Yes agreed, that is the whole point that I was also trying to make, I guess I wasn't clear - the "up-front" costs are usually higher for Power vs x86, but Power is a much better value especially when you consider "per-core" licensing costs, which the article I mentioned specifically discusses; the article also has "hard numbers" i.e. actual IBM list prices (at the time at least) to compare against, so really no one could argue that specific point; IBM also uses some "soft" TCO arguments to argue TCO is even lower on IBMi specifically because of lower complexity, bundled database/middleware, operator/dba costs/etc but those savings are harder to quantify (even though I agree with them btw)...
Just for some background, I'm working on an article for Dinoframe.io and wanted to compare/contrast Power/Z vs x86 from several perspectives including raw performance comparisons and also TCO; there seems to be very little hard data out there to compare the three architectures, and that specifically was what I was looking for.
So if anyone does have any data comparing performance/TCO (other metrics?) between the different architectures, your input is greatly appreciated, thanks again.
------------------------------
Andrew Clark
------------------------------
Original Message:
Sent: Fri September 13, 2024 08:27 AM
From: nigel griffiths
Subject: Power vs Z vs x86 common performance metric - (according to ChatGPT) Power *destroys* the "competition"...?
I think you are missing the whole point of the attached article.
Sure the hardware price of the Power server might be higher than the x86 server.
But the processors are 2 to 4 times faster, so you need less CPUs for run a particular transaction rate.
So you pay far lower prices for the software as most Enterprise solutions are based on the CPU count.
Summary the HW costs a bit more with Power but you pay far far lower prices for the software.
Thus reducing the Power solution Total Cost of Ownership below the price of the x86 solution.
Then you have other side effects:
- higher reliability (due to reduced numbers Power of parts)
- smaller foot-print for the lower number of Power servers (less racks) and
- also far better Green Credentials with a lower electricity bill for Power.
------------------------------
Nigel Griffiths - IBM retired
London, UK
@mr_nmon
Original Message:
Sent: Wed September 04, 2024 05:40 PM
From: Andrew Clark
Subject: Power vs Z vs x86 common performance metric - (according to ChatGPT) Power *destroys* the "competition"...?
Sorry, I don't seem to be able to delete/edit the subject line or the discussion itself(?? why can't I delete my own discussion??), but basically I had done some more research before it actually got posted (I submitted it a couple of days ago) and the numbers that I posted previously are not correct, so please ignore that previous post...
What I am interested in is if anyone has numbers that show a comparable metric between Power/Z/x86? I did a bunch of research and the best I could come up with from a pure-performance perspective was:
1 CPW ≈ 7 MIPS ≈ 100 SPECint_rate
Anyone have different/better calculations?
As far as Price/performance metric, this is the only link I could find with actual comparisons between Power/x86:
IBM Power10 Shreds Ice Lake Xeons For Transaction Processing
"If it costs X to buy an X86 platform of a given capacity, then it costs 2X to buy a similarly powered RISC/Unix, Itanium, or proprietary system, and it costs 4X to buy an IBM System z mainframe"
...and then he crunches a bunch of numbers and the overall TCO is 2x in favor of IBMi over Power - very cool!
Sorry again for the premature Post...