Hacker News Reader

Comment by _nickwhite | Hacker News Reader

Better performance [2] with less cores, much better pricing, and the kicker for me is PCI Express 4.0 [1]. Intel doesn't even support the newest generation of NVMe storage. So yeah, AMD is dominating for a good reason- for now anyways.

[1] https://www.pcworld.com/article/3400176/pcie-40-everything-y...

[2] https://www.cpubenchmark.net/high_end_cpus.html

View on HN

The "bang for your buck" value statistics are quite stark. 17 of the top 20 are held by AMD. Intel's i9s in particular look very expensive for the performance

https://www.cpubenchmark.net/cpu_value_available.html

I think you mean "with more cores", not less.

AMD has more cores available, but also AMD's cheaper 16-core beats Intel's 18-core in many tests.

There's this really fascinating paper about how AMD has made Intel to innovate more and that AMD itself has found cheaper ways to stay on the CPU quality frontier:

https://www.its.caltech.edu/~mshum/gradio/papers/GoettlerGor...

I have to correct myself: with the presence of AMD, Intel has been innovating less than it would if the market were a monopoly.

Plus ECC support

Any motherboards with LRDIMM/RDIMM support?

Unfortunately not. That is the hard part. Rumored for the 64core and it's chipset next year.

You can get 32GB ECC at 2666mhz Dimms that work. Giving you a max of 128gb ecc on an am4

Epyc is there; you don't even have to wait to buy it. I know a lot of people are expecting Epyc features at Threadripper pricing but they also expected Threadripper to be much cheaper than it is.

Threadripper is a workstation-class part currently capped at just 8GB per core (or 4GB per hw thread). That's not terrible, but it's also not great.

Epyc has much slower boost clocks. It's far less flexible.

And people should expect features like memory support to be the same across the chips.

Is there a reason nobody will take common 3600MHz chips and put 9 on a board instead of 8? Why is all the unbuffered non-ECC memory 50% faster than the unbuffered ECC memory?

I have a hunch it might be partially due to vendors being willing to fudge their numbers for the sake of marketing.

Selling into the gaming market, label it 3600 and if it crashes, they'll just tweak the clock and keep going. Selling into the server market, customers have a different expectation.

Also many DDR-3600 modules can only run that way if they're the only DIMM on the bus. Populate two and you're down to 3333 or slower. Nobody runs a server with one DIMM, so again, the faster number simply doesn't apply to the ECC market, even if it's the very same chips.

(Get more than two on the same channel, and you need to add buffers to drive the trace capacitance, which adds latency.)

This suggests, perhaps, that you might be able to get some ECC DIMMs and overclock them all to hell and gone, if you're only running a single one per channel.

Not everyone has use for those advantages. To me they are meaningless (I have no need for "better" performance, I have no use for PCI 4.0, etc). I buy Intel CPUs because lower heat dissipation and power consumption at the performance level I need. I have avoided buying the relatively cheaper but power hungry Intel CPUs for many years.

By all means, please buy AMD. The more people do that, the better deal I get for Intel models I want to buy.

There is likely some degree of humor in your message but I'm not sure. Why are you interested in lower power consumption? To save money on the electricity bill? Or for environmental reasons?

You mention you bought more expensive CPUs to get lower power consumption... Which points to the environmental reason. But you call others to buy AMD to get cheaper Intels, which would in turn rise the global power consumption.

I guess that last sentence is to be read as a joke but as a result I'm really confused by your comment :(

Reply on news.ycombinator.com