Intel Kernel Memory Flaw - "Looking at a Ballpark Figure of 5-30% slow down"

[Jozape]

Did this affect the performance on personal machines or just server architecture?

Technically, any machine using an Intel CPU that is not an Itanium is affected. Most of the time this is insubstantial for single-player games and web browsing. Multi-player games can be severely affected on the server side though. I read that some disk heavy activities on flash disks can be significantly slowed down, but I don't know about mechanical disks. If you don't run virtual machines, you don't run a server, and you don't do disk heavy activities, you are probably looking at a minimal performance drop for anything where performance is actually meaningful nowadays.

[Zoraptor]

There will need to be BIOS updates to fully implement protection as well, so the exact performance loss is still unclear at this point. But yeah, certainly shouldn't be much if anything for games since they're almost always limited by the graphics card power rather than CPU; and most basic productivity and home computing tasks are not taxing on modern CPUs either.

[injurai]

It would make turns in Civ V longer. Just because often rigs are gpu limited doesn't mean cpu specific tasks won't take a hit. In fact most animation is done one the cpu.

[Zoraptor]

There aren't many turn based games, and even fewer where there is a significant AI 'turn' to take into account, hence "almost all" games being GPU limited by the ability to draw and fill polys and apply effects. It may also have an adverse effect on frame rates if you're playing at 720p low details with a 1080Ti, but that combination is also pretty rare and practically only used if you're trying to make a point or doing CPU testing.

[Gorgon]

Yeah, so what about Photoshop, Zbrush, 3dsMax, Unreal Engline,  that sort of thing. A lot of people use those on their regular desktop.

 

Well, some anyway.

[Zoraptor]

They all depend on how much strain they put on the CPU- what you're using the program for, how demanding the program and its functions are, whether it's multithreaded or not and whether it uses the GPU and for what. If there's a different bottleneck than the CPU there shouldn't be much effect unless it hits one of the other areas Meltdown/ Spectre fixes effect like certain types of disk functions. From what I remember Photoshop is badly optimised and has no effective multithreading at all and little (only with a couple of functions, iirc) GPU based acceleration so I'd suspect it would be adversely effected especially for intensive processes; IIRC the others all have some degree of multithreading andor GPU acceleration built in so are likely to be effected less, dependent on what exactly you're doing with them.

 

It will also be dependent on what your hardware is, a well threaded program on an 8700k may be slightly slower than before but still faster than any other non enthusiast platform (though a very well threaded app on a r7 will likely be faster since SMT>>HT and the r7 has 2 extra cores). OTOH something a little bit older, less powerful and with fewer threads is likely to take a bigger hit since it's more likely to be struggling already.

Edited January 26, 2018 by Zoraptor

[Zoraptor]

Apparently there's (yet) another security flaw in Intel chips- 'Spoiler' (that's its name, not a warning for GoT S8 leaks etc). Similar attack vector to Spectre/ Meltdown but this one's a real doozy as well, since it seems likely it cannot be fixed except via direct hardware alterations and can at least theoretically be exploited via javascript.

 

AMD and ARM chips not thought to be effected, albeit they only tested an archaic A6 on the AMD side.

Edited March 5, 2019 by Zoraptor

[Bartimaeus]

That seems like a lousy writeup, but I guess that's no surprise. Literally the opening line says it affects only first generation Core processors which immediately made me go "yay", but opening up the actual paper says it affects all Intel CPUs *starting from* the first Core generation.

[Zoraptor]

The effected memory system dates back to the pentium pro (!) so in theory everything since- apart from Itanium- should be effected.

 

Choice of not so good writeups I'm afraid as at the time I posted none of the sites I'd usually cite had anything available.

[Zoraptor]

And another one. Fixable by yet another microcode update that will effect performance. Intel's HT really is an utter cluster.

Note: contrary to some articles (hello Tom's; though to be fair they've fixed it pretty quickly) almost all modern Intel processors [with hyperthreading, if it wasn't clear from the top line] are effected, but Intel is not recommending that everyone disable hyperthreading, only those at high risk.

