According to a benchmark, Sun’s Niagara processor is over 4 times faster at serving dynamic PHP pages than a dual Xeon server. “We did real production benchmarks using different servers. Servers were put into production behind load-balancers, then weights on load-balancers were changed so we got highest number of dynamic PHP requests per second. It must sustain that number of requests for some time and no drops or request queue were allowed. With static requests numbers for Opteron and T2000 were even better but we are mostly interested in dynamic pages.”
Well sure looks fabulous. Just that I would’ve appreciated if they stated how much the equipment of each cost.
Now surely Sun has lowered pricing quite a bit lately, but whatabout this piece of hardware vs the IBM HW?
If they are equally priced, then it’s really spectacular.
Now I don’t know how PHP serve it’s coffee, and this is just a benchmark. Would it be different assuming the complexity of the PHP scripts?
Now surely Sun has lowered pricing quite a bit lately, but whatabout this piece of hardware vs the IBM HW? <br/><br/> Also, this piece of hardware vs. quad-opteron (dual core) to level playfield on CPU count on both sides.
Unless you design processors, it’s not clear that normalizing the core count teaches you anything.
Here’s a hint. Anyone serving PHP as an application probably did so because they don’t have the $$ for something else that’s more easily scaleable.
As such, I’d doubt they’ll switch to Sun’s hardware, even if it is better. Xeons are cheaper.
Bottom line: I don’t know anyone running PHP on more than low price, (cheapest things you can find) commodity servers.
That is pretty stupid way of thinking.
PHP has nothing to do with scalability and small companies that does not afford servers. Think of all ISP:s that serves PHP on their servers for their customers, do you really think they invest in cheap hardware where you maybe don´t even know what guarantees you get.
• How many cheap “servers” can you find with dual power supplys?
• How many has ALOM?
• How many has an personallity card with the hardware id on it so that if the server breaks, you can take that card and insert into another backup server and continue using applications that has licenses bound to the hardware id?
Please, don´t come and say that companies invest in shitty cheap servers if they are using them for system critical applications. Even if that is “only” php.
Edited 2006-02-20 15:11
+1
Not to mention the IBM x335 is a dead product… not even sold anymore
Yahoo! uses PHP and it seems to scale quite well for them! Also, do the full math an Sun’s hardware is very competitively priced, if not cheaper than most already.
At least what i think.
here is an link to their prices:
http://store.sun.com/CMTemplate/CEServlet?process=SunStore&cmdViewP…
Examplpe:
—-
1 @ 1GHz, 4 core CPU
3 MB L2 cache
8 Gb ram, max 32 Gb
2 x 73 Gb SAS (Serial attached SCSI) disk drives
1 DVD-rom/CD-rw
4 x 10/100/1000 Mbit ethernet ports
1 x serial
3 x pci-e
2 x PCI-x
2 x Power supplys
ALOM
Sun Java enterprise system
Solaris 10
That is an good price.
We have had an T2000 server at ours (you can sign up for an 60 day testing period) and it sure is an quick server.
Well, this ones (MEDIUM) price seems even better in my opinion
http://store.sun.com/CMTemplate/CEServlet?process=SunStore&cmdViewP…
except the fact it has 1 PS only and less slots and no SCSI drives. $4,995.00 or $5,745.00 seems like a joke low here.
Just interested if anyone knows, no trolling.
How is about
– T1000 and Fibre Channel option?
– T1000 and Linux? I could even live if I knew it will be possible in one year and then port my app to Linux from Solaris.
I’m currently searching for some HW I need for my tests and currently deciding between 2xIBMp185 or 2xSunT1000.
Yes, Linux is a must, but in the end, my SW will have to run on both, Solaris and Linux (where Linux is a bigger must than Solaris (or at least first implementation scheduled will be linux based)). And since I think Solaris doesn’t work on those IBMs (but if it does I would appreciate correction), SunT1000 seems better choice here, at least for now.
you can sign up for an 60 day testing period
Is this option regional only, or is this Sun international policy by default? Or even better, could you provide some information on it?
Why on earth would you want to run Linux on an UltraSPARC T1?
Why on earth would you want to run Linux on an UltraSPARC T1?
