Intel releases 15-core Xeon E7 v2 processor

At ISSCC last week, Intel shed more light on its 15-core Xeon E7 v2 processor. Today, the Xeon E7 v2 lineup made its official debut. Intel says we can already look forward to more than 40 systems from 21 different vendors.

The Xeon E7 v2 family supplants the Xeon E7 series, which was based on 32-nm Westmere-EX silicon and featured up to 10 cores. Xeon E7 v2 chips have up to 15 cores and can populate 32-socket servers. According to Intel, they deliver up to twice the performance, three times the memory capacity, and four times the I/O bandwidth, not to mention support for Intel’s Data Protection Technology.

The full lineup of Xeon E7 v2 chips can be perused here in the official product brief. 15-core variants with the largest thermal envelope (155W) run at 2.8GHz with 37.5MB of cache and 8 GT/s QuickPath connectivity. The lowest-power models in the list have 105W TDPs and run at 2.3GHz with 24MB of cache and 7.2 GT/s of QuickPath bandwidth. There was also talk of 40W, 1.4GHz models at ISSCC last week, but I don’t see those in the list. Perhaps they’re coming later, or maybe I’m just not looking in the right place.

In any case, Intel says the Xeon E7 v2 series is fit for a "variety of industries from retail and healthcare to banking and transportation." The lineup’s improvements over the previous generation can "help businesses that run mission critical applications including business support systems (BSS), customer relationship management (CRM), and ERP to operate more efficiently, at lower cost and with faster response times," the company claims.

Comments closed
    • auxy
    • 9 years ago

    Why aren’t you getting upvoted? ( ̄д ̄;) This is informative AND on-topic. Thanks chuckula!

    • ronch
    • 9 years ago

    [quote<]I don't see how those 16 core opterons are any good at all[/quote<] Ok, I'm not taking what you said and adding to it, but AMD's current Bulldozer-based chips aren't really *bad*. Perhaps they're just not as good as Ivy Bridge or Haswell, but they'll get the job done in a timely fashion, especially when we're talking about highly-threaded code. They may not be as energy efficient either, but that doesn't mean they're useless. Shifting the comparison to desktop processors, one can't really say that an AMD FX-8350 is bad. Perhaps it's not as compelling compared to a Core i7-3770K or 4770K in terms of performance as well as energy efficiency, but really, it still works well enough and is a great CPU to own. I should know, because I am using mine right now.

    • BobbinThreadbare
    • 9 years ago

    Not exactly an apples to apples comparison. Toyota sells reliability, Porsche doesn’t.

    • the
    • 9 years ago

    I have a system at home with two 16 core Opteron 6376’s. It is used as a graphics workstation for image editing and 3D rendering. I picked up the motherboard (Asus KGPE-D16) with an 8 core Opteron 6128 and 32 GB of vanilla unbuffered DDR3 memory. Over the past three years it has gotten an upgrade to 32 cores total and 128 MB of registered ECC memory.

    The main barrier to entry to the Xeon’s is simply one of cost. At the time I put this system together, it was at the end of the socket 1366 era. Other costs on the Opteron side were also cheaper. The main reason to go with a dual socket motherboard was for large memory support as the 32 GB limitation on Sandy Bridge systems would be the initial starting point. The Xeon motherboards that supported two DIMMs per channel had a price premium attached to them nor did all them support unbuffered memory. Motherboard prices were a bit higher, but not enough to be a deal breaker on their own. The entry price of the 8 core Opteron 6128 was significantly cheaper than the entry level Nehalem-EP. I figured that the entry level socket 2011 chips would be roughly the same price as the socket 1366 parts (turned out they were slightly more expensive). Timing was also a factor as I was able to get all the parts three years ago for the initial build before the beefer Xeons and Opterons came out.

    So ultimately why did I get a 16 core Opteron? Significantly lower price compared to 8 core Xeons, it was an upgrade to an existing system, earlier availability and better motherboards to scale memory capacity.