Edited May 14, 2019 by Zoraptor

[Malcador]

Curious how come there are so many, guess I will have far too much noise to sift through and my comp org book to find before I can answer that for myself, heh.

 

https://www.redhat.com/en/blog/understanding-mds-vulnerability-what-it-why-it-works-and-how-mitigate-it

More info.

Edited May 14, 2019 by Malcador

[injurai]

At this point I just assume every advantage Intel still has is a massive trade-off against security.

[Malcador]

Wonder if any of these are related to performance (thus being a trade off) or are just oversights in design.  Rough times for Intel, but I guess there has to be valleys.

[injurai]

There is always security oversights if history tells us anything. I think it's obvious they chased the market with the designs they pushed. There is a reason why AMD had been stuck scratching their heads for so long.

[Zoraptor]

That was mostly on AMD for pushing a new architecture that was slower (!) than their old one and betting on more cores being critical too early. In terms of out and out speed advantage I haven't seen much suggesting that corners were cut specifically for speed with hyperthreading- and ironically, AMD's more secure SMT implementation is also a decent amount faster than Intel's HT. Most of the flaws seem to be 'buffer overflow' or not checking/ clearing the entirety of prediction memory space type issues that would presumably impact performance a little, but not much. And when it comes right down to it, AMD's implementation seemingly does (most) of those things properly while being faster.

Plus the newer Intel processors with mitigations for the first batch of vulnerabilities don't seem to have any detectable IPC drop. Intel do seem to be rather negative towards Hyperthreading, but it's hard to tell whether that's due to security or attempts at market segmentation.

Edited May 15, 2019 by Zoraptor

[injurai]

I didn't know about AMD's architecture managing to eke out performance over Intel's in those areas. Buffer overflow is often very much handled with the OS and compilers, most of the exploits recently seem to be around the timings of various inter-instruction value forwarding that can be exploited. Either by leaking information, or disrupting what should be atomic operations. Branch prediction is especially susceptible to this, as is the whole ILP chain. Especially with supposed "secure" sub-modules that may only exist for nefarious purposes in the first place.

[Malcador]

The fact that JavaScript is a method by which an attack like this can be made is sort of depressing.

[Malcador]

https://www.phoronix.com/forums/forum/phoronix/latest-phoronix-articles/1100545-spectre-meltdown-l1tf-mds-mitigation-costs-on-an-intel-dual-core-ht-laptop

Effect of mitigations.  Must be interesting times at Intel's engineering department in all this.

[Zoraptor]

700% increase in context switching time? That's going to be fun for some users.

[213374U]

Intel CSME bug is worse than previously thought - Researchers say a full patch requires replacing hardware.

(except if you want to jailbreak your CPU...)

[Malcador]

Would be nice to see how their internal teams are handling this.

[Zoraptor]

They probably wish they could just overtly blame the NSA for the problem.

I presume at some stage we're going to get books and articles about what's been going on at Intel in the same style as there have been about Boeing's decline. Certainly seems like Krzanich's tenure as CEO was a complete disaster on the technical side that they still haven't recovered from, even if it isn't significantly hurting their bottom line yet.

[Keyrock]

I fully expect Intel to bounce back, and it's not like the company is in dire straits, they're still profitable. That said, it's been a rough stretch of years for Team Blue after a long period of holding a near monopoly. In addition to all the security exploits, 10nm has been an unmitigated disaster for them. They're what, 4 or 5 years behind schedule and counting at this point?

[Zoraptor]

5 years.

Intel is probably too big to fail even if it were seriously threatened, much like Boeing. The big medium term threat is if their 7nm gets delayed or runs into 10nm like problems, they've now acknowledged they'll be behind AMD for nearly the next two years which is bad enough. But they've had problems now with their last two processes, albeit the Broadwell pipecleaner was far less an issue than 10nm, and if they make it 3 in a row there will be serious questions about them maintaining their own fabs, let alone the sort of damage further delays could do to their most lucrative markets.