Intel Unveils 10-Core Xeon Processors 128
MojoKid writes "Intel announced its new E-series of Xeon processors today, claiming that they will deliver nearly unparalleled advances in CPU performance and power efficiency. It has been just over a year since Santa Clara released its Nehalem-based octal-core Beckton processors. Whereas Beckton was focused entirely on performance and architectural efficiency, these new Xeons are more balanced. The new chips boost the core count to ten (up to 20 threads with HT enabled) and will be offered at a wide range of power envelopes. The new E7 series incorporates the benefits of the Sandy Bridge architecture, its support for new security processing instructions, and its improved power management technology. Intel has also baked in support for low-voltage DIMMs, which allows vendors to opt for 1.35v products."
unparalleled (Score:5, Funny)
What's the point of having 10 cores then ?
Re: (Score:1)
I was wondering if anyone saw the irony. Thanks :-)
Re: (Score:2)
There's a good reason they aren't touting it. Turns out that making logic gates that can sense 8 different voltage levels is much more complicated, area-intensive, and slower than just being able to slam the output to one of the voltage rails as hard as you want/need. They tried and tried, but it was a dog all around.
Then came the real breakthrough -- the realization that you can encode the 8 states in octal using 3 normal binary signals, and so simulate an octal circuit, only faster and smaller!
But then
Re: (Score:3, Insightful)
claiming that they will deliver nearly unparalleled advances in CPU performance
What's the point of having 10 cores then ?
I laughed at NEARLY. Nearly unparalleled is not unparalleled. Not unparalleled is... well, paralleled.
Specs (Score:5, Informative)
Re:Specs (Score:5, Informative)
Re: (Score:2)
SGI is going to use up to 256 of these in a supercomputer, bring some weenies and marshmallows.
Oracle’s 11g database can run 10 times faster at encryption tasks using the E7 series chips. Lots more performance comparisons and the 19 companies names Bull, Cisco, Cray, Dawning, Dell, Fujitsu, HP, Hitachi, Huawei, IBM, Inspur, Lenovo, NEC, Oracle, PowerLeader, Quanta Computer, SGI, SuperMicro, and Unisys [digitalninjastl.com]
Re: (Score:2)
130W seems impressively low for 10 fast cores (20 hyperthreads) if you ask me. Just a few years ago that would have been an entire cabinet full of P4's complete with a dozen power supplies, etc.
Re: (Score:2)
It's called Moore's law.
Re: (Score:2)
Re: (Score:2)
I would not put too much faith on the theoretical 2x performance increase offered by AVX this generation. Although the processing hardware is there, the processor itself is not designed with enough fetch width to keep the units fed.
See Here [lostcircuits.com]
Summary: a few tests topped %20 improvement.
And Here [phoronix.com]
Summary: a couple of tests topped %20 improvement. Many tests produced slower results with AVX enabled.
The overall performance boost looks to be around %20-30, at least for now.
Re:Specs (Score:4)
Why is this better than a 12-core Opteron with quad-channel DDR3?
Because each Intel core is a lot faster than each of the 12 cores on the latest Opterons. For many workloads, single-thread performance still matters. Search for "SPECint2006 Rate [spec.org]" results on the latest processors - a latest-gen Intel core is about twice as fast as an AMD Bulldozer core. 2*10 > 12
Re: (Score:2)
Because that 12-core Opteron is only about equivalent to Intel's 6-core Xeon. [anandtech.com]
Re: (Score:2)
Because Intel chips are currently better than AMD chips, clock for clock.
The only place where AMD has an edge is Zacate (AMD integrated graphics + a duel core Bobcat - it's Atom killer), and *maybe* Llano (too early to tell).
Those are budget systems, which leverage AMD's ATI acquisition. Intel HD is miles ahead of the old Intel graphics, but I think Zacate is better.
However, those are low-power consumer chips. Not server chips. Intel still owns the server.
"nearly unparalleled" (Score:2)
Re: (Score:1)
Re: (Score:2)
Re:Unparalleled? (Score:5, Funny)
I'm sure some Sun engineers would disagree.
...if they weren't busy looking for a new job.
SB is no joke performance wise too (Score:5, Insightful)
I've got a SB desktop computer and it just screams. they made some sizable per-clock performance improvements. Also AES-NI is no joke. I am pretty amazed by the speed. Tryecrypt supports it and the benchmark difference is huge. With a 100MB buffer a pure software AES implementation benches at 649MB/sec on my system (553MB/sec for Twofish, 254MB/sec for Serpent). Same test with AES-NI on, 2.7GB/sec. That is 4.2x the speed.