    The more interesting question is: would I still get that system today? I would be inclined to say no as socket G34 is basically dead. We saw a few Warsaw CPU’s launch earlier this year for the platform and that’s pretty much it until 2015. I always prefer to buy with an upgrade path on the road map (and for the record, I’d skip Ivy Bridge-EP as Haswell-EP is roughly 6 months away). I was hoping to get a Steamroller update but as soon as I found out that Warsaw would be Piledriver based, I upgraded from a single Opteron 6128 to two 6376’s. The price/performance is still pretty good on the 16 core Opteron even though the Xeons ultimately wind up being faster overall. Since the workloads are not always massively parallel on my box, turbo speeds do matter. The Opteron 6376 hits a nice sweet spot between high turbo speeds and a good base clock under full load. The entry 8 core Sandy Bridge-EP/Ivy Bridge-EP’s tend to have a slightly lower base clock but a far lower turbo. Though if I were to go up to an Opteron 6378, prices take a jump and the price/perfomance benefits start to erode against the Xeons. Things swing in favor of the Xeon’s at lower core counts. Intel has been generous with higher clocks here and while not cheap, prices are not outrageous like the high end Xeons. Still I’d wait for Haswell-EP half a year from now.

    • maxxcool
    • 9 years ago

    Hahaha… Nope good luck serving 15000/sec web hits on 4 cores…

    • jihadjoe
    • 9 years ago

    The 15-core chip is right on the limit of TDPs. 155W.
    Now would you rather they made it 12 cores and 3.5GHz? Or 8 cores and 4GHz?

    Servers are exactly the sort of work load where they should be prioritizing more cores. Sure a 4GHz i7 desktop might serve ONE web page 2x faster, but the 2.8GHz 15-core Xeon will serve 100,000 webpages more than 2x faster.

    • JdL
    • 9 years ago

    I admit it. I’m trolling. Mainly because I don’t see any high-Hz server CPU’s. Intel / AMD are priortizing going sideways, not up. Boom.

    • JdL
    • 9 years ago

    All absolutely true points. But my trolling is still valid. [Distributed] computing is only one aspect of server computing. I get a little sick every time I hear IT or OPS saying “hey check out that new Xeon. Our website would be SO much faster if we upgrade.” Really? Seriously bro? You can handle more traffic, yes, but it won’t serve the traffic you already have any faster.

    Ugh.

    • JdL
    • 9 years ago

    Dude SERVER processors do a whole variety of things, and everything ultimately comes down to how fast the CPU runs. Pretty sad when a Core i7 desktop serves web pages 2x faster than the latest Xeon CPU.

    • JdL
    • 9 years ago

    Skeptics. Critics. Democrats. I wonder how many of you actually work with servers! (OK… I admit to playing devil’s advocate here partly just for the fun of it, so just play along for the sake of the game 😉

    Here’s my thing. Here’s why I’m underwhelmed. I want a reason to upgrade my infrastructure, but I don’t see it here. Why?

    I have “upgraded” countless Java, .NET, and PHP architectures, MySQL and MS SQL databases, and more over the last several years from Core 2 Quad-based Xeons all the way up to the latest i7-based Xeons. You know what I find? Performance is roughly flat.

    Most of our apps are web sites / applications using custom dashboards, content management systems, etc. which frankly make up of the majority of the cloud computing market. How fast these can do database lookups and serve up web pages has not improved much. What incentive do we have to upgrade, other than keeping our power supplies and hard drives fresh?

    Recently we have started doing more with “big data” using technologies like Hadoop, Elastic Search, and Solr, and those certainly are more capable of scaling sideways across more cores. But this is not an “upgrade” case. This is a “new feature” case.

    The new CPU’s with greater density give us the ability to put more VM’s onto a single box, reducing space, cost, and power consumption. But for us, space isn’t an issue right now. Power consumption isn’t a huge expense either (remember, most of these are sitting idle, not running “compute”.

    SSD’s have provided the biggest boosts thus far, but now we are back to the CPU being the bottleneck. So going back to my previous statement… what ever happened to that 10 GHz transistor IBM was working on? What about Intel’s promises of 6 GHz or 8 GHz?

    Waaa waaa waaa yes I’m whining. But for real – true story bro.