??? Now, this is a funny surprise. I really hope you’re not the heart of Solaris community. I really don’t know why would I deserve being trolled for simple honest question.
As I said. I’m writing CS HPC service. Two of three already ordered installations will run on Linux (Already decided IBM-PPC). One on Solaris (Already decided SPARC).
In the final phase I will have to test complete thing on at least near implementation standard to see if everything is correct. Since I will need to buy at least two machines (but it might scale to four of them) for my test environment you can guess it won’t be cheap (and those aren’t even thrird of my initial test environment costs).
Now I hope you see my problem.
:: If I buy p185 setup I end up without any option to test Solaris. (or at least I think it is so).
— Positive thing: I get linux on the almost exact environment, and I could probably make Solaris env. on few Opteron boxes, but as I said I don’t really know much about diff between Opteron and SPARC, so I would like to mimick original environment as much as possible to avoid surprises.
— Bad thing: I don’t know how everything will work on SPARC at the end
:: If I buy T1000 setup and Linux doesn’t work (or at least it won’t start working in one year) I can only say goodbye to another >$12000 which would really badly broke my plans. But let’s say Linux would start to work
— Positive thing: I got similiar hardware for both environments and I can do my final PPC testings on my G5 (or simply add one more, hell they are cheap). Meaning, I got both environments. Better HW, almost the same price. Much better
— Bad thing: I don’t see one
Since I don’t have much experience with Sun hardware (for now I’ve been trying Solaris on Opterons, but as I said Opterons are not in question here), well… I posed this honest question. But, seeing the result of my question, I think going with the p185 (and buy few old SPARC models afterwards) where at least I know the HW setups and community does not make stupid jokes of honest questions would be better.
Sinecerelly, thanks for your time, you really gave me the answer I needed.
Consider that David Miller just got the kernel to boot on a Niagara machine and the blog entry was linked here two days ago I can see why some people woud not take you seriously:
http://www.osnews.com/story.php?news_id=13721
It will probably be at least a year before Linux actually runs on a Niagara machine (a full distro).
First to thank you for the second answer you provided.
Consider that David Miller just got the kernel to boot on a Niagara machine and the blog entry was linked here two days ago I can see why some people woud not take you seriously
Why, because I didn’t succed to read all the news about Solaris (and to be honest, I always skip articles Linux booted on XXX (where I even skip what XXX is), I can only expect it on my home broom the next day)? I do have days without internet too sometimes and I don’t have time to subscribe for all news about every OS I need in the line of my work. And even in the case I would notice that, it is only booting. I asked about yearly timeframe.
But,… damn, if it is so, then I better move off. I don’t hide the fact that Solaris became really interesting for me when Sun announced GPL version (CDDL is not interesting to me, I always decide for commercial apporach over free), before that it I was satisfied with commercial approach to Solaris only where Linux was no go.
But to be truthfull. Somehow I find my self more or less depressed whenever or wherever I ask Solaris users about some things. Sorry to say this like that, but mostly I get the feeling that Solaris community is a “piss-poor excuse for group of people zealoting about the same thing”.
Actualy you’re the only “human” person I encoutered when posting any question about Solaris. As soon as I mention I’m not Solaris die-hard and I what really need is crossing the borders between platforms, trolling and modding down starts. I don’t mind being modded down, but it is far different from BSD or Linux communities I’m used to with (all my knowledge there is mostly community based, in Solaris case, well I only learned how to become dissapointed). It comes with a lot of grain and salt with it.
Or to translate it, as much as Sun gained me with their latest actions, communities are turning me back away.
It will probably be at least a year before Linux actually runs on a Niagara machine (a full distro).
This is kind a “just in time” or “just too late” and I don’t know if I’m prepared to take the risk. Could you maybe recommend something else for my case? I would appreciate it very much. Or at least some similiar but cheaper SPARC HW with Fiber option (can be older model, since in this case it will run Solaris only), where I could fit p185 and those in my plan (or better say wallet). As I said, when SPARC arch is in question I’m completely new there without a single minute spended on it.
Edited 2006-02-21 21:08
It is hard to recommend something that you haven’t used yet (I expect to see the T2000 I asked for on Thursday) and have little to compare it with. The benchmarks are interesting, but that is only the beginning of many tests we could come up with where I work.
