When Oracle announced the eight-core Sparc T4 processor and the four systems that make use of the new chip to run Solaris workloads two weeks ago, the one thing that was missing on announcement day was pricing for the systems. Having lived through Oracle OpenWorld, El Reg figures now is a good time to circle back and take a look …
is good. It forces every vendor to deliver better products, and us consumers in the middle benefit. :o)
So much for cool threads the heat is cranked on these....
T4 is twice the software price. Given our software stack of EE, partitioning, compression, security., etc The price of software dwarfs the hardware price. In looking over the charts Sun sent over I see the roadmap says: 1X which means same performance per box as T3 and
5X single strand which means if only one app is running with only one thread you will get 5X the performance. Cheers to Oracle for coming out with new boxes, but the T has never had good value for Oracle software and the 2X price increase means they have no desire to improve the price performance. We will keep focused on virtualization.
I am quite certain lot of people dont agree with your claim that SPARC T has never had good value for Oracle software, because T cpus holds some world records. Why dont you start again with "the Niagara is cache starved and slow"? How is this possible if Niagara holds world records?
Regarding Oracle increasing prices - this is a sign of strength. A company shows its worth by charging a premium. If a company is lowering their prices, or even do things for free, then it is a sign the company dont trust their products.
POWER6 servers costed 5-10x more than x86 servers. POWER7 costs 3x more. POWER8 will maybe have even lower prices - this is not a good health sign. Meanwhile, Oracle is increasing their prices, confident that customers will still buy the products. So you can complain on Oracle's high prices compared to IBM's low prices - but that is how business is.
Lack of basic benchmarks
SUN used to at least publish T-series friendly basic processing benchmarks (like SPECjbb). Now Oracle are even suppressing these. These show-piece application benchmarks are all very well, but they bear no resemblance to how these ERP packages tend to get configured and used in real life.
Sensible people will be asking how do these machines compare with x64 on throughput, response times, costs, power consumption and so on. That's where the real competition is.
nb. T series performance can be a very tricky thing - once those H/W threads start competing for the core resources, then you can get dramatic worsening of response times (depending on app). Of course a carefully tuned app benchmark is not going to be entering those dangerous waters. It will be tuned for the marketeers, not systems people...
A quick look and it seems that the T4 beats both Power/Itanium/X64 on TPC-H on a per-core basis (IBM's favorite measure), and by 2-6 times on a per socket basis . I'm sure they're working on a SPECjbb benchmark. Can't do them all at once!
I won't count any chickens
No SPECjbb's were released for the T3 - I don't see them doing it for the T4.
SPECjbb (and Specint for that matter) is a very easy benchmark to run - it takes a tiny fraction of the resources of one of those big app benchmarks.
When I was playing with 286 based Hypercubes, I herd a comment about benchmarks from the guys who made the computer: "if the benchmark says it will do 1000 Foos per second, I can bet you can not find a way to ever get it to do 1001 Foos per second".
I love this stuff...
I cannot wait for Oracle to get their result published on tpc.org. That will mean that their result was replicated (always important with these claims). The IBM result was put out by IBM, so they cannot cry 'foul'.
Both are in trouble with Westmere-EX (ref: http://sqlblog.com/blogs/joe_chang/archive/2011/10/10/tpc-h-benchmarks-westmere-ex-versus-risc.aspx ).
I agree. It looks like IBM is holding back Power 7+ a little to counteract this outcome. If this pushes IBM and Intel to work harder against Oracle, that is GOOD for my shop...
"...Both are in trouble with Westmere-EX (ref: http://sqlblog.com/blogs/joe_chang/archive/2011/10/10/tpc-h-benchmarks-westmere-ex-versus-risc.aspx )...."
I would instead reformulate this as "both are in big trouble with SPARC T4" - because the T4 results are only 4 cpus. And T4 is almost the fastest, even though all other vendors are using twice the number of cpus. Imagine the next year when 8-socket SPARC T5 comes out. It will kill everything. Not to speak of the 64 cpu, 16.384 thread, 64TB RAM Solaris monster in a couple of years - I am convinced that Intel and POWER will have trouble. Yes indeed. Big trouble.
The big question...
Will T4 enable moving off of UltraSPARC IV+ platforms without upgrading to M series? Crunching the numbers, the answer appears to be: YES
@Steven - The value in the T processor is far beyond integer/floating-point performance.
When comparing performance with x86 platforms, the crypto units provide a significant throughput boost with real world applications. With the T5, the inclusion of compression/decompression will increase throughput boost with real world applications, again.
I guess the message is: if you are marketing, do x86. If you are business, do SPARC.
I think a simplistic view has been taken on the performance comparison in the article. One thing that isn't obvious about the T4 is that for smaller workloads, the CPU will turn off the extra threads, dedicating the resources of the core to a single thread.
In real terms, this means that you can run the system all out (max threads) at one performance level, and, at 1-thread-per-core at a much higher performance figure.
I believe in a comparison of T3 and T4, the all-threads performance figure and the 1-thread-per core should be used. I believe you'll find the performance for the 1-thread-per-core to be far superior to the T3.
Re: Performance skewed
Well, yes, it was a simplistic view. And it said so in the article.
Capcaity versus throughput
Well, the capacity seems to be around the same for the T3 and the T4.
Where capacity is the SUM of throughput for all the threads at the optimum number of threads.
Now the behaviour of the T4, I guess, will be much more like POWER7, when you add threads and surely that will mean that a single thread per core will have the full benefit of the core cause it'll be running SMT and not fine grained multithreading.
So.. T4 is a huge step away from what Oracle have been singing for these last years. So hopefully there are a lot of people who didn't go the T1,T2,T3 way who now will have a much smother migration onto new hardware.
- Product Round-up Smartwatch face off: Pebble, MetaWatch and new hi-tech timepieces
- Geek's Guide to Britain BT Tower is just a relic? Wrong: It relays 18,000hrs of telly daily
- Geek's Guide to Britain The bunker at the end of the world - in Essex
- Review: Sony Xperia SP
- FLABBER-JASTED: It's 'jif', NOT '.gif', says man who should know