Talk:UltraSPARC T1
![]() | This article is rated C-class on Wikipedia's content assessment scale. It is of interest to the following WikiProjects: | ||||||||||
|
Article title — Moved
[edit]I believe the title of the article should be "Niagra (processor)", and we should probably move it sooner, rather than later, as the page gets linked more and more. – Fudoreaper 20:56:56, 2005-09-05 (UTC)
- I disagree. I say Sun gets to name their own product. --Mike Van Emmerik 07:00, 15 December 2005 (UTC)
- I like the Niagara name better too. Most people that know of this product talk about "Niagara", not the "T1"...not only does it sound cooler and less like a prequel to the T100 and T1000 from the cinema, but wiki policy advocates using the common name. Justforasecond 19:27, 7 April 2006 (UTC)
- Just for the record, the original title of this page was "Niagra Processor", which i thought was lazy, and should have (parentheses) around the word processor, and lose the caps on 'Processor'. However, before anyone cared very much about it, the true name came out, and the article was thus moved to "UltraSPARC T1" on 2005-11-14. This is the proper name for the chip, which everyone now calls it, though its code name is still used in some circles. Sorry Justforasecond, but the issue, as i see it, is dead. —Fudoreaper 05:17, 1 November 2006 (UTC)
Integration
[edit]Items which need integration:
- Niagara2 will support 64 threads
- Niagara comes with a hypervisor which allows to run independet OS images per core (e.g. Linux, Solaris and so on)
- http://www.theregister.co.uk/2005/09/09/niagara_many_cores/ says Sun will ship Niagara1 with 8, 6 and 4 cores enabled (yeah, we can be happy that these chips really run beyond the 1GHz barrier when I think about who (=which fab) makes the chip)
- Niagara is SMP-on-a-chip
- There can only be one Niagara chip in a system, multichip SMP is done by it's cousin "Rock"
- Niagara was developed by the Alfara people
- More info comes with the "Niaraga - an Overview" artcile on TheRegister.co.uk on monday! :-)
Correction: Sun had a prior multicore multithread processor: MAJC --JustAnotherJoe 05:46, 10 December 2005 (UTC)
Performance
[edit]MIPS = ?
GFLOPs = ?
Criticism
[edit]FPU performace is less then 110Mflops and each core can only access a fraction (1/8) of this, rendering Niagara1 completely useles for applications like desktop, xterminal server, sunray server, render farms and so on. Niagara1 mainly targets applications such as web-, database-, mail- or filesystem servers - anything beyond that needs extensive testing whether this chip is suited for such applications. —Preceding unsigned comment added by 84.174.122.102 (talk • contribs)
- True. I'm not sure if that could really be called a criticism though. How can you criticise Sun for making the product they promised? And one that appears to be able to handle the tasks it was designed for. It looks like Sun picked their niche and successfully delivered a product for it. If you want a render farm, then go get a bunch of Opterons or Xeons. This T1 looks to be well suited to the types of servers you listed. Imroy 03:40, 24 February 2006 (UTC)
- With the exception of render farms, most of the applications you list are not particularly floating point intensive, and should have acceptable performance on Ultra T1. Although a sunray server could run anything, so an Ultra T1 may or may not be suitable, depending on what applications users are running on it.--HorsCat 04:43, 1 November 2006 (UTC)
is this an ad ???
[edit]"At the time of its release in December 2005, a single-chip, eight-core, 32-thread, 1.2 GHz UltraSPARC T1 server performed similarly to a two-socket, four-core, eight-thread, 1.9 GHz IBM POWER5 server, performed similarly to a four-socket, eight-core, sixteen-thread 3.0 GHz Intel Xeon "Paxville MP" server, and exceeded the performance of a four-socket, four-core, four-thread 1.6 GHz Intel Itanium server. Arguably, this made the UltraSPARC T1 the world's most powerful general-purpose commercial server processors, when considering multithreaded commercial workloads." This is pure theory. Everybody that really use this CPU knows that in real world this is completly wrong. Mainly because there is only one FPU for all 8 cores.Mr.damien (talk) 12:01, 10 September 2010 (UTC)
- No, the article is not an advertisement. But, the statement in question is very problematic. Firstly, it does not state what benchmark(s) were used to measure the performance of the various CPUs compared. And secondly, even if it did, the article cannot assert, or imply, that it is the highest performing CPU for whatever since the benchmark(s) used only prove what was measured. I think that the article can, however, include statements attributed to reliable sources that have, from comphrehensive benchmarking and/or analysis of the architecture, made educated statements about performance on various applications. Additionally, (and I seem to encounter this claim everywhere) the lack of good floating point performance does not inhibit the UltraSPARC T1 from performing well on certain applications. Really. I do not think that there is any CPU that excel on every application, and this should be obvious to everyone. Rilak (talk) 04:13, 15 September 2010 (UTC)
cache npov
[edit]The cache stuff all seems either NPOV or speculative.
- "A large cache is not required because the UltraSPARC T1 addresses memory latency by multithreading."
A large cache is never required. Even single-threaded procs can have zero cache as long as you don't mind losing performance. Anyway, this statement doesn't have the needed caveats. maybe something like "under certain workloads, the T1s multithreading obviates the need for large amounts of cache memory.
- Cache misses are masked by switching to another thread, so cache misses are not as significant.
Thid id only true if there is another thread available, and if the thread that misses is not time sensitive.
- For a single threaded processor, cache misses are much more significant, so large caches are necessary to reduce cache misses.
