Intel Core i9 9900K Linux Benchmarks - 15-Way Intel/AMD Comparison On Ubuntu 18.10

[ddriver]

Curious them power figures, as other reviews put the 9900k power usage at anywhere from 50 to 100% higher than then 2700k...

[ddriver]

When fed optimized code, zen has higher ipc than current intel *lake cpus (I've seen 3,5 (zen) vs 3 (kaby lake) reported by "perf" on a special testcase compiled with clang). But intel still has better memory controller (lower overall latency) and much more refined power management, coupled with a very mature manufacturing process (and better chipsets. the amd ones are meh at best, especially the asmedia usb controllers are garbage).

IPC wise, amd is pretty much on par with intel now. Intel has lower memory and execution latency because amd takes a hit from the modularity of the zen architecture. Zen 2 is likely to boost the core count per CCX to 8, thus allowing to both put 16 cores in AM4 but also produce an 8 core chip with a single CCX, thus evading the inter-ccx latency penalty.

The chipset is irrelevant, as everything important, that is the "north bridge" is integrated into the CPU - memory controller, PCI-E and whatnot, the rest is just fairly low demand peripherals that do not have any effect on system performance.

Another thing to take into consideration is how much of those benchmarks are tailored to one way or another give intel an edge. It might be optimizations, it might be software architecture design, it might even be the choice of workloads to be incorporated. The depth of intel's pockets is a sad fact of reality, and deep pockets mean long reaching hands. Don't take benchmark results as indication of the absolute ability of a CPU to perform. As someone who runs independently produced HPC software, I tend to find that the bulk of the most popular benchmarks out there to be significantly skewed towards making amd performance look worse than it is. Of course, absolute performance is also kinda irrelevant when you don't have the software to leverage it, but it is important to keep this in mind when talking about how the chip can perform from a purely hardware standpoint.

[msroadkill612]

.

Michael, as the ability to multitask increases and people do more simultaneously with their systems, have you thought about setting up benchmarks with multiple tasks in parallel? So for instance, web browsing to multiple sites (stored locally to eliminate network variability) with concurrent download and unzip of a zip file while a music or video player is running all while an antivirus is scanning in the background. That might be a more meaningful comparison for modern concurrency and usage of CPUs than single application stress tests. I think it might also provide some additional differentiation between AMD and Intel due to the higher context switching in this sort of scenario.

yes, u verbalise it better than I, but I have long thought the benches miss this important point.

Its not the end of the world if app A & app B don't parallelize well, and app C doesn't scale well beyond 4 cores, IF (realistically as u say), users are using all 3 apps, and perhaps multiple instances.

That blinkered view also conversely over prioritises IPC - the ability to do a single non scalable task fast, to the exclusion of all else,

It happens of course, but in a biz environment, most would be happier with reserve cores that can be loaded up with tasks w/o slowing.

Its a common sentiment from power users after bedding in their 8 core ZenS - "its a new mindset - u can just keep giving it more jobs & it doesn't slow down."

Zen sales indicate the market is getting jaded about the IPC premium (~8700k vs 2700x - mobo & cooling factored in) for 6 cores w/ slightly better ipc vs 8 core Zen.

Even keen gamers love that they can get good performance, while running money making apps concurrently. Does it matter much if a compile is a bit slower, if it doesn't mind if you game concurrently?

[5parks]

I recently purchased this same setup: the Asus Prime Z390-A and Core i9 9900k. It's running great under Linux with one exception: as the review mentions, sensor support in mainline Linux is a problem. Has any discovered a way to expose this board's sensors?

Thanks