“The introduction of Intel’s new Hyper-Threading technology is providing Microsoft with a handy mechanism for getting more money from Windows 2000 Server customers, for encouraging users to switch over to Windows .NET Server, or both. It kind of depends on how .NET Server pricing pans out, but as far as Win2k Server goes it’s already clear that people wanting to use the extra oomph in the new Xeons are going to have to stump up. According to a Microsoft backgrounder available here, Win2k Server handles Hyper-Threading by using the processor count from the machine’s bios.” Read the rest of the report at TheRegister. Our Take: Problem is, a SMT Xeon is not as fast as a real SMP Xeon 2x system at the same Mhz. Therefore, paying double the OS License price for something that only provides a hardware-level feature (Hyper-threading), but not the actual speed you would expect from its SMP counterpart, it is IMNSHO, unjustifiable.
SMT Xeons Count Double for Win2k Server Licences
About The Author
Eugenia Loli
Ex-programmer, ex-editor in chief at OSNews.com, now a visual artist/filmmaker.
Follow me on Twitter @EugeniaLoli
the only group benifiting from this is intel and microsoft. The only way somebody is gonna benfit is if OSes like linux support it.
This was a hack shortcut Intel used to get SMP-type software up and running without requiring some massive rewrite of compilers and other such things. It acts “like” a second processor, so they can treat it “like” a second processor. Now WindowsXP will have to be rewritten very slightly to be able to tell the difference.
I’m no Microsoft fan, and not a 100% Intel fan, but this is hardly a grand money making scheme by the two. It looks like a simple case of good engineering conflicting with marketing and sales.
If Intel can squeeze more oomph out of the hardware by making the OS believe that there are two processors, then why does one CPU not perform better? It sounds like a software problem to me!
Intel comes up with a new technology, and you expect a 2 year old product to be rewritten to take advantage of it, so that the licensing works out?
And especially when the new version of this product (.Net Servers) will take this new technology into account, and not require additional CPU licenses, and is about 3-5 months from being shipped?
Not support the new technology in the old software, but support it without any licensing issue in the new software. Duh, me no genius, but it makes sense to me.
uh, why would anyone use windows as a server?
to keep you telling such bs.
Hyperthreading is the only recourse left to Intel & all cpu designers to keep all funtion units of the cpu busy while the suspended thread waits for memory etc to catch up. Since all single threaded cpus stall when cache accesses fail & since DRAM accesses are 100x slower or more than current cpus, there are few alternatives. If cache misses m reach say 3%, then a cpu will be idle 300 out of 397 cylces, ie 4x slower. This will get much worse with 10GHz cpus only a few yrs away, before getting better. I have seen my 800MHz Athlon reduced to 200MHz P6 levels when running highly optimized cache busting asm code on large data sets. Clearly to reach max speed, m must reach 100.
This is extremely difficult, it can be achieved by, 1) bigger much more expensive caches ie Xeon, Alpha, Sparc, Itanium, or 2) add Level4 cache main memory say 16-64Mbyte SRAM very expensive, or 3) switch DRAM to RLDRAM ie faster RAS access time also expensive , or 4) smaller code size ie BeOS v Bloatware & or very careful coding of critical loops, or 5) Hyperthreading the cpu core.
Hyperthreading is atleast 20yrs old in HW computer architecture circles, but the SW people have resisted multiple cpus for so long it has finally come to crunch time. Hyperthreading is a darn good way to back out of the extremely complex designs now prevalent, since branch prediction & predication can now be replaced by fast context switches. Double whammy here, HW now can reach 100% busy, & much simpler designs can be run even faster.
This is not so very different from what Inmos did 20yrs ago in the Transputer, putting a thread switcher in HW but in that case, the threads (or Processes) were communicating through messages. It forced the programmer to get with the Occam parallel model & was probably too early.
Hyperthreading in HW is really no different than mainframe OS multitasking & sharing use of slow peripherals such as disks, printers etc. Single threaded OSs ie MacOS were thrown out yrs ago, so will Single threaded HW.
Intel’s version though is barebones, 2 independant threads. Intel should allow any no of ready Processes to be queued, not just 2 or 4. But that would be stepping on OS guys turf. If the OS can use the HW kernal, || programming could get a real boost.
As for MS, they will have to adjust their cpu count policy or face the consequences of more defections to other OS.
MS must really have a stranglehold on the market to get away with charging more for Windows because it is going to be used on a multiprocessor box. What’s next? Are they going to start charging more if you use processors with a faster clock speed? Is there going to be a surcharge if you add more RAM. Perhaps Windows should check how much disk space you have on the system, and require you to purchase a license upgrade if you have too many gigs for your edition.