meltdown Meltdown and Spectre patch shows significant impact on SSD performance

[N0rm]

1 hour ago, vorticalbox said:

why the exploit is getting blocked at a kernel level a BIOS upgrade probably isn't needed. Could be wrong though.

Kernel patches are a sufficient workaround for Meltdown, but not Spectre.

 

Spectre breaks security within a process, like a browser tab could read data from another if they're in the same process.

 

Vulnerable software needs to be individually updated, either to give each security context their own process (like the Site Isolation option in Chrome that gives each tab their own process), or to add code that prevents the vulnerability. The BIOS upgrade is necessary to patch the CPU microcode, which prevents the vulnerability even in software that has not been updated.

[Noirgheos]

1 hour ago, N0rm said:

Kernel patches are a sufficient workaround for Meltdown, but not Spectre.

 

Spectre breaks security within a process, like a browser tab could read data from another if they're in the same process.

 

Vulnerable software needs to be individually updated, either to give each security context their own process (like the Site Isolation option in Chrome that gives each tab their own process), or to add code that prevents the vulnerability. The BIOS upgrade is necessary to patch the CPU microcode, which prevents the vulnerability even in software that has not been updated.

So if all I really do is use Chrome to browse, Steam to play a few games, and Office for some school work, should I bother updating?

[INTELisINNOCENT]

My 4 Intel U.2 750s have actually gained performance.

[PSUGuru]

I never update my windows 10 lmao

[TheCherryKing]

I haven't noticed any performance reductions from my storage. My storage runs of PCI express.

[Bit_Guardian]

On 1/12/2018 at 3:43 AM, Zodiark1593 said:

I'll definitely be waiting on patching until the performance hits are mitigated. Neither of my systems are connected to the internet much at all, soo...

They can't be mitigated except by getting better hardware. Remember though these patches don't hamper performance for AMD since the software workarounds for Intel aren't required.

[Zodiark1593]

20 minutes ago, Bit_Guardian said:

They can't be mitigated except by getting better hardware. Remember though these patches don't hamper performance for AMD since the software workarounds for Intel aren't required.

It's a little early to be saying that, is it not?

 

Not sure what bringing up AMD has to do with me in particular. DDR4 is too expensive to justify, and Ryzen doesn't take DDR3, so AMD isn't particularly relevant for myself.

[genexis_x]

My 850 EVO has dropped some speed in 4k write with the update.

[Bit_Guardian]

52 minutes ago, Zodiark1593 said:

It's a little early to be saying that, is it not?

 

Not sure what bringing up AMD has to do with me in particular. DDR4 is too expensive to justify, and Ryzen doesn't take DDR3, so AMD isn't particularly relevant for myself.

No, as a high performance computing programmer and has worked on the Linux kernel, no it isn't early to say at all.

 

You asked for mitigations. I gave them to you.

[Zodiark1593]

10 hours ago, Bit_Guardian said:

No, as a high performance computing programmer and has worked on the Linux kernel, no it isn't early to say at all.

 

You asked for mitigations. I gave them to you.

One of my use cases does benefit from high random read speeds, so I would prefer to see how it effects others before applying any patches myself. My home internet comes from my 5 GB a month data plan, so I probably can't update even if I wanted to without murdering my data limit. Still need to download that big October update actually. This is why my desktop rarely ever sees internet btw.

 

My windows tablet on the other hand will probably get the update. Though with it already being quite slow (Intel Atom and slow MMC storage), I wonder if the patches will make it perform any worse.

 

Perhaps I could indulge a bit in grabbing a Raspberry Pi for general internet surfing now, seeing as it's immune to pretty much everything targetting high performance chips.

 

[jagdtigger]

3 hours ago, Zodiark1593 said:

Perhaps I could indulge a bit in grabbing a Raspberry Pi for general internet surfing now, seeing as it's immune to pretty much everything targetting high performance chips.

Good idea, where did i put that little sucker...

 

/EDIT

Well if you have a RPI2 expect it to be slow, and forget about FHD video... (Maybe openelec could play FHD but im not sure about it). And give more ram to the GPI in raspy-config.

Edited January 13, 2018 by jagdtigger

[Bit_Guardian]

4 hours ago, Zodiark1593 said:

One of my use cases does benefit from high random read speeds, so I would prefer to see how it effects others before applying any patches myself. My home internet comes from my 5 GB a month data plan, so I probably can't update even if I wanted to without murdering my data limit. Still need to download that big October update actually. This is why my desktop rarely ever sees internet btw.

 

My windows tablet on the other hand will probably get the update. Though with it already being quite slow (Intel Atom and slow MMC storage), I wonder if the patches will make it perform any worse.

 

Perhaps I could indulge a bit in grabbing a Raspberry Pi for general internet surfing now, seeing as it's immune to pretty much everything targetting high performance chips.

 

Some Atom's aren't affected since they don't have speculative execution. Did you find your model in the affected list?

[Zodiark1593]

48 minutes ago, Bit_Guardian said:

Some Atom's aren't affected since they don't have speculative execution. Did you find your model in the affected list?

Silvermont and after are still dual-issue, like the originals, but use OoO (and by extension, Speculative Excecution). So 2013 and later Atoms. Very big improvement over the originals, but are also affected by Meltdowns and Spectre. My tablet is based on Bay Trail, which uses the Silvermont cores.

 

This thing gets a lot of use on the internet compared to the desktop, so I'll probably have to patch it. It's likely that the storage drive is slow enough though that the patch won't make a difference even in heavy IO (cpu waiting on storage) vs the NVME drives in high end machines.

 

As for my use case, much of my gaming is done via emulation (pcsx2 and Dolphin). Some titles are startlingly io bound, with plenty of random frame drops when a game image is run from HDD. I don't have much information from the emulation community as to the impact of the patches. I considered grabbing an m.2 drive for this purpose actually. Though if I do patch the desktop and performance is notably worse, I would probably instead double my RAM (from 8 to 16 GB) and use a RAM drive to load the disc images onto, the unfortunate bit being the money is wasted during the eventual (at least a year if not much longer) move to a DDR4 platform.

 

I don't mind spending the money if necessary and the direct benefits are worth the cost, but I want what I paid for, and I don't like wasting money needlessly. If I can work around spending the money with a bit of work on my end, I will most certainly do so.

[Serin]

It certainly feels like opening blender files and reading textures is taking longer.
Most disappointing.

So, lets assume that they've been working on this patch since they found out about it... its been a while, and this is the best they could do?
Makes me worry for future performance. 'lol nope sorry this brand new 8 core is slow because hackers'. As if that's going to fly.

[Bit_Guardian]

33 minutes ago, Serin said:

It certainly feels like opening blender files and reading textures is taking longer.
Most disappointing.

So, lets assume that they've been working on this patch since they found out about it... its been a while, and this is the best they could do?
Makes me worry for future performance. 'lol nope sorry this brand new 8 core is slow because hackers'. As if that's going to fly.

Switch to ThreadRipper or Epyc if it's immediately concerning. Cannonlake was already designed and taped out, so it probably won't have a hardware fix, but I'd be surprised if Ice Lake or Tiger Lake doesn't get one, and then we get the brand new x86 on Sapphire rapids on Intel's 7nm in 2021/2022.