Could be really useful for web servers, particularly if you are looking at going all SSL all the time.
crypto performance (Score:1)
Could be really useful for web servers, particularly if you are looking at going all SSL all the time.
Sure, except you also have to worry about algorithms like RSA, DH, SHA1, SHA256, SHA384, SHA512, and providing a good source of random bits as well (hardware RNG).
That's one thing I like about Sun/Oracle's T-series chips: they have all the above algorithms. A ~4 year old T2 chip can do RC4 at 81 Gb/s, AES-128 at 44 Gb/s, and AES-256 at 31 Gb/s: that's enough to saturate 10 GigE connections (in both directions at times). SHA-256 at 41 Gb/s and RSA-2048 at 6400 op/s make for very fast web secure servers.
http
Re: (Score:3)
You should clean or replace the fans then.
Re: (Score:2)
Yeah, get rid of the fans you have and install some golf fans instead. They don't tend to scream. Or maybe some Lions fans, since they don't really have much to scream about?
Re: (Score:2)
Could already buy 12 core AMD chips last year (Score:2)
processor : 47 vendor_id : AuthenticAMD cpu family : 16 model : 9 model name : AMD Opteron(tm) Processor 6174 stepping : 1 cpu MHz : 2200.000 cache size : 512 KB physical id : 4 siblings : 12 core id : 5 cpu cores : 12 apicid : 75 fpu : yes fpu_exception : yes cpuid level : 5 wp : yes flags
Unfortunately AMD's performance is lagging (Score:5, Informative)
Hopefully Bulldozer will fix it but right now, they don't do so well. Have a look at this HardOCP article on the new SB processors (http://www.hardocp.com/article/2011/01/03/intel_sandy_bridge_2600k_2500k_processors_review/3). They tossed in a high end 6 core AMD CPU too. It just gets killed. In many tests, it is below the older 4 core i7 CPUs, in pretty much all of them it is below the 4 core SBs and I don't see a one that it beats the 6 core i7 (the 980X).
AMD offers more cores, but their cores don't do as much. Don't buy in to core hype any more than MHz hype or anything else. More is not automatically better. Have to run benchmarks on it and see how it actually does.
Like I said, hopefully Bulldozer will change that. Hopefully it'll be competitive with Intel per core, per clock and so on. However right now Intel processors just kill.
Re:Unfortunately AMD's performance is lagging (Score:4, Interesting)
Re: (Score:3)
Not necessiarly (Score:2)
Not all tasks are of an equal load at all times. You can have situations where there is a strong load on less cores, and then later ones with a load on all cores. That's the reason for such a thing to exist.
Not only that but turbo boost doesn't just work with how many cores are loaded, it has to do with electrical load and thermal load as well, hence why it can be active even if all cores are loaded. So it allows for a chip to meet a given TDP, but then if cooling is good it can go faster, and scale back if
Re: (Score:2)
Re: (Score:2)
Re: (Score:3)
no, we don't have any real info on how fast BD is yet. some technical papers on it indicate it's designed specifically for high clock while maintaining control on power dissipation. the shared FPU is somewhat faster than an unshared FPU would be, so this is a good choice, especially for code that's not always FPU-bound. and AMD has said that BD definitely has thermally-limited clock boosting.
Re: (Score:3)
But the other poster is right.. you dont generally want turbo features on servers. This 10 core Intel server chip is in the same boat as AMD's 12 core server chips, as it will under-perform for single threaded tasks. These chips simply arent made for single-threaded performance.
AMD has been king of the multi-CPU solutions for awhile now [cpubenchmark.net] so we will have to wait and see how Intels new line will stack up in 4xCPU
Re: (Score:2)
Yeah, Sandy Bridge has slaughtered AMD for the current generation.
I've been a long time AMD supporter, just because I don't want there to be only one kid on the block (competition is good and all that), but about three hours ago I just bought a 2600K and matching mobo and ram on Newegg. First Intel CPU I've bought since the old $90 Celeron 300A that was 50% overclockable out of the box.
Re: (Score:2)
supporting some company because you want competition is a pretty bad was to have competition. Buy the bast bang for your buck, if only one player can deliver, so be it,.
