Thursday, May 16th 2019
Apple: Protecting Macs from MDS Vulnerabilities May Reduce Performance by up to 40%
Apple has advised users that they should disable Intel's Hyper-Threading feature on the company's computers due to the recently exposed MDS vulnerabilities. Citing internal testing, Apple said that users can expect an up to 40% performance loss in such a scenario (depending on system and workload, naturally) in various benchmarks and multithreaded workloads. The performance loss is understandable - you're essentially halving the number of threads available for your CPU to process data.
Like Intel said, it becomes an issue of how much users value their performance compared to the security risks involved: a classic risk/benefit scenario, which shouldn't ever be in the equation, after all. If users buy a system with a CPU that has known performance levels, they will obviously expect those to be valid for the longevity of the product, unless otherwise stated and considering operational variances that fall within a margin of error/product obsolescence. Halving your performance because of a design flaw that resulted from an effort to achieve higher and higher IPC increases doesn't strike as a way to inspire confidence in your products.
Like Intel said, it becomes an issue of how much users value their performance compared to the security risks involved: a classic risk/benefit scenario, which shouldn't ever be in the equation, after all. If users buy a system with a CPU that has known performance levels, they will obviously expect those to be valid for the longevity of the product, unless otherwise stated and considering operational variances that fall within a margin of error/product obsolescence. Halving your performance because of a design flaw that resulted from an effort to achieve higher and higher IPC increases doesn't strike as a way to inspire confidence in your products.
23 Comments on Apple: Protecting Macs from MDS Vulnerabilities May Reduce Performance by up to 40%
Still, no one says you can't put your render farm on a subnet, NAT it to the internet for updates only, and not give anything on the outside direct access to it.
But yeah... them javascripts and leaky processors. Might go AMD this next round, but as soon as I do there will security flaws exposed there so meh...
Fixed the bug that "old generation CPUs being too powerful it affects the sales of new machines"
Adjusted CPU performance like we did on battery performance for a better user experience.
The problem is there is no legal remedy. If a decision or statute existed, most cars on the market today would be exposed as long term fuel economy even performance drop due to direct injection carbon build up on intake valves. Vacuum cleaners would be vulnerable as they slowly clog over time. Keurig coffee makers clog over time. There is a magical belief a processor should perform like new for the life of the product. It is a consumer product just like the products above. Hence, consumable and by statute degrades with time. The only legal remedy to this stuff is if Intel willingly knew about it and, if asked, didn't disclose or conspired to withhold the info. Nothing requires them to disclose voluntarily even if they knew.
Not flaws that affect performance.
>>The performance loss is understandable - you're essentially halving the number of threads
>>available for your CPU to process data.
>>...
This is Not right.
I'm a C/C++ Software Engineer and I've been using Intel HTT since 2007 year. In a period from 2007 until 2014 on different R&D projects and since 2014 until now on production quality HPC projects. So, when Intel HTT is disabled number of Logical CPUs equals to number of Physical Cores of a CPU. Depending on an Intel architecture when Intel HTT is enabled it increases number of Logical CPUs. For mainstream architectures by 2x, for KNL architecture by 4x. When Intel HTT is disabled a Processing Power ( PP ) of an HPC-like application with correct threads management is The Same! Period! This is because PP is a function of Physical Cores and is Not a function of Logical Cores.
I think Apple Corporation is talking about UI-like application and these applications never do correct thread management similar to what is done in HPC-like applications. If you ever worked with ESRI ( geoinformatics ) or AGI ( aerospace ) software you will understand what I'm talking about.
Processing Power ( PP ) of a CPU is calculated as follows:
Base Clock in GHz * Number of Physical Cores * Number of SP/DP elements per vector * Number of FP ops per Clock
For example, for Intel Core i9-9900K:
3.6 GHz * 8 * 16 * 2 = 921.6 GFLOPs
and it will be 921.6 GFLOPs regardless state of Intel HTT!
Search the Web with a Tag "Intel Hyper-Threading Technology and Processing Power of a Computer System"