Microsoft: Haswell and older CPUs are slowed down significantly by Spectre fixes

Let’s see what happens with AMD.

It’s vulnerable to Spectre even to Variant 2 but they define it as a very low risk what ever that means and only issuing fixes for Variant 1.

It looks like the second variant of Spectre the out of bounds check is the culprit here; if AMD isn’t vulnerable at all it would be great but to me it looks like they are sitting on the fence to wait and see where the exploits go.

don´t give them ideas... I see great new benchmarks with every security-incident.

And my main system is one such 4790K, which would still have been fine for a few years, but I guess I'm looking at 2019/Zen2 now.

I was thinking exactly the same thing, but regarding the OS rather than machines. It's their best excuse for making everyone upgrade to Windows 10.

CPU performance isn't important for 'typical home and business' tasks.

They're estimating 6% slowdown based on SYSmark 2014. I quickly skimmed the whitepaper and think this is probably reasonable; there's a mix of CPU-bound and I/O-bound tasks, and the CPU-bound tasks won't be impacted. Most home users are still on spinning disks, so that's your bottleneck for I/O, not the additional syscall overhead.

A 6% slowdown is imperceptible to a human. For tasks that don't need a lot of CPU power (like office applications and web browsing) even a 30% slowdown is not that big a deal.

For most of those users the affected part of the system is not a key bottleneck: syscalls such as filesystem or hardware access.

For most home/office tasks (web browsing, email, office documents) most of the time is spent waiting for user input or network responses. Even on a really slow old Atom-based netbook I suspect that the difference will be at worst "noticeable" and not significant for these tasks (i.e. it isn't going to make any difference while the app is waiting for your next key-press).

Older games that make many small calls to the graphics hardware will be affected but unless you are running them on really old hardware you'll not notice that either as they won't tax newer hardware even with the extra work. Newer games will be using more efficient techniques anyway.

Startup times are likely to be affected, both OS and individual application, but that is generally a once-per-session matter, and of course there will be some applications (or some workflows using applications otherwise relatively unaffected) that will see a difference, but not the majority.

Older OSs may be more starkly affected because of design changes since. For instance until 10 Windows performed font parsing and rendering in the kernel so text heavy applications might see some extra display lag after these patches. Low-memory situations might be more affected too, as RAM starvation increases the amount of IO happening during normal operation.

Developers are far more likely to notice, I've seen some bad before/after benchmarks for build processes. Long-running tasks, particularly those performing IO such as video encoding, are more likely to make the differences visible too (10% extra on a many-hour encoding task could add up to something pretty inconvenient). It is expected that certain server loads are going to be the worst hit. But none of this paragraph's "things that will be affected" are typical home/business use cases for the majority.

On the bright side, finally a reason for me to upgrade my desktop 3770k?

6850k has launch date 2016. So you are fine.

Microsoft claims CPUs post 2015 only have unnoticable slowdown.

Alternatively: It could be that Windows 10 has various performance optimisations that Windows 7 does not, and these have had to be undone as part of the fix.

In other words: If the fix would cause Windows 7 to lose X% performance, Windows 10 would lose that same X% plus any Y% advantage it had over Windows 7.

(This is just a guess, of course.)

I interpreted it the opposite. Maybe someone can clarify. The question is why would there be a difference between Win7/8/10 - it's afterall the same WinNT kernel architecture and has the same HAL hardware abstraction layer, it's a matured software code base, so any fix should be possible on all supported versions in the same perfection.

The original post offers some more details on this matter:

> For context, on newer CPUs such as on Skylake and beyond, Intel has refined the instructions used to disable branch speculation to be more specific to indirect branches, reducing the overall performance penalty of the Spectre mitigation. Older versions of Windows have a larger performance impact because Windows 7 and Windows 8 have more user-kernel transitions because of legacy design decisions, such as all font rendering taking place in the kernel. We will publish data on benchmark performance in the weeks ahead.

Source: https://cloudblogs.microsoft.com/microsoftsecure/2018/01/09/...

It's the opposite way around. If you read it closely they say on windows 10, some customers may see a decrease in performance, but for windows 7, most customers will see a decrease.

An article in the verge states "Windows 7 and Windows 8 will be the worst hit simply because these older operating systems have features like kernel-level font rendering that will be impacted by the Spectre and Meltdown mitigations even further than Windows 10."

From the original blog release

> Older versions of Windows have a larger performance impact because Windows 7 and Windows 8 have more user-kernel transitions because of legacy design decisions, such as all font rendering taking place in the kernel. We will publish data on benchmark performance in the weeks ahead.