As long as we continue to have good regulation so another competitor can rise up, all is good for the consumer.
Re: (Score:2)
>>supporting some company because you want competition is a pretty bad was to have competition.
If you have two equally good products (say the AMD64 vs. whatever Intel had at the time), it doesn't matter particularly which one you buy, so I always chose to support AMD. As I said, if we were down only to Intel, it would be a bad thing in general.
>>Buy the bast bang for your buck
Didn't I say I just bought Sandy Bridge? It's head-and-shoulders above anything AMD has right now.
Re: (Score:2)
Continue to have? CONTINUE TO HAVE? We haven't had decent anti-monopoly regulation since Teddy Roosevelt was president.
Re: (Score:2)
Apart from running twelve things at once?
I think you are comparing apples to aardvarks here. Something that is designed to run on a single core may run a bit better on a two or four core machine when everything else gets out of it's way but after that you won't notice any improvement.
I think you are looking at this in entirely the wrong way and are missing the entire point of these chips with a lot of cores and thus what you are using to measure their
Re: (Score:1)
I've been drooling over these; still a bit out of a college student's budget for a single processor motherboard.
What do you do that can use that much parallelization? (Not that I have anything I could use it for; I just want it anyhow. :)
Re: (Score:2)
What do you do that can use that much parallelization?
Having just spent $130 in processing time on Amazon EC2 to render a 50 second movie in 13 hours, I'd really like to have a few of these...
Re: (Score:2)
Neat chips, but for $400 you can get an i5-2500K+motherboard. You won't get 10 cores, but it's still stupidly fast ;)
Re: (Score:2)
Processing seismic data to see what is under the ground. The data is in the form of digitised audio tracks that are picked up by geophones that act just like an old moving coil microphone. Now imagine applying the same filter or other transformation to twelve million audio tracks. It's a task you could split to twelve million CPUs at the stupid extreme - that's how easily the tasks can be split to run in parallel. Of course in reality you want to dum
Re: (Score:2)
IF this was only about cores, you might have a point. I mean, the cores have to word well, hand threads well and so on. The AMD multicores don't do that as well as Intel.
Also, we ware talking about servers needed for high utilization. For a basic PC, or even a 'gaming Rig' the difference will be measurable, but not likely noticeable.
Re: (Score:2)
It IS only about cores or you are talking about something you don't need but others do. If you don't want to run a huge amount of things at once for sustained periods of time these machines are not for you. File servers, a typical web server or a desktop PC do not often see anything like those sort of loads so that's not what these things are for.
It's a completely different niche typically called High Performance Computing (misleading name because it's
E7 is not Sandy Bridge based (Score:3)
The new E7 series incorporates the benefits of Sandy Bridge
is a bit misleading, i think.
As far as i understood it uses the older Westmere EX architecture. So while it may have added instructions also available in the Sandy Bridge architecture, clock for clock it will likely be slower in most cases and probably won't reach the the clock speeds of Sandy Bridge based chips.
You do seem to be correct (Score:3)
Though these do not seem to have all the Sandy Bridge technologies. In particular, AVX isn't listed. Thus is does not seem to have new SB instructions. Maybe they are talking about the improvements to the existing AES-NI instructions (SB is faster with those than older i7s) however it does not appear to have new extensions.
I don't know that AVX is of much use to servers, but it does mean this isn't SB architecture. Which means it is not as efficient per clock (SB made some good gains in that area, not that
Re: (Score:2)
And *particularly* this series lags even "entry" level Xeons (the ones that go up to two sockets). Generally, about 6 months after an architecture makes it to desktop land the associated entry Xeons release, and then about six months after that the high-end Xeon implementation of that architecture comes out. Hence this announcement in the ballpark of six months after the low-end Xeon Westmeres (I think a bit longer, but too lazy to look up).
Re: (Score:2)
FYI, Nehalem-EX and Westmere-EP was released around the same time, March 2010 or so. For comparison, Nehalem-EP was released May 2009. Lynnfield was released September 2009 or so. The 45nm Clarkdale was released January 2010. The desktop Sandy Bridge was released January 2011. And now we finally have Westmere-EX.
Re: (Score:2)
For my scientific computing work, getting as many AVX units as possible onto a single motherboard is actually a big deal. So it would have actually been great if these 10-core chips had AVX.
Re: (Score:2)
Yea, why not call it the Xeon 7600?
