5

Recently I've been looking into some SPARC devices and their capabilities. But when I look for the differences between the two, I've seen a lot of people using the Ferrari vs bus metaphor

If your goal is to get 2 people from point A to point B as quickly as possible, then choose Intel. If your goal is to get 100 people from point A to point B as quickly as possible, choose SPARC.

Although a lot of benchmarks these days show that x86 outperforms SPARC in most of the cases, quite a lot of people(whom I saw on the internet) who use SPARC still believes that this is true and benchmarks do not reflect reality.

And Oracle still produces newer SPARC processors: T4, T5 and sells them for a huge amount of money compared to typical x86 servers. I wonder what kind of benefit people can get from using SPARC over x86 in 2015 and whether the metaphor is still correct.

5
  • The only time I've seen sparc in the wild is in legacy systems that are on Solaris and cost too much to move to any other platform. That's also why they're so expensive - smaller market share means less demand which means a higher price. Mar 20, 2015 at 2:48
  • I believe that SPARC also scales out better than x86 does. I've seen big iron SPARC machines marketed with 1024 cores at > 3GHz. Good luck getting that on x86. Mar 20, 2015 at 2:51
  • 3
    That Sparc T5 runs 128 threads /per socket/ (where x86 would be maybe 40 if you went big), and you can get 8-way systems. That's a crazy number of simultaneous jobs [numbers courtesy en.wikipedia.org/wiki/SPARC_T5 ]. Will your software run on it? Maybe.
    – Bill Weiss
    Mar 20, 2015 at 3:29
  • 1
    Another point: Oracle uses a great deal of time putting "software in silicon", i.e. they migrate operations from their db software into the SPARC chip. The SPARC boxes have always been good at I/O (a bit like what mainframes are good at too) and if you couple that with the "software in silicon" trend it means you can get superior db performance on a SPARC box compared to a X86 box.
    – peterh
    Mar 22, 2015 at 16:57
  • 1
    Does this help? It's a bit dated, but may still be useful.
    – ckujau
    Apr 9, 2015 at 4:10

2 Answers 2

3

Some would argue the metaphor stillholds true, others would argue otherwise.

There's clock speed, but you also need to factor in RISC vs CISC.

What people get, or hope to get with SPARC, is a more stable platform with less downtime. Tecoms were historic SPARC consumers due to nebs compliance.

As noted in comments above, it scales better, so you'll tend to see larger systems. Along with possibly products that require more resources. ie: A database in memory. Some of the newer SPARC CPUs are also much more power efficient -- power and cooling is a big deal in datacenters.

Someone noted "software in silicon", in the comments. Oracle expanded on Sun putting crypto units in the CPUs, by adding database instructions, and plans on adding features.

In the end, the architect(s) will determine what is needed for the system. Chances are it will be a mix of solutions that best meet the requirements.

On the CPU cost. AMD has kept Intel in check over the years. Oracle appears to be trying to be cost competitive to x86, as well as it's RISC competitors.

2
  • 1
    Not questioning you, the wikipedia article clearly says it's RISC... But isn't adding encryption, database, and other complex instructions the opposite of RISC? Why is it not considered CISC? Jan 20, 2018 at 2:50
  • ...and all of Intel's x86/x64 chips are now based on a RISC microcode, making x86 an abstraction anyway.
    – Dai
    Jan 4, 2021 at 5:16
0

No question that x86 single thread performance is better than T5 SPARC. So much so that SPARC is now a poor choice for most database workloads.

I'd post the benchmarks to prove it but Oracle licensing rules effectively prevent that.

1
  • Are you saying that no one is even allowed to post benchmark results? That is just crazy. Feb 5, 2020 at 14:36

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .