPECint rate=1 says otherwise. But yes, the generational single-threaded speed gains aren't as big as they were 20+ years ago. The proliferation of containers is adding quite a...

Now you're talking about ARM's GPUs, still not MediaTek or what they used. I think they modified them only a little. After a while, maybe not at all? "The Snapdragon 835, switched to using ARM’s Cortex A73 and A55. To frustrate anyone looking for information on the original Kryo, Qualcomm...

Do you know how many US states have laws that effectively prevent these sorts of arrangements from providing internet access?

I thought of this, but most laptops sold in the US don't have integrated 5G wireless. Maybe that's a bigger deal, for other markets? I think they used to be one of Imagination's customers, but dropped them before Apple did? Qualcomm switched to licensing ARM's CPU cores sometime around 2017...

Yes, here's one: https://rushpcb.com/ Perhaps there's not much sense in onshoring manufacturing of the lower-value parts, but some of these can certainly be made here. In something like a laptop, I think most of what's not integrated into the SoC are related to power-handling and some of the...

A relevant question might be: "for whom?" Do you care whether they hire US citizens and pay competitive wages, vs. getting migrant workers on H-2A visas to staff them? It seems to me that, if you have to bring in people to do the job, it's not much different than just manufacturing abroad.

The answer to this is probably too political.

Should we read anything into how "fat" it is? 🤣