Oracle Announces New T5 CPUs, But SPARC Fading From The Market
Oracle is launching a new set of SPARC processors it hopes will staunch its market share losses, even as revenue continues to slip. The new T5 chips are built on a 2nm process and refine the T4 core that Oracle debuted two years ago. Like the T4, the T5 emphasized improving single-thread performance and clock speeds, though the new core does double the number of cores per processor to 16, up from eight.
The SPARC T series from Sun/Oracle has always interested me, because it's an explicitly multi-threaded design that emphasized parallelism. After Intel announced the end of single-threaded scaling in 2005, Sun's early T1 processors flipped entirely in the opposite direction, emphasizing multi-threaded performance at the expense of single-threaded workloads. Long term, however, this didn't work. The SPARC T1, T2, and T3 designs were eventually judged to have prioritized multi-threading performance too highly -- while they excelled in certain narrow workloads, they weren't capable of meeting a broad spectrum of customer needs.
Oracle's T4, built on 40nm technology, addressed these flaws by sharply increasing clock speeds, reducing the number of threads per core, and boosting IPC. It was the first T Series chip to support out-of-order execution and it was 3-5x faster than the T3 depending on workload. The T5 refines this further. Built on a 28nm process, Oracle has boosted the T5's clock speeds to 3.6GHz (up from 3GHz) and switched to dual PCIe 3.0 interfaces (at x8), up from the PCIe 2.0 x8 interfaces that T4 used.
According to the datasheets, Oracle's T5 is up to 30% faster than the T4 thanks to a new FPU pipeline and lower memory latency. Oracle has published a number of aggressive claims of benchmark wins. IBM's response? "Yeah, right."
The Most Valuable Peanuts, Ever
Here's a bit of an eye-opener for you. The high-end server market is tiny compared to the PC consumer space, and the RISC market SPARC plays in is a tiny fraction of that. 9.5 million x86 servers shipped in 2012. The number of RISC servers (including IBM's Power, Itanium, and SPARC): 154,870. In such a small market, it's a little surprising to see Oracle attempting to destroy HP's Itanium business or IBM slamming Oracle's benchmark claims for the T5.
As IBM's Elisabeth Stahl writes: "Just as with every Oracle processor announcement, the benchmark results do the same thing. Most of the claims are Oracle’s own benchmarks that are not published and audited. There are a small number of industry standard benchmarks — and of course these are ones where it is extremely difficult, if not impossible, to compare to other relevant results. For price claims, Oracle, as they’ve done in the past, only factors in the price of the pizza box – make sure you add in the all-important software and storage."
Indeed, market evidence suggests that Oracle's best efforts haven't even slowed the company's decline. Hardware sales have fallen 22% in the first nine months of Oracle's FY 2013. That's after falling 13% in FY 2012 and 11% in FY 2011. Oracle's own lawsuits against HP and bad-faith attempts to disrupt the other company's business make more sense in this context -- companies that can't compete in the market typically resort to lawsuits as a backup strategy.
If the situation doesn't change soon, T5 may be Oracle's last hardware hurrah. The company has been de-emphasizing x86 sales in favor of pushing its own proprietary solutions -- and that's obviously not working.
The SPARC T series from Sun/Oracle has always interested me, because it's an explicitly multi-threaded design that emphasized parallelism. After Intel announced the end of single-threaded scaling in 2005, Sun's early T1 processors flipped entirely in the opposite direction, emphasizing multi-threaded performance at the expense of single-threaded workloads. Long term, however, this didn't work. The SPARC T1, T2, and T3 designs were eventually judged to have prioritized multi-threading performance too highly -- while they excelled in certain narrow workloads, they weren't capable of meeting a broad spectrum of customer needs.
Oracle's T4, built on 40nm technology, addressed these flaws by sharply increasing clock speeds, reducing the number of threads per core, and boosting IPC. It was the first T Series chip to support out-of-order execution and it was 3-5x faster than the T3 depending on workload. The T5 refines this further. Built on a 28nm process, Oracle has boosted the T5's clock speeds to 3.6GHz (up from 3GHz) and switched to dual PCIe 3.0 interfaces (at x8), up from the PCIe 2.0 x8 interfaces that T4 used.
According to the datasheets, Oracle's T5 is up to 30% faster than the T4 thanks to a new FPU pipeline and lower memory latency. Oracle has published a number of aggressive claims of benchmark wins. IBM's response? "Yeah, right."
The Most Valuable Peanuts, Ever
Here's a bit of an eye-opener for you. The high-end server market is tiny compared to the PC consumer space, and the RISC market SPARC plays in is a tiny fraction of that. 9.5 million x86 servers shipped in 2012. The number of RISC servers (including IBM's Power, Itanium, and SPARC): 154,870. In such a small market, it's a little surprising to see Oracle attempting to destroy HP's Itanium business or IBM slamming Oracle's benchmark claims for the T5.
As IBM's Elisabeth Stahl writes: "Just as with every Oracle processor announcement, the benchmark results do the same thing. Most of the claims are Oracle’s own benchmarks that are not published and audited. There are a small number of industry standard benchmarks — and of course these are ones where it is extremely difficult, if not impossible, to compare to other relevant results. For price claims, Oracle, as they’ve done in the past, only factors in the price of the pizza box – make sure you add in the all-important software and storage."
Indeed, market evidence suggests that Oracle's best efforts haven't even slowed the company's decline. Hardware sales have fallen 22% in the first nine months of Oracle's FY 2013. That's after falling 13% in FY 2012 and 11% in FY 2011. Oracle's own lawsuits against HP and bad-faith attempts to disrupt the other company's business make more sense in this context -- companies that can't compete in the market typically resort to lawsuits as a backup strategy.
If the situation doesn't change soon, T5 may be Oracle's last hardware hurrah. The company has been de-emphasizing x86 sales in favor of pushing its own proprietary solutions -- and that's obviously not working.
