by MattiasNYC » 21 Jan 2018 19:39
Fly Studio wrote: ↑
21 Jan 2018 18:47
I also read that it is the Spectre Variant 2 patching that is causing the biggest performance impacts on systems; since the patches can't 'fix' the problem, they are using 'voodoo' techniques to mitigate it; these techniques are causing the performance hits.
I don't recall reading or hearing that. As far as I know it is the patching of Meltdown that's causing potentially large performance drops, not the other two.
I read so many whitepapers, discussions, benchmarks, etc. that I don't remember exactly
what led me to conclude that, but on one of the tech sites (Tom's Hardware, Tech Republic, Zdnet, etc.), they tried to assess the impact of the patch in question by running a series of SSD benchmarks; they had Reference (pre-patch) results, OS update results and OS & BIOS update results. The only Variant that requires a BIOS microcode update is Spectre Variant 2; that is most likely why I made that assumption. The performance was varied with different test parameters (and the system was brand new with an NVMe SSD). Since the OS updates attempt to mitigate all 3 Variants and ASUS is unlikely to release a BIOS update for my P8-Z77 V Pro motherboard, it is presently irrelevant to me which Variant mitigation strategy is likely to impact performance (if future updates are more targeted to specific Variants, I will revisit the question).
My first comment about OS developers taking advantage of the CPU design was purely speculative, & not an attempt to point the blame at them. Obviously Intel is responsible for the security flaw. I was just wondering if it occurred to the OS developers that taking advantage of the methods in question (e.g. Speculative Execution) might have potential security risks.
I understand that many DAW users have never run a benchmark or would even know how to interpret the results, but some here are up on these things. On one thread, someone said they loaded a 'Stress Test project with 17 instances of Ivory running to compare pre-update & post update results (they saw a very small performance drop, but their OS & CPU were newer, AFAIK); I have screen captures of AAS Benchmarks from each of my SSDs taken at 3 month intervals for comparison, as well as other test results archived. I'm sure others here have similar practices to maximize DAW performance.
BTW, I (apparently) successfully uninstalled both the Cumulative Update KB4056984 & the NET Framework Update 4055532 and regained the performance lost with both my SSDs and other applications (e.g. Desktop operations were slower after the updates - they returned to their snappy state after the rollback); I had just cloned my OS to a new SSD, did benchmarks and DAW tests, then installed the updates - then I read about the possible performance impact, so I had an opportunity to experiment since I had the original OS SSD intact & offline.
I am a musician first & recording - mixing - mastering engineer second; my computer knowledge is long but sporadic and I don't claim to be an expert, so anything that wasn't quoted from a reliable source is speculative to some degree. I.E. I am not here to have a *quiz* contest about these issues. Simply want to find out what people are experiencing, especially when using a massive amount of sample streaming from SSDs, which is the bulk of my work.
Many of my projects reach the upper limit of the 16 GB of DDR3 Dram I have, so any hit in SSD read performance is a big concern for me; if I have to raise the buffers in Play 5, for instance, I will run out of memory with some of the projects I am working on.
I used the term 'voodoo' since software patches are being employed to deal with a low level hardware problem. I also am taking into account the changing landscape of computing; 25 years ago, security was a very different (and largely minimal) concern. I doubt many software (and hardware) designers accurately saw the future trajectory of security and the extent that the 'Net' would have in the future; how many predicted online banking and teams of hackers in foreign countries when DECs were the big boys on campus? And even those DEC CPUs had these basic 'flaws', according to several industry articles I've read.
MattiasNYC, Hope this clarifies my previous post & your point by point response to it. it seems you know more about current OS architecture than I; that's fine by me. I've had my fill of dealing with the tech; and I surely don't mind having an inaccurate statement being corrected. If fact I appreciate it. I could have been more specific and precise, but my post would have been way too long, so my comments were largely 'shorthand'. Besides, this is a 'new' issue and I expect the available info will continue to change for a long time to come.
Fun fact: My band Machine Screw was the first group to multicast a live performance over the Internet (Sept. 7, 1994), using software developed at Cornell and modified - implemented by an Austin ISP. The forerunner of the live podcast I suppose.