    • UberGerbil
    • 9 years ago

    Well, you could leverage the work of others by either porting your data analysis to something that offers parallelism (R, say) or by using parallel libraries from your language of choice. That doesn’t entirely free you from having to learn parallel programming, but it may give you a benefit (at least, most of the time; of course, the odd case where you have to debug a cryptic race condition can still eat you alive).

    • UberGerbil
    • 9 years ago

    David Kanter’s [url=http://www.realworldtech.com/forum/?threadid=140143&curpostid=140143<]executive summary[/url<] over at RealWorldTech: [quote<]So Ivy Bridge EX launched today, and it's a pretty big step up from Westmere-EX. Here's a bit of a summary: 1. Brand new platform for EX platform, uses LGA2011 2. Four channels of SMI2 @ 2.66GT/s, which fans out to 8 channels of DDR3. The DDR3 has max bandwidth at 1.33GT/s, which is around 80-90GB/s peak theoretical. 3. 1.5TB/socket of memory, and will probably be upgradable to DDR4 in the future. 4. 3 QPI links per-socket, since I/O is now integrated 5. 32 lanes of integrated PCIe 3.0, which is strange as hell since IVB-EP has 40 lanes 6. Upto 15 cores, with upto 37.5MB of L3 cache 7. Cache slices are enabled independently of the cores, so that some SKUs are available with 6-10 cores and all the cache. These are targeted at databases with per-core licensing. 8. The power management is way better, so the actual frequency is quite higher than base frequency (whereas Westmere-EX didn't have good turbo at all). 9. System idle power is way way lower, nearly 100-150W less than the previous generation. Overall, it's a big improvement, probably around 2X and in some cases much more. [/quote<](Edit: added link)

    • Klimax
    • 9 years ago

    Didn’t even get much right… (It isn’t AES-NI you are looking for)

    There’s far more to Xeons E7 then you think. Primary hint: One of targets is POWER.
    Suggested reading:
    [url<]http://www.intel.com/content/dam/www/public/us/en/documents/datasheets/xeon-e7-v2-datasheet-vol-1.pdf[/url<] [url<]http://www.intel.com/content/dam/www/public/us/en/documents/datasheets/xeon-e7-v2-datasheet-vol-2.pdf[/url<] As for socket: Another variant of LGA2011, this one uses special link to memory buffer chips which then provide interface for RAM itself and there are other changes incompatible with other LGA2011 variants. Note: Memory buffers are used for signal integrity.

    • Klimax
    • 9 years ago

    And subsequently wondering why those licenses cost suddenly significantly more…
    (Although if one uses MS software like MS-SQL, then each AMD core is counted differently – as 0,75 of Intel’s. See [url<]http://download.microsoft.com/download/F/F/2/FF29F6CC-9C5E-4E6D-85C6-F8078B014E9F/Determining_SQL_Server_2012_Core_Licensing_Requirements_at_SA_Renewal.pdf)[/url<]

    • maxxcool
    • 9 years ago

    “SPU” was the literal name for the “off cores” .. 😉

    • maxxcool
    • 9 years ago

    Until silicon is replaced, 4-5ghz is the heat wall and power barrier. More threading is the “current” and only means of getting around that. Even with process reductions and power reductions frequency increases cost a ton of power to push those electrons around..

    • maxxcool
    • 9 years ago

    While many things are serial, The scheduler should be spreading out more tasks rather than waiting for core-0.. but then many tasks are low priority so that won’t happen. But the ultimate point is to the OP that he is not likely waiting for disk IO, but waiting on a Adobe filter, or May render or a game engine.. where-in my reply is more valid.

    • the
    • 9 years ago

    Cell had a slew of problems. It only had a 512 KB of L2 cache for its single PowerPC core. A bit on the low side even in 2005. Each SPE had 256 KB of local memory. This was not a cache as the programmer had to manage what its contents were directly. Similarly, that 256 KB of local memory had to contain both the running code and data for execution. There was no active coherency between the SPE units as that had to be managed manually. The SPE units had remote DMA functionality to get information from the XDR main memory without overhead to the main CPU but it could easily become bottlenecked with requests.

    The PowerPC core had a branch unit but was still an in-order design and would stall when a branch was encountered. The SPE units on the otherhand had no branching hardware and it would stall for upwards of 28 cycles. The SPE’s were purely vector units and vector hardware has to be used for scalar computations, leading to some inefficiencies.

