I won't argue he's wrong, but I think as fast as CPUs change you'd have to have across the board reductions in workload capacity by a significant number (ie, the 30% touted initially) to be able to claim harm and justify a recall.
So far what I've read is that performance in some storage applications has taken a small hit, but generally it's not a meaningful hit overall in CPU performance.
And even if you could prove a performance hit, how reliable are the numbers? Workloads vary a lot. And if you're not operating at the limit of your hardware, is going from 45% idle CPU to 25% idle CPU really a performance "loss"? It's arguable that you had surplus capacity to begin with and that unless you frequently max out your CPUs, you didn't really lose any capacity.
Datacenters which already patched are apparently getting a performance hit in the high two-digits. Epic, for example, has people complaining [epicgames.com] because users cannot connect to game servers after the patch doubled servers CPU load.
It's hard to understand if this is Epic finding a convenient whipping boy for their problems or an actual problem associated with the patch. There's always complaining about gaming server performance, outages, etc, so it's not like they couldn't have other problems and that this is just being used as an excuse.
Idle CPU uses less power. When CPU idle time is lowered, that directly translates into additional power consumption. Remember some high-end CPUs have TDPs of 100W or more at full tilt. Now take a whole datacenter of those, decrease their idle time by 25% and you'll see a substantial increase on your power bill.
Now take a whole datacenter of those, decrease their idle time by 25% and you'll see a substantial increase on your power bill.
And in your mind, what percentage of a server's power usage is tied up in the CPU? I'd speculate about 20% after you figure in RAM and drives.
So if the CPU is 20% of the total power of the server, and cpu power consumption increases 25% of the 30% low-power idle time (7.5%), that means the servers power usage will increase 1/5th of 7.5%, or 1.5%...
If a 1.5% increase in server power usage is a "substantial increase" you can overcome that increase by running your data center 2-3 degrees warmer.
I won't argue he's wrong, but I think as fast as CPUs change you'd have to have across the board reductions in workload capacity by a significant number (ie, the 30% touted initially) to be able to claim harm and justify a recall.
Man, have you ever drunk the software caveat-emptor Kool-Aid whole cloth.
The specifications for hardware and software are not that it will work as specified.
The specification is that, with sufficient user cleverness (and sweat streaming off a bulging forehead, and unbroken vigilanc
"I have just one word for you, my boy...plastics."
- from "The Graduate"
"I want repaired processors for free" (Score:5, Insightful)
You know, he's not wrong. This is, in impact, way bigger than Intel's FDIV fiasco and that ended up in recalls.
Re:"I want repaired processors for free" (Score:2)
I won't argue he's wrong, but I think as fast as CPUs change you'd have to have across the board reductions in workload capacity by a significant number (ie, the 30% touted initially) to be able to claim harm and justify a recall.
So far what I've read is that performance in some storage applications has taken a small hit, but generally it's not a meaningful hit overall in CPU performance.
And even if you could prove a performance hit, how reliable are the numbers? Workloads vary a lot. And if you're not operating at the limit of your hardware, is going from 45% idle CPU to 25% idle CPU really a performance "loss"? It's arguable that you had surplus capacity to begin with and that unless you frequently max out your CPUs, you didn't really lose any capacity.
Re: (Score:2)
Datacenters which already patched are apparently getting a performance hit in the high two-digits. Epic, for example, has people complaining [epicgames.com] because users cannot connect to game servers after the patch doubled servers CPU load.
Re: (Score:3)
It's hard to understand if this is Epic finding a convenient whipping boy for their problems or an actual problem associated with the patch. There's always complaining about gaming server performance, outages, etc, so it's not like they couldn't have other problems and that this is just being used as an excuse.
Re: (Score:2)
Well, it is likely. Still, that graph is horrible to look at.
Re: (Score:2)
the posted CPU utilization graphs look pretty obvious :/
I think they were on the edge and this shoved them over, but now they need to spin up more servers and deal with that, plus killed margin.
Re: (Score:2)
Idle CPU uses less power. When CPU idle time is lowered, that directly translates into additional power consumption. Remember some high-end CPUs have TDPs of 100W or more at full tilt. Now take a whole datacenter of those, decrease their idle time by 25% and you'll see a substantial increase on your power bill.
Re: "I want repaired processors for free" (Score:2)
Now take a whole datacenter of those, decrease their idle time by 25% and you'll see a substantial increase on your power bill.
And in your mind, what percentage of a server's power usage is tied up in the CPU? I'd speculate about 20% after you figure in RAM and drives.
So if the CPU is 20% of the total power of the server, and cpu power consumption increases 25% of the 30% low-power idle time (7.5%), that means the servers power usage will increase 1/5th of 7.5%, or 1.5%...
If a 1.5% increase in server power usage is a "substantial increase" you can overcome that increase by running your data center 2-3 degrees warmer.
Increased power
the solder contract (Score:2)
Man, have you ever drunk the software caveat-emptor Kool-Aid whole cloth.
The specifications for hardware and software are not that it will work as specified.
The specification is that, with sufficient user cleverness (and sweat streaming off a bulging forehead, and unbroken vigilanc