Single core benchmark cpu11/25/2023 The Alpha, SPARC and MIPS all ran up to 8x faster. Up until 2002, we see a huge difference in floating-point performance between mainstream and workstation CPUs. After that, it leveled off at the same 21% per year. Prior to 2004, it climbed even faster than integer performance, at 64% per year: a doubling period of 73 weeks. Here are the results, which I’ve also adjusted to eliminate autoparallelization: Floating-point performance has always been important for heavy-duty computation such as scientific simulation or 3D rendering. SPEC also publishes SPECfp, an equivalent benchmark for floating-point performance. So far we’ve only looked at integer performance. Bulldozer, their latest microarchitecture, is meant to shine in multithreaded workloads. This is not too surprising, as AMD is pretty open about their stance on single-threaded performance. If you remove all Intel CPUs from the data, a different picture emerges: Here are a few machines located along the red line in the graph: Hardware AvailabilityĪs you can see, Intel deserves credit for squeezing out the most single-threaded performance since 2004. Compare that to the 28x increase between 19! Things have really slowed down. If you’re willing to trust this line, it seems that in the eight years since January 2004, mainstream performance has increased by a factor of about 4.6x, which works out to 21% per year. I drew it manually, using the less-than-scientific method of eyeballing the points for Pentium, PowerPC, Athlon and Core. The red line is meant to represent mainstream CPU performance. The graph incorporates results from three different benchmark suites (CPU95, CPU2000 and CPU2006), but I’ve normalized the results in order to see historic trends. It consists of 5052 test results from 715 different CPU models, all gathered over the last 17 years:Įach test result is plotted according its hardware availability date, and the vertical axis uses a logarithmic scale. 7, grouped the results by CPU brand, and generated the following graph. I’ll share the method at the end of this post, and you can let me know if you think it’s valid. I took some pains to remove those results from the dataset. That sounds perfect, except for one catch: many licensees use automatic parallelization. One of their benchmark series, SPECint, was designed to measure the single-threaded integer performance of a machine. SPEC licenses their benchmarking software to various companies, collects results back from those licensees, and makes those results available on their website. ![]() It’s the same benchmark used to plot a few data points on the above graph. I turned to SPEC, an industry-standard benchmark that’s been going strong since 1989. So I decided to crunch the numbers myself. Moreover, I couldn’t find another source to corroborate them. These figures aren’t really consistent, and both struck me as a little low. ![]() Last year, Chuck Moore of AMD presented this graph, suggesting that single-threaded CPU performance recently started going backwards: Bill Dally of nVidia threw out a few numbers in a recent presentation: He had predicted 19% per year, but says it’s turned out closer to 5%. Everybody acknowledges that single-threaded CPU performance no longer increases as quickly as it previously did – but at what rate is it actually increasing? So, what’s happened since 2004? Clearly, multicore computing has become mainstream. Concurrent software would continue its meteoric rise, but single-threaded software was about to get left in the dust. Not all software will run remarkably faster year-over-year anymore, he warned us. Later that year, Herb Sutter wrote his now-famous article, The Free Lunch Is Over. Perhaps the turning point came in May 2004, when Intel canceled its latest single-core development effort to focus on multicore designs. But eventually, at some point in the mid-2000’s, progress slowed down considerably for single-threaded software – which was most software. Your 486SX-16 was almost obsolete by the time you got it through the door. Throughout the 80’s and 90’s, CPUs were able to run virtually any kind of software twice as fast every 18-20 months.
0 Comments
Leave a Reply.AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |