New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Intel Hardware Bug
Well, better keep the popcorn ready:
https://news.ycombinator.com/item?id=16046636
-30% performance on some CPUs, let's see how much of it is true.
Comments
If this turns out to be exploitable all i can say is nice one Intel. First AMT now this. Well done.
Given that AMD seems to be ‘safe’ and the fixes introduce a significant performance overhead, it could be great for AMD in the server market.
Intel has some issues since Ryzen, then they fucked up 2 times, now China comes into the race with x86 cpu's. This gonna be good.
If only they for once decided to focus on the power consumption. Customers aren’t eager to pay more for AMD either.
@Clouvider I haven’t really looked into the new EPYC range much so I don’t really know how they compare performance per watt wise. Is the gap really still that big?
The big EPYCs are expensive enough (like big E5's) that the silicon cost dominates the power bills.
We tested them on Bios 1.0 and returned. Even the Idle power consumption was dramatically higher than Intel with the only difference being the CPU and the Motherboard. I haven’t checked since, but my SM aff Manager is not pushing them hard any more which implies that they still don’t fall in line with our requirements.
I disagree, investment is a one off cost, but power + cooling is very much recurring
I hope AMD benefits from this, since the only useful thing Intel did in the last 6 years is increasing efficiency, adding new instruction sets and higher (turbo) clocks. If we factor out the increase in clock speeds (approximate IPC comparison), there is less than 5% net performance gain in over 6 years, which is beyond ridiculous.
Even the crappy Bulldozers were in fact faster than Sandybridge for their intended workloads, but many people believe games or shitware from Adobe is appropriate for benchmarking. Of course, typical single core workloads were far worse on bulldozers, but it was not nearly as bad as Intel's fuckups in the past 6 years.
Perhaps this (along with Zen 2, POWER9 and A11) will put more pressure on Intel. But I agree that the recurring operational costs are more important for businesses and AMD will hopefully further increase their energy efficiency, which is their biggest weakness in my opinion.
If the cpu uses 100W of power on average (more when running at 100% but it's idle some of the time), then over a 3 year lifetime it will use around $500 worth of electricity at budget data center rates, depending. The top Epyc processor is $4000+ and the top E5 processor I believe is $8000+. At those prices the cpu cost dominates. Obviously the math varies depending on the cpu. But even the top consumer Threadripper is around $1000.