Slashdot Mirror
AMD's 12-Core Chip Cuts Software Licensing Costs
CWmike writes "AMD released on Monday its 12-core chip code-named Magny-Cours, doubling the number of cores over the previous-generation Opteron chip. While a doubling of performance is nice, another key benefit delivered by a chip with a dozen cores may be in reducing software costs. For Matt Lavallee, director of technology at MLS Property Information Network, a company that supplies real estate data, upgrading to the 12-core Opteron chip from his current quad-core chips will allow him to cut the number of servers — and his software licensing fees. While the 12-core chip costs a little more than an eight-core chip, it's 'nowhere near as much as a SQL server costs,' said Lavallee, who has been beta-testing the new chips. MLS operates 60 servers, and Lavallee said he could theoretically cut the number of servers by half but will likely reduce his server count by a third with the chip upgrade."
Reader adeelershad82 adds that AMD is hoping the new Opterons will compete with Intel in the high-volume server market.
Has MS updated their licensing to be "per-core" instead of "per-CPU"?
http://www.bynarystudio.com
Fair enough, but my Linux licensing costs won't change!
From hell's heart I fstab at /dev/hdc
Almost all of the enterprise software we buy charges by the CPU and by the seat. For this purpose a CPU core is the same thing as seperately socketed CPU. Whatever about OS savings I think you'd save more in hardware and running costs than you would on software.
I never get used to these constant resurrections
Oracle, MS and others change the licensing to require a charge per core.
No sig for you!!
Is that software licensing is a rip off to begin with.
upgrading to the 12-core Opteron chip from his current quad-core chips will allow him to cut the number of servers — and his software licensing fees.
Really? You mean, as computers get faster you *might* need fewer of them?
With the advent of the T1, you didn't need 24 DS0 lines, which saved me money on my telecom fees!
I would have thought the real-estate market downturn saved him a bundle on licensing.
THL phish sticks
They license per-core, so more cores per CPU can be more costly.
Why the heck is he paying anything? Just use MySql and be done with it. It is certainly easier to use/setup/maintain than that crappy SQL Server stuff. And it is free to boot! sheesh.
In my experience, it's rare for SQL Servers to be CPU bound, they're almost invariably IO bound, and having more cores won't help you when your disks are the bottleneck. I could see excitement over lowering per-machine costs for something like a renderfarm, but it doesn't seem likely to materialize for Databases.
I read the internet for the articles.
Very clever, AMD. Naming your chip after a location in Europe as usual, but this time making it able to be read as "Many-Cores" (or possibly more accurately "Many-Core", I don't really know how to pronounce French words). Very clever indeed...
Will I need to buy more SCO licenses for this one chip? This could get expensive.
Who would win this election: Andrew Weiner vs Andrew Weiner's weiner.
What are these? Is this something that afflicts Windows people still?
Stick Men
The other part is that until we have better tools (or devs as many of the ones I know are REALLY dependent on Visual Studio and .NET languages doing the hard stuff for them. I know that with the advent of the newer proc archs ASM is damn near impossible, I don't think it's unreasonable for someone other than kernel or driver guys to understand the ramifications of multithreaded app design. I've been looking at what it'd take to consider each proc it's on VM and use transparent memory sharing (like VMWare does) to treat each proc like a system unto itself and then treat things more like a distributed computing problem.
My only issue is that it's NOT a distributed compute problem so maybe I'm approaching it incorrectly but this highlights my problem, not many are well trained and experienced in this type of dev.
"Chinese Amazons, power armor, laser swords.... things just meant to be." - Shampoo, A Very Scary Bet
Anandtech has an excellent review of the new chip. The AMD chip is compared against the latest Xeon. In some situations such as OLTP and ERP, the AMD offering lives down to it's name Mangy Cores. In HTP and data-mining, Anandtech gives the nod to AMD.
So choose depending on your needs.
Really? You mean, as computers get faster you *might* need fewer of them?
No really. Please provide evidence for the thesis that as computers get faster, people need fewer of them.
Second point. It's usually the I/O performance anyway. A 12 core server is unlikely to be able to push as much throughput as 3 quad cores, given the same I/O technology.
Deleted
Huh, good point... I wonder if down the road you'll have to pay more for DB licenses that run on SSDs.
Why would the software vendor not be paid for the increased performance afforded by being able to access cutting edge technologies like Solid State Disk? They've to test their software against this equipment to make sure the additional performance doesn't break something don't they? Not charging would be like giving value away for free.
Ok, seriously I'd like to see somebody benchmark a Postres cluster running on a couple Westmere 2-socket boxes and backed by a well-engineered 6G SATA SSD OpenFiler 10Gbps iSCSI cluster against an Oracle backed by SAN. That would be hilarious.
Help stamp out iliturcy.
Wont help oracle licensing costs .. in fact it will raise costs, unless you virtualize..
---- Booth was a patriot ----
Someone needs to put "Advertisement" at the top and bottom of these posts of PR copy.
I really don't understand why these companies insist on changing from a Per-Socket to a Per-Core pricing scheme.