Almost any machine that has a PCI (32/64-bit) slot can support Fibre Channel, it is a case of how much money are you willing to spend to get a level of performance and whether the desired hardware is supported by the OS of choice.
Depending on what you are looking for in terms of performance for a specified workload, there might be no equivalent to a T1000/T2000.
Almost any machine that has a PCI (32/64-bit) slot can support Fibre Channel, it is a case of how much money are you willing to spend to get a level of performance and whether the desired hardware is supported by the OS of choice.
Well, I knew that about PC and PPC (those two I know). But with Sun I was a little confused (All my excursions with Solaris were constricted to Opterons so far). Sometimes they specify optional Fibre, sometimes not. I wasn’t sure, so I asked. I’m not hiding the fact that I don’t know much about SPARC.
Depending on what you are looking for in terms of performance for a specified workload, there might be no equivalent to a T1000/T2000.
I noticed that too. This is why I was noting them as better option than p185.
The only thing that bugs me there is the timeline you specified for linux to start fully working and I can’t afford that. To be truthfull, I didn’t even notice that they use Niagara chips (this is why that article pased by me, I went looking after you noted this fact and suddenly picture became much clearer, again thanks for this pointer). As I said, I can be considered a moron when SPARC is considered. But I suspect it has some specific quirks just as PPC has in case of low level optimization, but what bugs me even more is the fact that I wouldn’t like to be surprised when actual implementation starts (I can optimize later when I get more familiar with SPARC, but unfortunatelly I have to provide working version before that).
You can safely correct me when I speculate this. I said I’m more or less guessing about SPARC. Optimizing or at least coding on older SPARC that could be obtained for $3000 would mean practicaly the same as optimizing/coding for Niagara (no, I don’t do low level kernel work) except the fact that it has more cores which means more threading and segmentation (no problem here, all my threading, etc. is safely made to be runtime modifiable and externaly controled during processing. This is the only way I can tune up to the max after workload presents it self and shows its real picture which varies from one need to another). And I think I could probably scrap up to $6000 after buying 2-3 p185 for two older SPARCs with Fibre (hopefully 2GBit but I could live with slower) and SCSI drives are not needed (and the fact that I have an option to get to used SPARCs, that is, if I would know what I should be looking for). This is what I had in mind when I asked if you can suggest a SPARC model. So, I would appreciate at least a pointer in which way/models should I look in this price range, or better what would you look at if you would be constricted to $6000 and similiar needs as I am (oh, yes. In this case Linux poses no question, I don’t care. They will run Solaris only, Linux will be on p185).
T1000/T2000 are unfortunately too expensive for me, since they can’t boot linux too, or at least time frame poses too high risk for me. But,… I won’t hesitate when customer will pay the bills for those or something better:)
Again, thanks for your answer and your time.
As I said. I’m writing CS HPC service.
This implies (to me) that you really care about the performance of a small number of threads. If you have a single-threaded (and single-process) workload, your workload is going to perform very much like you were running on a 2 year old single processor V240. In other words, the T1000 or T2000 would be dog slow. If you are running something that deals with dozens of running threads (32 is optimal) the T2000 will do quite well.
If you are after good performance for a small number of threads, look toward Opteron, Xeon, Power, or UltraSPARC IV+ systems. The first two on that list are likely to be the cost effective options.
FWIW, yes I have used a T2000 (8 cores @ 1 GHz, 16 GB RAM, < $20k list) and it did at least as well as a V490 (UltraSPARC IV @ 1.35 GHz, 32 GB RAM, > $80k list) on a heavily threaded workload (Oracle Apps). The really cool thing is that Oracle charges for 2 CPU’s on the T2000 or 6 CPU’s on the V490. With Oracle licenses way more than $10k per CPU, the T2000 wins when it comes to licensing as well.
just if someone is surprised to see that a 8-core CPU beats 2x/4x configurations….
No, T2000 is not only 8 cores per cpu.
You can get it in “smaller” versions also.
And, even if it is 8 cores per cpu, why not use it for benchmarking, it is still an one cpu server, but with several cores.
That is like saying that you are not allowed to benchmark an HyperThreading enabled intel CPU with an single AMD cpu!
Making use of Intel’s SMT implementation is not like having two processor cores. Comparing a dual-core Opteron to a single-core Pentium 4 with HTT might have some informative value but it is not a “fair” comparison. One approach attempts to schedule more work on the same number of functional units, while the other schedules work on twice the number of functional units.
That doesn’t mean that a comparison betweena dual Xeon and the T2000 is necessarily unfair. Each core in the Xeon is considerably more complex than one core in the Niagara. The question you should ask yourself, is if instead the benchmark were stressing the strength of the Xeon and weaknesses of the T2000, if people would come out and say “YEAH TWO CORES SMOKED EIGHT OF SUN’S!” Sometimes people treat computers like sporting teams.
Okay then. Half the score for the Niagra setup and it’s still more than double the Xeon!
Okey, I caught up on the pricing.
The Sun System = $8,295.00
The IBM System (336 series) = $3,319.00*
(Where the IBM system now is 3,2ghz).
But, thinking in terms of 4 times as fast, and less than 4 times the price, it sure is ok.
However, it’s like comparing apples with oranges. More interesting would be to check a higher end box from IBM which comes at about $8,000.
However, looking at the X2100 which also beats the IBM box hands down is priced $1,895.00.
http://store.sun.com/CMTemplate/CEServlet?process=SunStore&cmdViewP…
(Extra large model)
This would mean IBM is (like I said before) far to expensive to work with, and Sun is a true bargain.
Okey, I caught up on the pricing.
The Sun System = $8,295.00
The IBM System (336 series) = $3,319.00*
(Where the IBM system now is 3,2ghz).
But, thinking in terms of 4 times as fast, and less than 4 times the price, it sure is ok.
Not really, to equal the performance of the T2000 you would need 4 IBM x336 so 4x $3319 = $13,276
Also the one x336 consumes more power than the T2000. So 4 IBMs would consume 4 times more than one T2000. More power consumption more cooling and electricity costs for the same performance.
Anyway you slice it the T2000 is the cheaper solution to get the performance mentioned in the benchmark.
If the price point is about the same, then this comparison is relevant, but the blog article doesn’t make a point of that.
That said, comparing eight cores vs two processors then hailing it’s a massive victory seems pretty lame. What if this was eight cores vs eight Xeons? What would the performance comparisons look like then? If price isn’t an issue, it looks like a xeon is as fast as a Sun core, so eight Xeons should be nearly as powerful as Sun’s eight core.
I wouldn’t point this out but the blog entry says nothing except comparing the performance of disimilar machines, with a wide technology gap no less.
The right way would have been to compare
-the dual Xeon machine against a dual core Sun processor to see the advantages of both cores on die,
-then comparing the eight core processor to the dual core one,
-then comparing the dual processor Xeon against a dual, two core Sun box
-and finally comparing to an eight, eight core Sun box to give a complete perspective.
Yes, that is correct, but anyhow, it is also interesting to see how well an 1 cpu with 8 cores does in an test, because, up until now all cpu:s has (almost, if not counting IBM:s highend cpus) only had 2 cores.
But, anyway, in the test we can clearly see that the ibm server is slower than the sun boxes with AMD.
1. IBM x335, 2x Xeon 2,8GHz
2. Sun x2100, 1x Opteron 175 2,2GHz
3. Sun x4100, 2x Opteron 280 2,4GHz
So, SUN is cheaper and faster, even if you don´t choose their T2000 (T1 cpu) cervers to do the work.
Not to be too picky, beating a P4 these days isn’t too hard.
But the price point is what really shows the difference, the only thing is the blog article says nothing about price, only comparing various Sun servers to a single model of IBM.
I’m impressed by Sun, and I hope they make their way back, their products are looking good.
I do look forward, however, to see how Sun compares to AMD’s multicore offerings. I’m not so sure Sun’s eight core would best AMD’s eight core. Then again, this is now and AMD doesn’t have an eight core yet.
I’m not so sure Sun’s eight core would best AMD’s eight core.
It wouldn’t. But that would not be a fair comparison, because AMD’s cores are bigger, more complex, and thus more expensive (although economies of scale might distort that.) The relevant comparison would be with a 2-core or possibly a 4-core Opteron.
Actually you are completly wrong. If you do have a lot of web servers one of the main factors is price/performance. Serving WWW is horizontally scalable so in most cases you buy servers with best price/performance. And comparing just server prices these Niagara boxes are looking quite good. Now even if a purchase price is similar you’ve got to manage 4 times more x86 servers, pay lot more for cooling, power and loose much more server room space (see that UltraSparc T1 with 8 cores consumes 72W at most). When you add all these things than Niagara boxes look much better in price/performance.
Now if you compare larger x86 boxes then Niagara price/performance is even better.
I agree that the price point is attractive, but as I said, it wasn’t brought up as a point. I don’t know why you said I was completely wrong?
You wrote “That said, comparing eight cores vs two processors then hailing it’s a massive victory seems pretty lame.” – I can’t agree. If you can get 1U or 2U (T1000/T2000) server with 1x CPU (8 cores) and price is similar to 2-way x86 server box (adjusting for perfrormance) then such comparison is relevan. Add maintaince costs (4x less servers), power and cooling costs and you’ll see why this benchmark is relevant.
If you take 8-way or 4-way x86 system it’s going to cost you much more. Not to menchion rack space, power/cooling costs.
Good information, you should add this to your blog entry instead of housing this in a comment section of an article that’s not directly linked from your page.
And don’t assume the audience can read your mind. If you want us to see your point, you have to convey it first.
The benchmark seem mainly if not only oriented to test multithreading capabilities, no wonder that more cores brings the best results!
But anyone interested in machines of that level knows that 10 men cannot dig an hole 10 times faster than a single man and that many problem cannot be parallelized well (and some, not at all).
See i.e. the comparison between Opteron and Xeon, this is an old duel and lot of benchmark were published, but this is really biased toward the areas where the Opteron excells and the Xeon fall shorter due to a well known and documented worse capability of Xeon dealing with concurrent tasks
In those years, no benchmark trying to be realistic and complete (if a benchmark can be…) had claimed a 2×2,4 GHz Opteron to be 2,5x times faster than 2×2,8 GHz Xeon as in this benchmark!
Ok, now forget the Xeon and see the Opteron (being it a Sun machine, we can assume that the test was quite friendly to the Opteron architecture too…), it’s interesting saying that in this test the dual Opteron 280 (two cores at 2,4 GHz) scores approx 250 points against the approx 420 of T2000 (8 cores at 1 Ghz)… I wonder how a 8 way server based on Opteron 8xx would score since it scales so well (see the single Opteron for comparison)!
Odds are for awesome performance for AMD based solutions, staying away from possible vendor’s lock in, enjoing the huge software base for the huge OS families ported to x86 systems, maybe having also a better cost/performance ratio…
With AMD providing also some good low voltage families of CPU and mature 64 bit implementation, it seem that AMD has really good offers in 2x 8x – computing!
If you read blog entry more carefully you would see that x4100 was equipped with 2 dual core CPUs (280) – so there’re 4 cores at 2,4GHz each. That’s why we got 2.5x performance of 2x Xeon 2.8GHz (signle core).
This benchmark was all about WWW – nothing more.
And most important factor is price/performance.
Please note that T1000 is an 1U rack server, and T2000 is a 2U rack server. Now add to it power and cooling costs, multiply it by lot of servers and see what price/performance difference these servers could make.
Please take also note that we actually put these servers into real production (larges internet portal in Poland – http://www.wp.pl).
Ehrm….
As far as I know seeing from previous benchmarks. Suns Niagara architecture is already the winner on the Watt<->Performance deal.
This would only show that Sun is even better again….
“That’s why we got 2.5x performance of 2x Xeon 2.8GHz (signle core).”
So, a Sun x2100, 2 cores @175 2,2GHz, 1,3x faster than a 2x Xeon 2,8GHz?
Yes, well, no: that advantage is realistic only in a test heavily biased to multiple threads execution.
“This benchmark was all about WWW – nothing more.”
OK.
But that means: this benchmark is mainly (if not only) about capability in dealing multiple threads; as I prepended, no wonder that more cores = best results.
(in the case of Opteron vs Xeon, this advantage for Opteron on certain kinds of task is well known and well documented)
And no wonder that a newer design specifically meant for low power ultradense systems is more adapt for that use than CPUs like Opteron or Xeon not meant for low power ultradense systems, it’s obvious. If you want to buid a TDP whise x86 system today you should go on EE Opteron family or on mobile Intel processors (or wait for Conroe).
In other words, I belive facts in the benchmark are true and correctly reported, however I wonder if it makes sense (and if it may be not misleading for an unskilled reader):
– make a comparison based merely on multithreading capabilities from 8 core with to 2-4 cores system… on the other hand would it not be misleading running a signle threaded encoding task on a budget PC with single 3,8GHz P4 saying “see, it beats 4 Opteron cores an the new 8 cores SUN CPU (in application xyz)!”?
– make a comparison of something build ground up to be a low power cpu for dense systems with something that IS NOT MEANT to be low power… no wonder it will win!
– make a comparison between a product that impose (afaik) a vendor lock in and something that doesn’t.
This compairson is more revelent then you are giving it credit for. The reason I say that is because IBM was out at our company last week trying to sell us x335 as a solution for our WWW hosting. This is the server/model that they they seem to push in these environments. The T1000 is the model that sun pushes for these environments.
That does not mean that I totally disagree with you from a pure technical perspective there are ways to make a more technically correct compairson. But this is a good real world compairson based on what vendors are pushing for this soluction, and how they stack up against eachother.
As Milek has said, I don’t think the point of this is
to show that Sun’s 1 socket/8 core machines trounce
a single (or dual) core x86 box. The relevant points are
that the T1000 and T2000 occupy 1ru and 2ru, have very
good power and cooling requirements, have very good
reliability and RAS features, good price AND give great
performance on this type of workload. These are the
factors most applicable when choosing hardware for this
purpose, not how many cores it has. When you get can
an 8core x86 box in 2ru then that would be a relevant
comparison.
That is the whole point of the benchmark, to show that a computer that is cheap to buy and cheap to operate (not to mention that with ALOM and such it’s enourmeously more practical to manage) is also performing well.
Maybe price/performance wasn’t explicitly written all over the place, but that’s because the author (erroneously, it turns out!) thought he was dealing with readers who have a bit of common sense.
Edited 2006-02-20 16:37
“Maybe price/performance wasn’t explicitly written all over the place, but that’s because the author (erroneously, it turns out!) thought he was dealing with readers who have a bit of common sense.”
This is one of the worst attitudes a person can have. I emplore this of everyone, don’t assume your readers know what you know. Friendly discourse and the willingness to listen and learn are hallmarks or progression. Expecting someone to know what you know then saying something condescending like this should be common sense really isn’t productive.
If you tear apart the blurb and the blog, there is NOTHING written about price or price/performance. In fact, the blurb misleadingly states that the (singular) Niagra processor trounced the dual Xeon processor by four times the performance without mention of price or price/performance. If that was the point of the blurb-blog, then why as I as a reader, didn’t get that from the article, especially since the blurb-blog nevers mentions that?
These Sun boxes look cool to me. I’m shopping around for new servers right now, and a 6-core T1000 seems to be priced at a nice point.
The big problem I have though is according to Sun’s on-line shop they ship with a single 80GB hard disc. Where are the build configuration options? The other contender for this server role I’m considering is a Linux box with 2x500GB hard discs with dual Opteron processors… I need more than 80GB storage – looks like I’ve got to call Sun up to get anything besides their 4 pre-configured options.
Seems Sun has fixed a few things. They are beating all competors in the rack space market. If someone disagrees think about this what is the most important things in the datacenter.
* Size
* Power
* Heat
As shown by benchmark Sun Boxes on power race, check the specs they win the others as well. Though lots of people want to dismiss the benchmarks as being apples to oranges but i dare you to find a faster box that uses less power and fits in a 1RU or 2RU space than a t2000 doing typical network based workload. If your workload isn’t suited to the T1000 or T2000 there is always the X4100 and X4200 opteron powered boxes that beat all the intel boxes, and ones that will be released in the next 12 months.
