AMD's Six-Core Istanbul Opterons

EconolineCrush writes "AMD's latest 'Istanbul' Opterons add two cores per socket, for a grand total of six. Despite the extra cores, these new chips reside within the same power envelope as existing quad-core Opterons, and they're drop-in compatible with current systems. The Tech Report has an in-depth review of the new chips, comparing their performance and power efficiency with that of Intel's Nehalem-based Xeons. Istanbul fares surprisingly well, particularly when one considers its performance-power ratio with highly parallelized workloads."

Comments

Istanbul runs your shells

[smittyoneeach]

Istanbul runs your shells
Through shaves as tight as Dardanelles.
Use Opteron and the gallant foamy,
And thus avoid Gallipoli.
Burma Shave

Re:Istanbul runs your shells

[drinkypoo]

Istanbul was Constantinople
Now it's Istanbul, not Constantinople
So if you were waiting for a core called Constantinople
It's been released as Istanbul.

Re:Istanbul runs your shells

[smittyoneeach]

The Turks are casual, until you name a chip "Armenian" or "Kurd".
Then, not so much.

Wasn't it called Constantinople?

[OzPeter]

Or isn't that anyones business but the Turks?

Re:Wasn't it called Constantinople?

[underqualified]

no no no. that was the beta version.

I won't be impressed until it is...

[Anonymous Coward]

Over 9000!!!!1

Fuck Everything, We're Doing Six Cores

[iamdrscience]

You think it's crazy? It is crazy. But I don't give a shit. From now on, we're the ones who have the edge in the multi-core game. What part of this don't you understand? If two cores is good, and four cores is better, obviously six cores would make us the best fucking processor that ever existed. Comprende? We didn't claw our way to the top of the CPU game by clinging to the two-core industry standard. We got here by taking chances. Well, six cores is the biggest chance of all.

Here's the report from Engineering. Someone put it in the bathroom: I want to wipe my ass with it. They don't tell me what to inventI tell them. And I'm telling them to stick two more cores in there. I don't care how. Make the cores so thin they're invisible. I don't care if they have to cram the sixth blade in perpendicular to the other five, just do it!

No.

[Timothy Brownawell]

Harnessing muli-cpu machines with these installed is going to be.... Interesting.

No more interesting than existing many-core machines.

Seriously, having a couple dozen or more cores is nothing new.

Re:No.

[default luser]

You've already made this comment before, and I've already responded, so I'll keep it short and sweet.

If you're using a slow 2.2 GHz Quad core, that's not the fault of the industry, that's the fault of YOU. I have already made it clear that the top-end Core 2 Duo chips would run circles around your P4, but apparently you'd prefer to pretend they don't exist. As for your dog-slow quad core, that was YOUR purchasing decision. You can purchase MUCH FASTER quad cores today for reasonable prices, but apparently you're still suck in the year 2006.

The reason Core 2 / Quad destroys the P4 despite having a slower clock speed: Core 2 ups the Instructions Per Clock versus the Pentium 4. The increase is between %60 and %100 more IPC. If you read my previous response to you on the subject, you'd actually know that, instead of continuing to spout your ignorant bullshit.

And if you can't find a video codec with multiple core support, you're looking in the wrong place. Video decode is one of those embarrassingly-easy things to parallelize, and so your "boast" is really just outing you as a lazy bastard who can't take five seconds to search Google.

Re:Finally

[Kotoku]

I'll be finally able to run Crysis at a decent framerate.

Just in time to be behind the curve for Crysis 2!

Another test at anandtech.com

[IYagami]

http://it.anandtech.com/IT/showdoc.aspx?i=3571

Includes information about virtualization performance: http://it.anandtech.com/IT/showdoc.aspx?i=3571&p=9

Conclusion:
"The six-core Opteron is not an alternative to the mighty Xeons in every application. The Xeons are more versatile thanks to the higher clockspeeds, higher IPC, Hyperthreading and higher bandwidth to memory. The Xeon 55xx series is clearly the better choice in OLTP, ERP, webserving, rendering and there is little doubt that it will continue to reign in the bandwidth intensive HPC workloads. There are two types of applications where we feel that the AMD six-core deserves your attention: decision support databases and virtualization."