For years everything was single cores, and every 18 months the performance has doubled and the number of cores stayed the same. Yet licensing was still done per socket.
Now that the performance per-core is coming closer to a brick wall (per-core performance has gotten better over the years, but it's not doubling every 18 months anymore) the only way the chip makers can keep improving performance to pack more cores onto the die. How is the situation today different than it was 5 years ago when dual core processors took off? It reeks a lot of "dying-business-model-must-squeeze-every-penny-while-we-still-can".
...for Moore's Law.
No, AMD is not going to save software costs, on Oracle for instance, by using a 12 core processor when an 8-core Nehalem-EX processor outperforms the AMD at two-thirds the per core license cost.
This is AMD trying to get out in front of the issue that the overall throughput per core is much lower than Intel's current Westmere-EP 6-core. 2-socket or Nehalem-EX 8-core, 4/8+ socket cpus .
In virtually all per-core licensing scenarios (most of HPC and many big DB ( Oracle, DB2 ) and ERP apps) AMD Magny Cours is not competitive
Sure you can get more threads and CPUs for your SQL Server, but eventually it's going to be a cloud like environment where there will be massively partitions, very small databases, each using up 1 virtual CPU. Massive monolithic databases are how things are right now, but in the future, things like CPUs or cores will be foreign concepts.
well, this is nice to jack up the number of cores, but what about access to memory and other system resources? is hyper-transport getting "fatter" to allow more concurrent access to ram and such, i.e. will each core have a dedicated access to memory, for example?
maybe i'm just not getting it ("you're too old, go back to your punched cards", yada yada yada), but what's the point of upping the number of cores on a die if too many of them have to wait in line to access resources?
we're still dealing with micros, here, not mainframes. there are still echoes of the original bone-headed ibm 5150 design that have to be maintained for software to run. it's not as if we can go with a radical re-architecturing of the "wintel" microcomputer to accommodate these new multi-core cpus. so how do "they" go about it to ensure we are indeed getting more performance and not some hobbled design pretending it is faster?
Some IT manager will make a commitment to server reduction.
They'll then discover that consolidation of servers running the same application exhausts CPU or IO or networking or memory or [...] so you're forced to combine application loads on the same server to achieve a reasonable consolidation load.
Then they'll discover that 95% of windows and linux users are fairly single threaded users with simply threaded apps run one or two or three at a time. And that loading up the equivalent of 6-10 servers worth of varying applications on a single box pushes the software and hardware architectures well past what most users do. Which means you discover all sorts of new bugs and inefficiencies. And they get relatively little attention from the suppliers since they dont affect most of the users.
As you drive up the consolidation, the uptime not only suffers due to the architectural issues but the introduction of these complex issues stresses the support people, requires more sophisticated (and expensive) support people, and one previously minor problem now affects 3-10x the users and multiple applications all at the same time.
We used to have these things and they were called 'mainframes'. Those were designed from the ground up to have a thousand users and dozens of different applications and very high uptime despite the complexities. Unfortunately windows and linux and most servers arent made with mainframe like robustness. And yet we went away from those platforms because they had too many limitations.
McKusick's second law:
"The number of MIPS delivered to the keyboard has remained constant since 1974" -- Kirk McKusick
-- Terry
too far
will respond with new licensing schemes... Not a chance in hell.
it's amazing that intel has improved the speed of the their cores twice that of amd. amd now has to compete by putting two cores for intel's single core. now if intel would actually package 12 cores in a single cpu (much like the core2 quad days,) then that would kill amd. of course, intel wouldn't do this since they are doing well and enjoying very very big margins compared to amd.
this has been an exciting week, it's amd vs intel on the cpu side and now nvidia and amd on the graphics side with the release of fermi. this week has been a fast one! kudos to them and hope they continue to make better products so we can all benefit from it. :)
Live your life each day as if it was your last.
which computer would be the best for that? add your own computer if you like, these are just what i have found. Budget is under 700 bucks. Forex Trading
http://www.theonion.com/articles/fuck-everything-were-doing-five-blades,11056/
Just replace blades with cores...
These "savings" are irrelevant and temporary at best. People should remember that the monopolists selling them their crack can set the prices however they please. You can feel perfectly sure that if there's a strong trend to replace a dozen dual-core machines with 4 more powerful machines, doing the same job at 1/3rd the software-licensing-cost, the vendors will simply change their licensing to per-core.
So yes, short-term you may be able to save some money by such trickery. Medium-term it's a zero-sum-game though, the price will fluctuate back to the same point it always was: that point where the vendor believes (rightly or not) that the maximum profit can be made.
I would replace the oracle/mssql (+windows) monsters with better alternatives
like Ubuntu/Debian (or insert your preferred distro ) with an open source SQL database like Firebird , postgresql or mysql
i would really need a 12 core monster for my servers and with Firebird Classic or with SupperClassic
http://www.sinatica.com/blog/en/index.php/articles/firebird-superserver-classicserver-or-superclassic
developer http://flamerobin.org