You can not just use IBM Mainframe MIPS to compare against normal x86 MIPS. So that is not really comparable. Also, IBM marketing will try to make it difficult to compare performance between x86 and Mainframes. For instance, it would be easy to run Dhrystones / SPECint / SPECfloat / any other benchmark on a Mainframe, but you will never see such benchmarks published. IBM marketing would never allow publication. But we know that whenever IBM has anything advantageous to say, IBM says so. So, why are IBM not publishing common benchmarks run on IBM Mainframes? IBM Mainframes crush everything in terms of performance, right?
The new IBM Mainframe Z196 cpu, is the "Worlds fastest cpu" according to IBM:
It has crazy specs. 5.26GHz, ~300MB cache (L1+L2+L3). Close to half a GB of cache! So, how fast is it? Let us see. It is 50% faster than the previous Z10 cpu. How fast is a Z10 Mainframe cpu? The biggest Z10 Mainframe with 64 cpus, give you 28.000MIPS. Now, that is a lot! How much is it?
If you use an 8-socket Intel Nehalem-EX server, and use the IBM Mainframe software emulator "TurboHercules", then you get 3.200MIPS:
That is under software emulation, which is 5-10x slower than running native code. So, an 8-socket Nehalem-EX server would actually give 5-10x higher performance if you ported the Mainframe applications, so we could run native code. That is: 16.000-32.000 MIPS. Thus, an old 8-socket x86 server would give you as many MIPS as the biggest Z10 IBM Mainframe with 64 cpus. If you build an Emulator for IBM Mainframes, then you know the ins and outs of Mainframes.
Another developer that ported Linux to IBM Mainframes came up with this rule:
1 IBM Mainframe MIPS == 4 MHz x86.
So, the biggest IBM Mainframe Z10 with 64 cpus, correspond to 28.000MIPS == 112.000MHz x86. But, a single 8-core Intel Nehalem-EX running at 2.5GHz, has 20.000MHz in total. Thus, you would just need a few Nehalem-EX to reach 112GHz. Again, we see that you need a few Nehalem-EX to beat 64cpus of the Z10 Mainframe.
And, Here is a source from Microsoft
"we found that each [z9] mainframe CPU performed 14 percent less work than one [single core] 900 MHz Intel Xeon processor running Windows Server 2003."
The z10 is 50% faster than z9, and the z196 is 50% faster than z10, which means a z196 is 1.5 x 1.5 = 2.25 times faster than a z9. This means a z196 corresponds to 2.25 x 900MHz = 2 GHz Intel single core Xeon. But todays modern server x86 cpus have 8 cores, which means they have in total 8 cores x 2 GHz = 16 GHz. Again we see that you need only a few 8-core Nehalem-EX to match the biggest z10 Mainframe.
Now the Z196 cpu is 50% faster than the z10 cpu. Instead of using 8-socket x86 servers, you need to use 12-socket x86 servers to beat the biggest Z196 IBM Mainframe, utilizing the "worlds fastest cpu". In short, if you can migrate your workload to x86, then do it.
Sure, IBM claims their Mainframe can virtualize 1.500 x86 servers. If you read the entire study, you will see that IBM assumes all x86 servers to idle, and the Mainframe is 100% loaded! Now what happens when a powerful x86 server does some work? The weak Mainframe can not virtualize that load. The Mainframe cpus are to weak.
Sure, IBM claims the Mainframes have brutal I/O and can handle 400.000 logged in users. But I would not be surprised if 95% of the users idle and only a few of the users do some work.