Intel claims each core can perform 16 FLOPS per cycle, at least at SP. Each core has a 512-bit wide vector unit. I'm not sure where their DP claims are coming from, though.
The whole point of the Niagara was to provide zero-impact context switches and to effectively hide latency, and to get close to 100% utilization as a result. On embarrassingly parallel workloads (web serving was the one Sun hyped hardest) a 32-thread T1 or 64-thread T2 did quite a bit better than its Intel contemporaries. The problem is that a lot of workloads expect to be able to consistently issue more than 200 million instructions per second per thread to do well, even when you are hiding latency by doing fast thread switches. Databases, a workload you cite as parallelizing well, tended to run like crap on Niagara.
With Bulldozer, you effectively have 16 2-issue cores, since each module has a shared 4-instruction decoder. I'm skeptical that it will perform all that much better on multithreaded integer workloads than the T3 did, which had 16 1-issue cores with aggressive multithreading and latency hiding on top. On the other hand, Westmere-EX is 10-core, 4-issue, 2-way SMT-capable per core, and has big caches with good latency numbers. On both a technical level and based on early benchmarks (SPECcpu), things don't look good for Interlagos.
They both have the same issues, including that each module (two 4-issue cores) has a single 4-instruction decoder in front of it. Cache latency is also likely to be similar if not the same.
It doesn't matter. If the shares were promised, they were promised - if there was no intention to deliver, they shouldn't have been promised in the first place. Honor is honor, whether or not you think the other person "deserves" it.
Re:What Sun built in goodwill, Oracle destroys.
on
Solaris 11 Released
·
· Score: 2
The VIIIfx and IXfx are uninteresting for commercial applications due to the irritating fact that they don't support SMP.
The IBM Power 795 usually outperforms the M9000, and you're comparing a 64-socket machine to a 32-socket one.
Your evaluation of the T4 is actually much worse than the reality - it's a significant improvement over the T3. But the lack of speccpu or TPC-C benchmarks is interesting.
Re:What Sun built in goodwill, Oracle destroys.
on
Solaris 11 Released
·
· Score: 2, Interesting
I think I have. I've seen that the "latest and greatest" SPARC64 VII+ still gets regularly spanked by Power7 and Itanium and even commodity systems in performance, despite being considerably more expensive - and I've seen vague roadmaps for the future of M-Series. I've seen that the T1/T2/T3 performance promises never really panned out (see: SPEC results vs the much cheaper Magny-Cours), and that the T4 has so far largely been hidden behind the veil of vague benchmark-fu while being far more expensive than its competitors.
What hardware have I been missing?
Re:What Sun built in goodwill, Oracle destroys.
on
Solaris 11 Released
·
· Score: 1
AIX at least runs on decent hardware. SPARC sucked for years before the acquisition, and continuing to beat that particular dead horse seems unwise.
Palm never really did much with their BeOS code, and HP doesn't even own it (PalmSource, makers of the Access Linux Platform, do.) webOS was killed by a fickle market and the incompetence of HP and Palm, not by any kind of curse in an unrelated product.
Microsoft has stated that they won't sue over C# usage, and they've demonstrated no inclination whatsoever toward doing so. Nobody has ever presented evidence that they will - only vague accusations about how EVILLLL Microsoft is.
Mainframe market share is huge (about 3 billion a year in hardware sales) and growing rather quickly, especially since "cloud" and "virtualization" became buzzwords. It's the only part of the high-end non-x86 niche that's really having solid growth right now - SPARC and Itanium have been tumbling for a while, and Power has been more or less flat.
People tolerate it because it works. Mainframes have compelling performance characteristics, especially for virtualization- or I/O-heavy workloads, and most people don't need a full unlocked processor (a CP.) Linux is the fastest-growing OS on z, and a Linux specialty engine (Integrated Facility for Linux) is relatively cheap compared to the other types of specialty engines. zAAP/zIIP/CP only really matters if you're running z/OS, in which case you're probably a large enough company tat you can afford it, or if you're running z/VSE, in which case you're probably only using one or two processors anyway.
The summary misses something fairly important, which is that Windows isn't running on the z mainframe itself. This allows Windows blades to be inserted into an external chassis (zBX) and managed by a software component called the Unified Resource Manager.
Re:The next new airplane to get axed...
on
The F-35 Story
·
· Score: 1
I'd say they do have the money. They're continuing production of the Tu-160, which is easily the best strategic bomber in the world, and if I recall they're ordering several hundred PAK FA as well as modernizing their Flanker fleet to the Su-35 standard, which should make it one of the best 4+-generation fighters out there.
Re:The next new airplane to get axed...
on
The F-35 Story
·
· Score: 1
Russia is building the PAK FA 5th-gen fighter.
Re:The next new airplane to get axed...
on
The F-35 Story
·
· Score: 1
The F-22 had a few issues, namely that Congress wouldn't permit it to be exported and it wasn't really suitable for carrier-based ops (no STOL/VTOL capability.) The F-35 was supposed to be the successor to the F-16 as the near-standard fighter for NATO and non-NATO US allies, and provide a replacement for the F/A-18 for the Navy.
The T3 is crap for anything compute-intensive. The whole point was "throughput computing," not processing power, and even at that it's pretty poor. Look at the speccpu benchmarks sometime.
My Vaio Z doesn't boot from USB. That's incredibly weird, since it's a "premium" high-spec machine, but it makes using Linux a pain in the ass. (So does their disregard of TRIM in favor of a custom SSD garbage-collection system, and their proprietary switchable graphics, and their out-of-the-box RAID 0'd SSD's, and...) It's like Sony had a serious case of NIH syndrome.
Slashdot Mirror
Intel claims each core can perform 16 FLOPS per cycle, at least at SP. Each core has a 512-bit wide vector unit. I'm not sure where their DP claims are coming from, though.
