SPARC Hardware

There were several SPARC adventures, but for now there is just this page. We also placed it amongst the 32-bit chips, as it was 32-bits when we were considering using it in embedded work, as well as in the SPARCstations that we had bought.

A brief history of the SPARC will help place it in our adventures. The V6 and V7 were introduced by Sun Microsystems mid 1987 after many years with Motorola’s 680x0 family. In 1990 the V8 was released, with the 64-bit V9 in 1993. This was almost ten years before AMD released their 64-bit x86-64. In 2006, the multi-core and multi-threaded T1 was released and made open source. The T2 followed in 2008. The T3 and T4 were developed after Oracle acquired Sun, and they are not open source. In June, 2009, there was only one SPARC supercomputer in the TOP500 list, ranked number 28. In 2011, Fujitsu would change that and be ranked number 1.

We bought a SPARCstation ELC in the early 1990s. We had tracked the Berkeley RISC-I and RISC-II, then the AMD29000 and Ross Semi/Cypress Semi CY7C601 processors. Fujitsu launched the SPARClite for embedded work soon afterwards.

Sun Microsystems had a range of VME boards which were sold to Force. Later the CompactPCI range was launched. Sun Microelectronics provided datasheets of their support chips (interrupt controllers, memory drivers and refresh, clock buffers and their SPARC cores). The SPARC 32-bit LEON was announced with Sun’s blessing, but prior SPARC chip vendors were not as fortunate.

SimplyRISC provided their S1 core based on a stripped down version of the T1 core. Sun’s fortunes collapsed with the infamous Dot Com bubble, but their prior treatment of Solbourne and the Taiwanese clones (after courting their business) were not good for small businesses looking to standardise on their architecture.

There were many Sun academic and internal papers on tracing, architecture exploration, profiling and performance analysis. For embedded work, the large number of register windows and keeping track of window spillage after more than seven calls deep did not seem like a simple problem to solve. Some SPARC cores ran without sliding windows and instead of overlapping calls, used a normal stack calling convention. We had little interest in modifying compilers.

Compilers, assemblers and linkers were readily available. The SPARCstation was cheap, and for debug, not bad. At the time it was much faster than the i386 PC, plus it ran 32-bit software whereas Microsoft was trolling 16-bit windows in i386 PCs.

Sun had tried to make the SPARC attractive to anyone wishing to adopt their architecture, but that only happened after it started to slide off the radar. For embedded work, it is not a good choice, even if open source. The high-end brochures looked impressive and Open Solaris had many excellent features, but Oracle, their new masters are not as nice as Sun was to the open source community. For the high-end, probably not a bad choice, but no longer on our radar.

Installation

We installed a SPARC SLC at Delta Motor Corporation for a server for the CNC data from the CAD department and the digitizer, which was then made available to PCs next to milling machines connected via Sun’s NFS. The system is briefly described in Toolroom Software.

The ELC (successor to the SLC) was damaged in shipping back from a software demonstration, and the local agents could not identify the problem. We took the system apart and found that there was a crack in one of the PCB tracks on the CRT tube, but that was not easy to fix or put back. The packaging was not meant to be opened! We repackaged the motherboard in a housing with the power supply and large SCSI disk drive. It was sold to Toyota Tool and Die for their toolroom server.