KVR Audio

bmrzycki · Post by **bmrzycki** » Wed Jan 03, 2018 8:29 pm

EvilDragon wrote:
bmrzycki wrote:AMD's Ryzen does this too but I don't understand yet why they are supposedly immune. Everyone online is pointing to a single LKML patch as the "proof" that AMD isn't affected: https://lkml.org/lkml/2017/12/27/2 . I am not certain yet if this truly is the case.
Why wouldn't it be the case?

Two reasons:

1. If you read the sophos analysis you see the problem is a fundamental issue related to speculative execution. AMD does it (differently) than Intel does. But the problem isn't with how they do it, it's with the fact that cached lookups are near instant and non-cached ones aren't. An attacker can detect the time delta on the lookup to determine where they are in physical memory.

2. ARM posted patches related to this issue (see previous post).

Cinebient · Post by **Cinebient** » Wed Jan 03, 2018 8:40 pm

EvilDragon wrote:It's not a fixed -30% performance decrease. That one will affect virtual machine users the most. Do you use virtual machines? Then don't worry that much. 5% is the lowest stated perfromance decrease. Gaming is not terribly affected, for example. Audio drivers, though, might be. We'll need to wait for benchmarks after the fix is rolled out on Windows (for macOS, that's already in 10.3.2.)

This is not fake news, it's not a theory. The bug is real, and the fix needed to do it software-side (as opposed to fixing it on the CPU itself) has sideeffects. Newer CPUs (Coffee Lake and onwards) don't have the bug at all. CPUs that have PCID (Process-Context Identifiers instruction) will have a reduced performance hit.

I´m still on a older version and won´t update then for a while (even if i would like due to some features i could need).
But even 5% would hurt a bit since i´m already often at the limit

I know iOS and bla bla bla but i really wonder when Apple finally get away from that freaking Intel for that 5% increase at best per year and then i will loose it again maybe while these "lame" ARM processors are close to outperform them (especially if i use it for music) and might have still increase in power of 30-50% each year. Can´t speak for windows and powerful desktops but i really would like Apple to see to ditch them for notebooks.
But maybe they exactly like when this happens and the mac will be even faster obsolete and they go full iOS. I´m sure they will one day. So i have to think about where i invest my money in the future.

lalo · Post by **lalo** » Wed Jan 03, 2018 8:54 pm

EvilDragon wrote:Audio drivers, though, might be.

Finger crossed for my arriving i7 7700K with Presonus Quantum...i bought that for very good reported latency

Cinebient · Post by **Cinebient** » Wed Jan 03, 2018 9:00 pm

O.k. i had a look at the macOS update which seems to only partially address this problem if i get that right.
It might be just the big plan to want people update to new and faster computers or Apple evil plan to make more people indeed jump the PC ship to live up to their post PC evolution.
I trust no one

rsmus7 · Post by **rsmus7** » Wed Jan 03, 2018 9:07 pm

I ordered all parts for a new i7 8700k DAW PC on sunday,
today I canceled my order.

I will wait a little more to see how that bug affects DAWs

demonsk · Post by **demonsk** » Wed Jan 03, 2018 9:18 pm

from intel

https://www.theverge.com/2018/1/3/16846 ... g-response

chk071 · Post by **chk071** » Wed Jan 03, 2018 9:21 pm

Doesn't surprise me one bit, TBH. This whole thing has "scandal press" written all over it, sorry.

lnikj · Post by **lnikj** » Wed Jan 03, 2018 9:26 pm

"Contrary to some reports, any performance impacts are workload-dependent, and, for the average computer user, should not be significant and will be mitigated over time"

We aren't average computer users when we have Diva, B2 etc on multiple tracks.

Cinebient · Post by **Cinebient** » Wed Jan 03, 2018 9:34 pm

lnikj wrote:"Contrary to some reports, any performance impacts are workload-dependent, and, for the average computer user, should not be significant and will be mitigated over time"

We aren't average computer users when we have Diva, B2 etc on multiple tracks.

This. I need every bit of cpu i can get with all these great tools

bmrzycki · Post by **bmrzycki** » Wed Jan 03, 2018 9:34 pm

lnikj wrote:We aren't average computer users when we have Diva, B2 etc on multiple tracks.

The real problem isn't the CPU intensive nature of Diva or B2. The problem comes in when we have to switch to the kernel to handle an interrupt. Every time you need to send a sample buffer to your USB audio interface you need to do that. So yeah, this is a big deal for sound design people. It's also a big deal if you stream data to/from disk. So Omnisphere and Kontakt users will feel this.

bmrzycki · Post by **bmrzycki** » Wed Jan 03, 2018 9:40 pm

TheVerge wrote:Intel says it’s working with AMD and ARM in a strongly worded statement, despite AMD engineer Tom Lendacky previously saying “AMD processors are not subject to the types of attacks that the kernel page table isolation feature protects against.” Intel says it planned to disclose this issue next week along with other vendors, but that it’s issuing a statement today due to what it angrily describes as “inaccurate media reports.”

As I feared. This is big and I had my suspicions about AMD being immune. It also probably impacts server-class chips from the big boys too...

jdnz · Post by **jdnz** » Wed Jan 03, 2018 9:43 pm

bmrzycki wrote:The problem comes in when we have to switch to the kernel to handle an interrupt. Every time you need to send a sample buffer to your USB audio interface you need to do that. So yeah, this is a big deal for sound design people. It's also a big deal if you stream data to/from disk. So Omnisphere and Kontakt users will feel this.

no it's not - you may think that, but the interrupt rate for BOTH those tasks is absolutely trivial compared to a decent size SQL server which is saturating i/o on NVMe storage.

also kontakt (and presumably omnisphere) caches sample data - which even further lowers the i/o load

EvilDragon · Post by **EvilDragon** » Wed Jan 03, 2018 9:44 pm

Of course Intel would want to downplay the issue, and suck in all the competitors. I don't trust them one bit. Onset looks like a typical PR machination.

chk071 · Post by **chk071** » Wed Jan 03, 2018 9:44 pm

Someone just posted this in the "New DAW" thread: http://www.tomshardware.com/news/intel- ... 36208.html

I find this resume most notable:

We expect, and have seen already, the normal level of hyperventilation that comes with such news, but it's best to wait for more information.

EvilDragon · Post by **EvilDragon** » Wed Jan 03, 2018 9:45 pm

jdnz wrote:also kontakt (and presumably omnisphere) caches sample data - which even further lowers the i/o load

Not when you're streaming hundreds of voices on the lowest DFD buffer size from your SSDs. That's a lot of random disk reads right there. Sure it might not saturate the bandwidth fully, but it IS a lot of streaming.

'Kernel memory leaking' Intel processor -- a serious cpu bug!?!