What is really amazing is that in the next year or two Sun will be releasing the “rock” platform that puts a floating point unit on each core, and you can only guess what the speed will be, imagine if they get it running at 2ghz, almost double the speed the ultraT1 chips, now you have a box would replace rackfulls of last years fast boxes in a 2RU space. Sun has allready began replacing Sun Enterprise 10000 boxes with a T2000 as documented at http://uadmin.blogspot.com/2005/12/300u-to-8u.html
The next generation will replace other large sun boxes perhaps Sun will end the datacenter space and heat problems singlehandedly with these chips. I wouldn’t want to be the one that dismissed Sun because of a benchmark you chose to ignore. (apples to oranges) doesn’t seem like a relavent argument when they save your company space, and money.
Of course, there is life below 1U; specifically blades which can pack quad-core Opteron systems into 0.5U.
Credit where credit is due, this is definitely the type of application that will flatter Sun’s new chip.
Now if they could just become a company that didn’t screw their own VARs, they’d be in business.
It’s true that it beats intel CPU but this doesn’t make sparc a winner in my opinion. What I really appreciate is performace price; Let’s Look at this:
1. Sparc based servers are almost less than 2x speed of AMD servers (see http://milek.blogspot.com/2006/02/t2000-real-web-performance.html )
2.SPARC based server (T2000) is 4.6 times more expensive than AMD (X4100) (see http://store.sun.com/CMTemplate/CEServlet?process=SunStore&cmdViewP… for X4100 most expensive, which will be 5800$)
(T2000 most expensive offer is 27000$ see http://store.sun.com/CMTemplate/CEServlet?process=SunStore&cmdViewP… )
Actually SPARC platform should be stopped for the sake of Sun survival in the future, I don’t want see them face SGI crisis.
x4100 config for 5800$ has only 4GB of RAM. T2000 for 27000$ has 32GB of RAM, and additionally it has faster CPU than I used (I used 1.0GHz, in this config there’s 1.2GHz).
So looking at the prices thel look like:
T2000, 8GB RAM, 1.0GHz T1 – 13,400$
T1000, 8GB RAM, 1.0GHz T1 – 8,000$
x4100, 8GB RAM, 2x 275 – 7,100$
2.SPARC based server (T2000) is 4.6 times more expensive than AMD (X4100) (see http://store.sun.com/CMTemplate/CEServlet?process=SunStore&cmdV… for X4100 most expensive, which will be 5800$)
(T2000 most expensive offer is 27000$ see http://store.sun.com/CMTemplate/CEServlet?process=SunStore&cmdV… )
As some one else also pointed out these two systems aren’t comparable. The T2000 box has 8x the memory of the X4100. The X4100 has no PCI Express slots. That much memory costs money in any system.
You wouldn’t be able to run the same load on the boxes because the 4GB X4100 would trash like crazy running a large memory load that the T2000 could easily run. You can’t even populate the X4100 with more than 16GB RAM. Not to mention that the T2000 in the article had 8Gb RAM and was the 1Ghz version.
The UltraSPARC T1 consumes 72watts each fo the dual core Opterons in the X4100 consumes a 100w totalling 200W for the cpus alone. All else being equal the 2x Dual core Opteron is consuming almost 2.5x the power for little less than half the performance of one 1.0Ghz UltraSPARC T1. I think that speaks for itself and why Sun should continue to invest in SPARC’s future.
Edited 2006-02-21 01:45
I see a little bar graph. No details, no disclosure, no analysis, nothing. So, where is this mysterious benchmark?
The T1 (aka Niagara) was specifically designed for this type of job so it’s no surprise it did well. There are several other (much better documented) benchmarks on which it does just as well if not better (See the TCP or SPECjbb results).
Chip for Chip, for heavily threaded web type apps it eats other processors for breakfast, there’s no competition.
If you want the best price/performence solution, a Sun Fire T1000 Server with 6 cores, and 2GB of ram cost only $3,625, which is both cheeper and faster then the ibm solution :}
It does only include a single harddisk, so you would need to use iSCSI or a nfs server, if you want to serve much data, but still a good deal :}
Looking at the results of various benchmarks I can’t wait to get my hands on the loaner I asked for, which I should hopefully get this week! Just in time for a proof-of-concept “dog and pony” for an upcoming requirement.
In the past I always assume that Sun is equal to an expensive machine, now it changes with the release Niagara. And it will be interesting later when each core has its own fpu.