    Essentially Cell was a fast chip but only if developers took the time to write hand coded assembler to avoid all of its performance nuances.

    • jihadjoe
    • 9 years ago

    Only because of idiots asking for the wrong stuff.

    When we get a consumer proc that prioritizes speed over cores, people come in and say “NOT ENOUGH COARZ!!”, and now we have a server proc where cores are more important, people come in and say “NOT ENOUGH GHz!!”. Drives me nuts…

    • Krogoth
    • 9 years ago

    Awesome server and workstation chip if you can afford it.

    Overkill for us mere mortals though.

    • UnfriendlyFire
    • 9 years ago

    From what I’ve heard, it was a pain to program for, and it had bottlenecks and insufficient caches for certain parts of the CPUs that screwed the programmers.

    • ronch
    • 9 years ago

    My, my, aren’t we in a bad mood today?

    • the
    • 9 years ago

    There are still design techniques that could increase IPC: micro-op caches, out-of-order retirement, more/wider speculative execution but they tend to have a significant power cost associated with them. Intel is currently operating on a 2% performance gain minimum for every 1% gain in power. Adding these features also increases the area of the CPU core, something CPU designers are keen to minimize.

    Even though the E7 v2 is a server CPU, the core designs stems from the mobile market. It is still Ivy Bridge that scales down to the 13W [url=http://ark.intel.com/products/64898/Intel-Core-i7-3667U-Processor-4M-Cache-up-to-3_20-GHz<]Core i7 3667U[/url<]. The server part simply scales the number of cores and cache sizes upward. With no expectation of actually being mobile, Intel scales up the clock speed too. Intel doesn't have a dedicated server x86 CPU design. (The IO portion of the CPU with its serial memory links, QPI links to other sockets and PCI-e controllers are distinctly server oriented.) With Intel letting the Itanium line wither and die, a pure x86 server core design could happen. Intel does have engineering resources to maintain several distinct core designs simultaneously. (Intel has had the Pentium 4, Core, and Itaniun all in development.)

    • ronch
    • 9 years ago

    Favorite car makers : Nissan, Mazda, Ford, Mitsubishi (for whatever reason), and maybe I wanna try a Suzuki Kizashi just out of curiosity (but i guess I can’t really take that risk).

    • jihadjoe
    • 9 years ago

    SERVER processors could always use more cores.

    You want 4GHz? The consumer space has plenty of those for far cheaper. You’re barking up the wrong tree here.

    • ronch
    • 9 years ago

    And you should have Krogoth sing you a song.

    • jihadjoe
    • 9 years ago

    Wrong thread…

    • ronch
    • 9 years ago

    People buy 16-core Opterons because they’re univaled, unlocked, and unbelievable. Oh wait, scratch the second one.

    • esterhasz
    • 9 years ago

    I find stagnating single core performance extremely depressing, not only because many things indeed cannot be parallelized, but also because many of those that can would require hefty investments to get there. Words like “crappy programs” or “lazy programmers” rarely explain anything because these are simply a question of investment and market dynamics. If a software company can expect to recoup the time lost in making their stuff parallel through more sales or a higher sales price they will probably do it.

    But continuously growing single thread performance was so awesome because it allowed us to be lazy in the sense of not making that investment and still get continuously more speed. I’m writing my own software for scientific data analysis and I simply don’t have the time and – admittedly – skill to write highly optimized and parallel code. Let’s say that I would need 200-300 hours to get really good at parallel programming (I wouldn’t know where to take them from, though) and then maybe add an additional 10% of time to development time for every program I write. Let’s say that I code around 25 hours per week, I’d add roughly 100 hours of additional work per year. At about $100 per hour, that’s a startup cost of 20-30K and 10K per year. That’s a lot of money.

