Slashdot Mirror
Intel Addresses CPU Shortage: 'Supply Is Undoubtedly Tight' (crn.com)
Intel interim CEO Bob Swan publicly addressed the company's CPU shortage issue for the first time since July, when he acknowledged that meeting additional demand would be Intel's "biggest challenge." From a report: In a message posted to Intel's website Friday, Swan said the "surprising return" to growth in the PC market "has put pressure on [the company's] factory network." He added, "We're prioritizing the production of Intel Xeon and Intel Core processors so that collectively we can serve the high-performance segments of the market. That said, supply is undoubtedly tight, particularly at the entry-level of the PC market."
Intel partners and at least one distributor previously told CRN they were seeing a shortage of Intel's current generation, 14-nanometer CPUs, most notably in lower-end client processors.
Intel partners and at least one distributor previously told CRN they were seeing a shortage of Intel's current generation, 14-nanometer CPUs, most notably in lower-end client processors.
And why does AMD always perform great in benchmarks but badly in real world tests compared to Intel?
That hasn't been my experience at all. The K6 blew the doors off the P2, clock for clock, when you compiled for it. Of course, no commercial software was, but if you run Linux you can compile most things yourself and reap the rewards. And the original Athlon likewise absolutely slaughtered the P3, clock for clock. The FP performance was hilariously superior. Today, AMD only outperforms Intel abusively dollar for dollar, and yeah if you need maximum single thread performance you have to go with Intel. That only matters to gamers though, since everything else that needs much performance is multithreaded now.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
Sure, now that all the good CPUs are gone, it is time for the second choice.
Are you running Windows? The Ryzen you tested use NUMA (especially the 2990WX which is four-way asymmetrical!) and sometimes fair badly because Windows supports it badly.
Maybe you used slow RAM, the AMDs perform better when you just ignore the official RAM speed and run faster RAM (the frequency of the RAM controller and the interconnects between packs of four core and between dies depend on it). Not sure if you can find fast ECC RAM though like DDR4 3200 or more. The effect is very big, there's major performance left on the table that can be seen between e.g. DDR4 2400 and DDR4 2933.
Funnily though, the Xeons W you quote (being variants of Core i8 7900X, 7920X etc. series) are lemons on single thread performance at least for something like games, because their cache latencies are slow. Maybe your workload likes a large L3. Or it fits well anyhow. Intel is also strong on things particularly SIMD heavy.
You could probably finds something where the same AMD that is 15% slower than your Intel would be 10% faster than the Intel.
Why is AMD so often trumpeted here but rarely seen in the real world? And why does AMD always perform great in benchmarks but badly in real world tests compared to Intel?
They lost their entire market share on servers and dual CPU workstation - consider that just a few years ago a dual socket Intel would beat quad socket AMD.
Now there getting beaten by 15% in a benchmark may be considered good enough.
(sockets grew bigger so 2950W/2990X or Xeon W are fit to replace older dual socket workstations like the old Mac Pros)
It's quite new, begins from zero and high end customers or OEMs didn't jump on version 1.0 hardware as is natural.
Another issue is for virtual machine farms : the hypervisors had to be updated and even if everything is updated, tested and works perfectly it will remain impossible to live migrate VMs between AMD and Intel. There were hardware bugs too but I already kind of say that refering to "1.0 hardware" (and 1.0.0.1 firmware etc.), every CPU has dozens of hardware bugs even the Intel albeit most may be obscure or easy to paper over. I tried to read Intel erratas once and understood almost nothing about what it says. OS and compiler writers may like such documents better.