Uncomplification. (Score:1)
"...they will deliver nearly unparalleled advances in CPU performance..."
Good to hear. Hopefully the next iteration will be fully unparalled. Much easier to program for a single core.
Re: (Score:2)
Yep.
Re: (Score:1)
My xmas list. (Score:1)
Hardware accelerated SSH
Hardware accelerated LZMA
Thats about it.
Re: (Score:2)
The AES instruction set [wikipedia.org] referred to as security processing instructions in the summary will accelereate ssh.
Re: (Score:1)
Thanks, missed that little gem. Will be interesting to see how a bulldozer handles this.
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
The AES instruction set [wikipedia.org] referred to as security processing instructions in the summary will accelereate ssh.
Is this anything like the accelerators Sun put in the Niagara line a few years back?
Re: (Score:2)
Shaving analogy (Score:2)
Re: (Score:2)
+1 - Oh yes.... What we neeEEEEEEd is points :)
It's not a power of 2 (Score:5, Funny)
Re: (Score:2)
Re: (Score:2)
What about multithreaded software that DOESN'T have engineering insight built into it? (ie. almost all multithreaded software)
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Your average game will use one core to keep the port to current generation of consoles simple.
Ummm... that might have been a current or valid point 5-6 years ago with PS2 and XBOX but not today.
Speaking as a game programmer, most of the industry has been working for years on optimizing games for multiple cores. The current generation of consoles (XBOX 360 and PS3) are multicore and the next generation promises to have even more cores. Heck even handheld gaming systems (Sony NGP has 4 cores), portables, tablets (iPad2 is dual-core), and phones are going multicore.
Re: (Score:1)
I'll take them off your hands.
Re: (Score:1)
Sad. From the comments, it seems nobody even remembers what a hypercube computer is.
Re: (Score:2)
Again proving there are 10 types of people in the world, those that understand binary, and those that don't.
Next Mac Pro? (Score:3)
I'm still quite content with my E5462-based 2008 model, thanks :)
Re: (Score:3)
I don't expect we'll see a Mac Pro announcement next week, as Apple's support for the AMD Radeon 69x0 cards is horribly broken in anything you can see outside of a lab in Cupertino. Announcing a system with "next generation performance" that uses previous generation video cards is something they are trying to get away from.
However, NAB is next week in Las Vegas, so who knows what they might do to get people to stop looking at CUDA-accelerated Adobe Premiere.
just what we need a $2500-$3000 1 cpu systems w (Score:2)
just what we need a $2500-$3000 1 cpu system with a low-mid range video card and like 2GB ram at the base price.
Re: (Score:2)
Re: (Score:2)
Indeed, while intel did introduce 8-core xeons the clock speeds sucked so much that really the only reason to use them was if you wanted a 4-socket system or had some other weird requirements (e.g. INSANE ammounts of ram). For two socket systems with up to 96GB of ram the 4/6-core 56xx series xeons were the best option, especially if the system is to be used as a workstation.
I suspect the next major revision on the mac pro will use some form of sandy bridge based LGA2011 xeon. Probablly with up to 8 cores p
octal-core? (Score:2)
Huh? Why are the core operating on octal? Does this new version run in decimal?
Rebooting (Score:1)
Big Deal... (Score:1)
Meh (Score:2)
Re:The NSA was running at 600mhz in 1950 (Score:4, Informative)
Re: (Score:3)
I skimmed the document, I didn't see anything particularly exciting about ATLAS or ABEL. ABEL apparently had drum storage and core memory. There's no way any of the stuff in that document was running at 600MHz.
The Cray-1 ran at 100MHz and the NSA and national labs snatched them up. There would have been no market for the Cray if there were secret machines running at 600MHz.
I worked in supercomputing in the late 80's and early nineties. At that time it was still possible to assemble processors out of dis
Re:The line of tinfoil hats (Score:2)
Nah, that was probably 100 years ago courtesy of F. W. Woolworth. Before that everyone liked quality too much. Unhardened steel hats!
Re: (Score:2)
First, MHz is a meaningless metric when independent of instruction set. Second, how much do you think the NSA was willing to pay for their 600MHz? I was talking to some IBM guys a bit about some of their high-end machine a while ago, and wondering who was actually buying them. Their reply was that they weren't allowed to disclose their customers, but 'most of them had three letter initials'. These machines had 1TB of RAM and up, and a price tag where optional features were measured in increments of mill
Re: (Score:2)