> In a finite resources world, with unsustainable levels of pollution and soon of climate change, I don’t understand why much more of EU regulation and enforcement resources are not spent towards mandatory hardware re-use. All computer vendors know 90% of IT users never even scratch the surface of computational power and functionalities.
well, newer hardware is more efficient than older hardware, but the cost and e-waste resulting from replacing working but older hardware with new stuff is also non-zero.
desktop usage sure, it makes sense to keep it a good long time. in datacenter, for many situations the cost is not worthwhile because DDR5 is substantially more expensive for a given tier of memory, pcie5 is way more expensive to implement, etc. the newer platforms are really also higher-cost ones, due to the complete collapse of moore's law and hitting the limits of physics in link rates etc. On the other hand power does matter and datacenters are highly power-constrained etc.
it's completely application-specific, maybe if you do something that benefits from AVX-512 it's super worth it to upgrade, but for a lot of people it isn't, so it isn't something you can make a blanket regulation on when is the Right Time to upgrade.
MLID has good guests on sometimes and this is an interesting one. Just before this he's talking about the power issues ("they just can't get power into the datacenters quickly enough to keep up with needs"), and he balances this concern against the massive price factor confounding the newer DDR5 stuff.
This engineer is a good reality check on a number of sacred cows with the AMD fanbase too - for example he is excruciatingly negative on AMD's Platform Vendor Lock. He was asked if the AI market dumped if they could scoop up any cheap gear and the answer is no - they don't use GPUs currently, and they wouldn't even be able to benefit from (eg) epyc cpus being dumped because of the platform lock. They are basically e-waste (by design) once they hit the market unless the provenance is known, and even then it destroys the market efficiency (by design) since now you have separate market for Dell Epyc, Lenovo Epyc, HPE Epyc, etc. Once the value drops, surplus places won't even bother parting them out and basically the channel for that stuff dries up and they become actual e-waste.
And remember, this affects Ryzen processors now too, and platform lock is becoming much more common now as AMD makes the deals with OEM providers to get them into work desktops etc. In 5-10 years there probably won't be too much of a secondhand market left, largely because of AMD... and there's really not much that can be done since this is all hardware-locked/physically fused, short of just pushing a firmware which disables the whole thing.
He also is not mincing any words about the Sinkclose/Ryzenfall exploits where an attacker can escalate from a VM guest to jailbreak/control of the PSP and BIOS persistence. Obviously that's a huge, huge issue for datacenter operators and it's bullshit that AMD just basically decided not to patch it for older chips. The amount of handwaving and corporate defense the AMD fan club runs is silly, of course those are major issues and need to be patched ASAP.
I remember the "root password lets you do root things, where's the exploit" and other insane cope/handwaving from HUB and GN and other tech media and social media. Shockingly, the people who actually own the servers aren't as keen on a VM guest being allowed to `sudo jailbreak psp`. And AMD just wanted to leave that unpatched on a huge number of chips, even though they had a working fix for that uarch they were already deploying!
It's unfortunately the same level of security focus that AMD has given to other exploits like the cache ways vulnerability or the PREFETCH+cache eviction vulnerability ("worse than meltdown", discovered by one of the researchers who discovered meltdown), which AMD simply left unpatched and insecure, and (very) quietly told people to enable KPTI if they cared. "Insecure by default" corporate mindset.
well, newer hardware is more efficient than older hardware, but the cost and e-waste resulting from replacing working but older hardware with new stuff is also non-zero.
desktop usage sure, it makes sense to keep it a good long time. in datacenter, for many situations the cost is not worthwhile because DDR5 is substantially more expensive for a given tier of memory, pcie5 is way more expensive to implement, etc. the newer platforms are really also higher-cost ones, due to the complete collapse of moore's law and hitting the limits of physics in link rates etc. On the other hand power does matter and datacenters are highly power-constrained etc.
it's completely application-specific, maybe if you do something that benefits from AVX-512 it's super worth it to upgrade, but for a lot of people it isn't, so it isn't something you can make a blanket regulation on when is the Right Time to upgrade.
MLID has good guests on sometimes and this is an interesting one. Just before this he's talking about the power issues ("they just can't get power into the datacenters quickly enough to keep up with needs"), and he balances this concern against the massive price factor confounding the newer DDR5 stuff.
https://youtu.be/evhkvGBljWI?t=588
This engineer is a good reality check on a number of sacred cows with the AMD fanbase too - for example he is excruciatingly negative on AMD's Platform Vendor Lock. He was asked if the AI market dumped if they could scoop up any cheap gear and the answer is no - they don't use GPUs currently, and they wouldn't even be able to benefit from (eg) epyc cpus being dumped because of the platform lock. They are basically e-waste (by design) once they hit the market unless the provenance is known, and even then it destroys the market efficiency (by design) since now you have separate market for Dell Epyc, Lenovo Epyc, HPE Epyc, etc. Once the value drops, surplus places won't even bother parting them out and basically the channel for that stuff dries up and they become actual e-waste.
And remember, this affects Ryzen processors now too, and platform lock is becoming much more common now as AMD makes the deals with OEM providers to get them into work desktops etc. In 5-10 years there probably won't be too much of a secondhand market left, largely because of AMD... and there's really not much that can be done since this is all hardware-locked/physically fused, short of just pushing a firmware which disables the whole thing.
https://youtu.be/evhkvGBljWI?t=5667
He also is not mincing any words about the Sinkclose/Ryzenfall exploits where an attacker can escalate from a VM guest to jailbreak/control of the PSP and BIOS persistence. Obviously that's a huge, huge issue for datacenter operators and it's bullshit that AMD just basically decided not to patch it for older chips. The amount of handwaving and corporate defense the AMD fan club runs is silly, of course those are major issues and need to be patched ASAP.
I remember the "root password lets you do root things, where's the exploit" and other insane cope/handwaving from HUB and GN and other tech media and social media. Shockingly, the people who actually own the servers aren't as keen on a VM guest being allowed to `sudo jailbreak psp`. And AMD just wanted to leave that unpatched on a huge number of chips, even though they had a working fix for that uarch they were already deploying!
It's unfortunately the same level of security focus that AMD has given to other exploits like the cache ways vulnerability or the PREFETCH+cache eviction vulnerability ("worse than meltdown", discovered by one of the researchers who discovered meltdown), which AMD simply left unpatched and insecure, and (very) quietly told people to enable KPTI if they cared. "Insecure by default" corporate mindset.
https://www.youtube.com/watch?v=evhkvGBljWI&t=3053s
https://www.youtube.com/watch?v=7HxkLlmh4EY
https://old.reddit.com/r/hardware/comments/849paz/assassinat...
https://old.reddit.com/r/Amd/comments/8goyuq/amd_ships_cts_l...