On the server, AMD is making a security play w/ EPYC arch.
- AMD not vulnerable to Meltdown
- SEV allows encryption of a VM’s memory, further mitigating read attacks
On the consumer side, AMD is selling on-package Vega to Intel. This should increase market penetration, and thus get more games optimized for the AMD arch.
SeV will not help in this regard.
Games are already optimized for AMD GPUs and pretty darn well due to consoles I’m not entirely sure how much difference would the KBL-G provide considering that at least in the mobile space it’s going after the MX150 it needs 100W to compete with the 1050 so that would be on desktop only.
Spectre is also much much harder to mitigate against Meltdown sucked but it was easily fixed and it will also be easily fixed in silicon Spectre especially variant 1 will be with us for a long time and no one knows what still is lying hidden.
AMDs main advantage is that none of the researchers really bothered reversing their branch predictor so none of the BP training reliant PoCs work on their hardware but there is nothing in their design that mitigates against Spectre specifically.
And judging by the guide they’ve released for how to manage branch prediction post Spectre at least the technical team is worried about what would happen when someone starts paying attention to them.
This would also be a pretty straightforward mitigation for Intel, so the whole "they need to redesign the whole core from the ground up!" thing may be a bit overblown. Cache behavior will need to change a bit, for sure, but we probably don't need to smash the looms and go back to using non-speculative ARM cores.
I tend to think AMD probably has some vulns lurking in their uarch as well, the difference is that Intel has a massive amount of marketshare in the server world and their hardware gets shaken down first as a result. I'm just a random internet nobody but Anders Fogh said the same thing in an interview with Gamers Nexus: Spectre is not much less severe for AMD than Intel, these are the opening shots of a whole new class of attacks and there are more to come now that we are looking for them.
This is incorrect. Spectre works at a cache level, memory is decrypted by that point - if it were encrypted you couldn't use it for the good kind of speculation either. This means SEV doesn't do anything in particular here, effectively the victim is decrypting its RAM for you.
The point of SEV is to make memory inaccessible after shutdown, whether that's physical (cold boot attacks) or virtual (VM instances reusing host memory). Not for Spectre.
Let say AMD makes a profit on their Vega Chip sold to Intel.
Intel will now have to sell this CPU+Vega chip ( Raw Cost ) @ Intel Margin.
So Intel either sell a very expensive package, or sell it at the same cost ( or slightly higher ) to customer as if they were to buy them separately, which hurts Intel margin. ( Because Intel earn less on Vega )
And it has been a long time Intel dont care about margin at all.
But yes, even still, Kaby-G is a very expensive part. I've heard ballpark like $500-600 for just the package. You can pretty much buy a whole Ryzen 2500U laptop for just that one item on the BOM. But it's also much faster - roughly 4x as fast as a 2500U.
Intel has already made a move to rework their GPU development by effectively starting over(see: recruiting Raja Koduri to work in the same kind of leadership position he had at AMD previously while developing Vega). But they need a stopgap for the coming years, and this is one way to attack.
The Vega M package is a technically interesting product(size, integration, possible economies of scale), but it's primarily a win for market share because Intel can cram it down the throat of OEMs to stop them from using Nvidia. They would happily cut their margins to do so.
Let hope they do.