Or at least the "right" to squat on someone else's property for several months while defying eviction orders.
Zucotti Park is privately owned.
The whole point of the Niagara was to provide zero-impact context switches and to effectively hide latency, and to get close to 100% utilization as a result. On embarrassingly parallel workloads (web serving was the one Sun hyped hardest) a 32-thread T1 or 64-thread T2 did quite a bit better than its Intel contemporaries. The problem is that a lot of workloads expect to be able to consistently issue more than 200 million instructions per second per thread to do well, even when you are hiding latency by doing fast thread switches. Databases, a workload you cite as parallelizing well, tended to run like crap on Niagara.
With Bulldozer, you effectively have 16 2-issue cores, since each module has a shared 4-instruction decoder. I'm skeptical that it will perform all that much better on multithreaded integer workloads than the T3 did, which had 16 1-issue cores with aggressive multithreading and latency hiding on top. On the other hand, Westmere-EX is 10-core, 4-issue, 2-way SMT-capable per core, and has big caches with good latency numbers. On both a technical level and based on early benchmarks (SPECcpu), things don't look good for Interlagos.
SPECcpu results, TPC-H, and personal experience.
Servers need single-thread too; think stuff like big database writes, joins, ERP, and CRM. Think outside the embarrassingly-parallel web-serving box.
If multithreaded performance was all that matters, the Sun Niagara chips would have done a lot better than they did.
They both have the same issues, including that each module (two 4-issue cores) has a single 4-instruction decoder in front of it. Cache latency is also likely to be similar if not the same.
Intel's server chips are 8- and 10-core, and outperform Opterons by a considerable margin.
It doesn't matter. If the shares were promised, they were promised - if there was no intention to deliver, they shouldn't have been promised in the first place. Honor is honor, whether or not you think the other person "deserves" it.
Coming from him, that's pretty damn funny.
The VIIIfx and IXfx are uninteresting for commercial applications due to the irritating fact that they don't support SMP.
The IBM Power 795 usually outperforms the M9000, and you're comparing a 64-socket machine to a 32-socket one.
Your evaluation of the T4 is actually much worse than the reality - it's a significant improvement over the T3. But the lack of speccpu or TPC-C benchmarks is interesting.
I think I have. I've seen that the "latest and greatest" SPARC64 VII+ still gets regularly spanked by Power7 and Itanium and even commodity systems in performance, despite being considerably more expensive - and I've seen vague roadmaps for the future of M-Series. I've seen that the T1/T2/T3 performance promises never really panned out (see: SPEC results vs the much cheaper Magny-Cours), and that the T4 has so far largely been hidden behind the veil of vague benchmark-fu while being far more expensive than its competitors.
What hardware have I been missing?
AIX at least runs on decent hardware. SPARC sucked for years before the acquisition, and continuing to beat that particular dead horse seems unwise.
Palm never really did much with their BeOS code, and HP doesn't even own it (PalmSource, makers of the Access Linux Platform, do.) webOS was killed by a fickle market and the incompetence of HP and Palm, not by any kind of curse in an unrelated product.
Strawman arguments for 100, Alex.
Microsoft has stated that they won't sue over C# usage, and they've demonstrated no inclination whatsoever toward doing so. Nobody has ever presented evidence that they will - only vague accusations about how EVILLLL Microsoft is.
Mainframe market share is huge (about 3 billion a year in hardware sales) and growing rather quickly, especially since "cloud" and "virtualization" became buzzwords. It's the only part of the high-end non-x86 niche that's really having solid growth right now - SPARC and Itanium have been tumbling for a while, and Power has been more or less flat.
People tolerate it because it works. Mainframes have compelling performance characteristics, especially for virtualization- or I/O-heavy workloads, and most people don't need a full unlocked processor (a CP.) Linux is the fastest-growing OS on z, and a Linux specialty engine (Integrated Facility for Linux) is relatively cheap compared to the other types of specialty engines. zAAP/zIIP/CP only really matters if you're running z/OS, in which case you're probably a large enough company tat you can afford it, or if you're running z/VSE, in which case you're probably only using one or two processors anyway.
The summary misses something fairly important, which is that Windows isn't running on the z mainframe itself. This allows Windows blades to be inserted into an external chassis (zBX) and managed by a software component called the Unified Resource Manager.
I'd say they do have the money. They're continuing production of the Tu-160, which is easily the best strategic bomber in the world, and if I recall they're ordering several hundred PAK FA as well as modernizing their Flanker fleet to the Su-35 standard, which should make it one of the best 4+-generation fighters out there.
Russia is building the PAK FA 5th-gen fighter.
The F-22 had a few issues, namely that Congress wouldn't permit it to be exported and it wasn't really suitable for carrier-based ops (no STOL/VTOL capability.) The F-35 was supposed to be the successor to the F-16 as the near-standard fighter for NATO and non-NATO US allies, and provide a replacement for the F/A-18 for the Navy.
The T3 is crap for anything compute-intensive. The whole point was "throughput computing," not processing power, and even at that it's pretty poor. Look at the speccpu benchmarks sometime.
The VIIIfx supposedly doesn't support SMP, so I'm skeptical as to how interested in it Oracle is.
Or HP-UX, or AIX, or GCOS 7, or z/OS, or OS 2200, or NSK...
Or a properly configured Windows or Linux. Proper administration matters far more than OS choice.
My Vaio Z doesn't boot from USB. That's incredibly weird, since it's a "premium" high-spec machine, but it makes using Linux a pain in the ass. (So does their disregard of TRIM in favor of a custom SSD garbage-collection system, and their proprietary switchable graphics, and their out-of-the-box RAID 0'd SSD's, and...) It's like Sony had a serious case of NIH syndrome.