Re:Another test at anandtech.com

[Vancorps]

I believe Anandtech is showing it's bias here. I had heard great things about the Xeon 55xx series CPUs so I went and bought a couple of servers. Specifically one web server and one database server. I also had Opteron-based servers performing the same tasks. My webservers are load balanced using a hardware load balancer. During January I was under an extremely heavy load scenario. I ended up having to weight more traffic to the Opteron servers because the Xeons were choking under 100% cpu load. I barely squeaked by as all my servers were quite overloaded but once you get to a certain threshold Xeon performance seems to drop sharply.

Rendering Intel has always been king but everywhere else Opterons have performed better for me again and again. I'm 100% 64bit though and I haven't tested virtualization performance yet.

Re:Another test at anandtech.com

[Zork the Almighty]

Hyperthreading shows you eight fake cores which map to four real cores. I benchmarked it extensively. Computationally intensive routines with a small memory footprint can gain up to 20%. Bandwidth or memory intensive routines can lose up to 50%. In the extreme case, 8 threads on virtual cores can be half the speed of 4 threads on 4 real cores on a Core i7. Keep in mind, this is on a crazy application that generates lots of data.

If your algorithm is designed to break up the problem to exploit the cache then hyperthreading is a bigger mess. The data for thread 1 and thread 2 (out of 8) might be complementary, but the operating system will run those threads different actual cores, because all it sees is the virtual cores. This can be very inefficient if you need the whole cache.

Perhaps worst of all, you are stuck always running 8 threads. 2-6 threads may not be distributed evenly across the real cores, leading to inconsistent performance. Therefore, you may lose performance by attempting to scale the problem further than it is efficient to do so. With real cores, I can decide (based on problem size) the correct number of core to use.

In conclusion, hyperthreading has its uses, but operating systems are oblivious to it and that's a major problem with more than one core.

And 14 cores is nothing compared to 64 threads

[IYagami]

From http://www.sun.com/processors/UltraSPARC-T2/features.xml

"Features and Benefits
With eight cores and 64 threads on one chip, integrated 10 GbE networking, crypto, and PCI-Express expansion, you have the jump on anything else on the market. The opportunities for system consolidation and virtualization are here like never before. Consumes less power per core and thread than any processor in its class - without compromising on performance. The UltraSPARC T2 processor gives OEMs a massively threaded, multi-core alternative to more power-hungry, less threaded processors from competing vendors."

Scary Quote from the Article

[Gazzonyx]

[...] Not only that, but it's hitting the market early. AMD had originally planned to introduce this product in the October time frame, but the first spin of Istanbul silicon came back solid, so the firm pulled the launch forward into June. Even with the accelerated schedule, of course, Istanbul comes not a moment too soon, now that Nehalem Xeons are out in the wild.

Does anyone else think that this seems a little convenient? I'm really hoping that they didn't just tone down the testing to make it to market. I'm thinking they'll go to market and then quickly release a new revision to fix the corners that they cut the first time around. I hope I'm wrong, but AMD has been slipping lately.

Any EE's out there know the process well enough to confirm or deny my suspicions?

Re:Scary Quote from the Article

[Narishma]

From an interview with bit-tech:

bit-tech: Has the launch of Istanbul been brought forward in response to Nehalem EX's updated launch date?

Patler: Istanbul being pulled in by five months is a result of excellent execution by our design and manufacturing teams who were about to take it from first stepping of silicon to production. Also, the fact that Istanbul is based on our existing socket infrastructure, enables our OEMs to save time on validation cycles that are normally associated with a new processor that delivers the performance Istanbul can.

Re:Scary Quote from the Article

[dpilot]

I'm in the silicon business. Not CPU, but still silicon.

It sounds as if AMD budgeted time for another pass at the design, and turned out not to need it. The amount of time they pulled out of the schedule looks more like a silicon pass than short-cutting testing and validation. Adding that extra pass, and making sure it was scheduled is probably a result of having been so badly burned last time, but that's good. You can always be a hero by doing better than plan.