Cache misses on the T1 are just as significant on a single threaded processor. Your single threaded program will suffer from cache misses just as much on a Niagara as on a single-threaded processor, and with smaller cache shared between more threads this will happen more often.
T1 sounds like a daring approach, but let's keep the article away from all this stuff about how it is superior to single threaded processors and how it is designed towards certain workloads and looks to be very good at those loads.
btw:
• On the SPECjbb2005 test of Java server software, the T2000 scored 63,378 business operations per second compared with 61,789 for an IBM p5-550 with two dual-core Power5 chips and 24,208 for a Dell PowerEdge SC1425 with dual single-core Xeon processors.
• On the SPECweb2005 test of Web server performance, the T2000 scored 14,001, compared with 7,881 for an IBM p5-550 with two dual-core Power5 processors, 4,850 for a Dell PowerEdge 2850 with two dual-core Xeon processors, and 4,348 for an IBM x345 with dual single-core Xeon processors.
• On the NotesBench test of Lotus Notes performance, a T2000 accommodated 19,000 users at $4.35 per user and got a NotesMark score of 16,061. In comparison, an eight-processor IBM p5-570 had 17,400 users, a cost of $10.19 per user, and a NotesMark score of 14,740. But the average response time of the IBM system was 270 microseconds compared with the slower 400 microseconds for the T2000, demonstrating the relatively slow single-thread performance of the Sun system.
Justforasecond 19:21, 7 April 2006 (UTC)
- I think you're largely splitting hairs over the exact wording. I'm too tired at the moment to work on this, but you're always welcome to have a go. I'll see what's happened after I get some sleep... Imroy 19:45, 7 April 2006 (UTC)
- Ok, I've just worked on the cores/cache section. I changed the order of the paragraphs because I thought one of the later ones made a better intro. And I've almost completely rewritten the cache description. I hope it explains things better than the previous version, and that I've also removed the words/phrases that Justforasecond thought were POV. I'm not the best writer though, so any improvements on the wording would be appreciated. Imroy 08:47, 14 April 2006 (UTC)
UltraSPARC T# Series Merge
[edit]Due that their is some overlap between the UltraSPARC T1 and UltraSPARC T2 along with a good probability the Niagra 3 will be similar; Should these pages be merged to allow for consolidation of similar information and comparison/development in the series? WikipedianYknOK 19:05, 7 May 2007 (UTC)
- That sounds like a good idea. Sun also released the T2 design under the GPL, just like the T1. So it does make sense to merge all or most of the two articles. But "UltraSPARC T series" sounds a little awkward. Sun uses the term "Coolthreads", but perhaps that's too market-speak. How about "Sun Niagara"? --Imroy 05:43, 11 August 2007 (UTC)
- The only collective terms of record seem to be "CoolThreads" and "Niagara" although you could also make a case for "UltraSPARC CMT".
- "UltraSPARC T series" (note no capitalization of "series") seems the most logical but I don't see any record of the term ever being used. Niagara's a code name and CoolThreads does sound market-y, but then again, all brand names are market-y.--NapoliRoma 06:44, 11 August 2007 (UTC)
- Possibly "CoolThreads processors" or "Niagara processors" for consistency, since their is already a related article titled Rock processor . "UltraSPARC T series" is not an official term, that was just a quick section title I had put for this discussion topic. WikipedianYknOK 02:08, 12 August 2007 (UTC)
Intel 80386 and Intel 80486 are similar, yet each has its own wiki page. -- Rayson Ho 02:40, 17 August 2007 (UTC)
- Indeed. But we're not talking about merging everything into one article, at least I didn't think we were. Both processors share a lot in how they work and some other issues e.g releasing the designs on OpenSPARC. So it makes sense to create a common article for the family, like x86-64, and leave the specifics to the individual articles. Although, if enough material cannot be found for the separate articles, it might make sense to just merge everything into one article, which is what happened with Itanium. For now we're discussing just what to call the common article. I prefer something based on the project name "Niagara" instead of the marketing term "CoolThreads". --Imroy 07:51, 17 August 2007 (UTC)
Purpose
[edit]Article says:
- The UltraSPARC T1 was designed from scratch as a multi-threaded, special-purpose processor,...
(though it doesn't say which purpose...) – but also:
- ..., this made the UltraSPARC T1 the world's most powerful general-purpose commercial server processors,...
Now I find it hard to believe, that an engineering team designs a Processor for specific tasks, but somehow magically ends up with the best overall processor. It's difficult enough for the general public to understand where different processor architectures have their pros and cons as it is, so I would find it a good idea for someone in the know to clear up this apparent paradox. --BjKa (talk) 10:55, 27 May 2009 (UTC)
- The T1 was designed for specific commercial server workloads, thus it is "sort of" a special-purpose microprocessor, so the first statement you claim is problematic is not really that much of a problem, although the article could make this more clear. Their second statement you claim is problematic is also not that problematic because it says "general-purpose commercial server processor", which I interpret as a microprocessor without special fixed-function hardware for accelerating particular workloads. Again, the article could be made more clear, but it is not particularly concerning in my opinion. Rilak (talk) 05:19, 28 May 2009 (UTC)
Add historical reference, Afara Websystems ?
[edit]The technology that ended up producing the sun4v / Niagara systems has come to Sun through the acquisition of Afara, http://wiki.riteme.site/wiki/Afara_Websystems ; better historical references (pre-2005) would improve the article. — Preceding unsigned comment added by 205.228.82.170 (talk) 13:22, 28 May 2012 (UTC)