Re:Scary Quote from the Article

[Anonymous Coward]

Actually, testing was increased for 6Core... As our 4Core tests no longer stressed a system with 50% more cores the same way.

What changed was Process. 6Core uses all of the 'good' tech from Shanghai, then implements a few things differently (rev upgrades, etc). The reason 6Core launched soo quickly, is we learned all of our lessons on the initial quad core fiasco. We did things 'right' this time, and the result is... a launch date that is nearly 12mos ahead of the initial schedule (which was set 2yrs ago).

Personally, I trust the 6Core parts moreso than our 4Core parts... but maybe that's cause i've been testing the 6Core parts for over 8 mos and have had realatively few problems.

Trust me, we can't afford to fail, so there's no way they're going to cut corners. The last thing we want is another Cache Disable fiasco. Mark my anonymous word, 6Core is a fully tested and mother approved processor.

Re:EPT?

[KonoWatakushi]

AMD has supported nested page tables since the Shanghai series processors.

'so what' is that :

[unity100]

ancient egypt is THE source of many of your philosophies and sciences. from 1000 BC and onwards, early greeks were coming to egypt for education. egypt had 2 schools - school of life, and school of death. school of life was teaching stuff related to this world, ie, medicine, land registry, writing, government, and school of death taught stuff pertaining to abstract world. not to mention that many of the professions people identify themselves today originated in egypt.

even before knossos was known, medicine men and wise men of egypt were world renowned, even legendary in their time. a LOT of stuff that is ascribed to greeks were what greeks learned in egypt.

brush up on your history.

it's interesting, but not becase of 6c

[markhahn]

the real news here is not the extra couple cores, but coherency snooping. this feature will make 4/8s machines far more attractive; it doesn't hurt that with 48 cores and 32 ddr3/1333 dimms, you have quite a monster. _and_ incidentally something that Intel can't currently answer.

there's no question that nehalem has put a serious dent in the market, but Intel's going quite slow in rolling out higher-end products. yes, a nehalem socket delivers about 50% more bandwidth than a current opteron socket, but show me the 8s nehalem machines. nehalem-ex is coming, but how soon and at what price?

one thing I haven't seen is any attempt to measure real SMP performance on new-gen chips. I don't mean something like Stream or VMs, where there is no real sharing inherent to the workload. how long does it take to exchange a _contended_ lock between cores (in the same socket vs remote)?

finally, the real question is whether there is actual demand for more-core chips. I'm in HPC, and we always want more, and throw good money. but it has to be smart more - the 6-core core2, for instance, was just asinine because even 2c core2 is drastically memory-bandwidth-starved. nehalem-ex seems quite promising, but if it's cheaper to cluster dual-socket machines rather than pay the premium for 4s's, the 4s market will be stunted and less successful in a self-fulfilling way...

I'd rather have faster disk I/O

[PeeAitchPee]

We run a lot of commerical OCR (as in millions of images), which is extremely processor-intensive, disk-intensive, memory-intensive, you name it. Our current main OCR server is a dual quad-core Xeon X5355 box with 16 GB of RAM. Our OCR software multithreads and the processor is no longer the bottleneck -- it's now disk I/O. While current drives continue to increase in size, their read / write speed is what keeps us from getting work done faster. It now takes several orders of magnitude longer to build, and then export, for example, a 2 GB batch than it does to recognize it, and the holdup is entirely due to disk I/O.

SSDs help. We recently upgraded our server's OS drive to two Intel Extreme 64GB SSDs in RAID 0 (also using part of the array as a "scratchpad" for the OCR batches), and that cut the disk I/O time approximately in half -- but we're still talking almost an hour for your typical 2 GB batch. Time is money, and we'd gladly throw more money at faster infrastructure were it available. SSDs are still way too expensive to replace our existing main storage arrays, though.

So, while I appreciate continuing work in processor speed and density, I'd say I'd rather see a commensurate increase (and reduction in cost!) in disk speed at this point. Just my .02.