    Previously I could still get a nice boost every year, but that’s over now. Fortunately I can still throw money at the storage part of the equation 🙂

    • chuckula
    • 9 years ago

    [quote<]physics simulation test[/quote<] Kaveri would get blown out of the water by a desktop CPU much less a huge multi-core monster like Ivytown. The reason for that is that any physics simulation that isn't a toy relies on double-precision computations. One of the dowsides of Kaveri is that AMD basically nuked double-precision capability in favor of single precision. My desktop haswell has about twice the double-precision capability just in the AVX units compared to the combined score for Kaveri including all the CPU and GPU cores running at theoretical maximum performance levels. That's not even counting the fact that Haswell's GT2 IGP... while certainly not as strong as Kaveri at playing games... actually has more than twice the theoretical double-precision performance of Kaveri's GPU. See Anand's article for more details: [url<]http://anandtech.com/show/7711/floating-point-peak-performance-of-kaveri-and-other-recent-amd-and-intel-chips[/url<]

    • Chrispy_
    • 9 years ago

    Kevin could struggle if he’s trying to use a workstation CPU or GPU in this instance.

    Kaveri’s IGP would win by default because these Xeon’s aren’t for workstations and they don’t have an IGP at all – what would be the point in having eight processors with intel HD graphics in them?

    IGP’s are for single-socket products only, which is a shame because Datacenter administrators could probably run Crysis on an 8-way HD4600 SLI (or whatever you call a multi-GPU solution for intel IGP’s).

    • blastdoor
    • 9 years ago

    Depressing because it’s true.

    • Scrotos
    • 9 years ago

    I dunno, would you go for the Cell? That had 1 main core and 6 or 7 sorta-cores. I see debate whether that can be called a multi-core CPU or if it’s 1 CPU and a bunch of specialized processing units.

    • Terra_Nocuus
    • 9 years ago

    Accounting authorized those AMD servers, because they were cheaper?

    • Scrotos
    • 9 years ago

    Power use, maybe?

    • Scrotos
    • 9 years ago

    Yup. Like he says, this is for a server which typically serves multiple processes/users. Per-thread performance has to be decent but being able to handle more threads is more important. This is probably a beast running VMWare.

    • ronch
    • 9 years ago

    You want a 4GHz processor? You can get one for just $200.

    • chuckula
    • 9 years ago

    I find your fractional-core processor intriguing and would like to subscribe to your newsletter.

    • blastdoor
    • 9 years ago

    I wish this were true but I just don’t see it. That is, I don’t see how those 16 core opterons are any good at all. I’d take an 8 core xeon, maybe even a 6 core xeon, over that opteron beast any day.

    But I would seriously like to know — why does anyone buy a 16 core opteron instead of an 8 core xeon? Are there cases where it really makes sense?

    • Grigory
    • 9 years ago

    They could make a CPU with 18 cores on a 3 x 6 floor plan, disable the worst 2 cores on each chip and rake in billions from the OCD crowd on additional charges over the 15 core variant.

    • Chrispy_
    • 9 years ago

    Hey, American cars are getting way better now.

    You do occasionally still use leaf springs and horrible quality plastics but they’re no longer a joke in terms of fuel economy, handling or build-quality.

    • UberGerbil
    • 9 years ago

    [url=http://money.cnn.com/2013/10/01/autos/bmw-brake-recall/<]Four months ago?[/url<] Toyota's recalls are massive because they sell so many cars. The entire production run of some BMW models would be considered a rounding error at Toyota. I lost count of how many times my friend's Porsche Cayenne got recalled but I never saw a news story about any of them. A 4Runner recall, however, is going to affect enough people it makes the nightly news.

    • UberGerbil
    • 9 years ago

    Many tasks are inherently serial. A llot of low thread programs aren’t “crappy”; they just are. Unless you know a secret way around Amdahl’s Law…

    • NeelyCam
    • 9 years ago

    Intel is synonymous* with efficient [b<]performance[/b<]. European cars represent that perfectly. And have you missed those massive Toyota recalls? When was the last time you heard of a BMW recall...?

    • NeelyCam
    • 9 years ago

    You need to be punished

    • maxxcool
    • 9 years ago

    Kev would lose horribly since this WS CPU would be paired with a WS GPU…

    • maxxcool
    • 9 years ago

    More than 4ghz has way too much heat to disapate with silicon. Worse the more dense you make the chip (smaller) the harder it is to cool. The issue is not core or speed its crappy low thread programs

    • UberGerbil
    • 9 years ago

    Power required (and thus heat dissipated) grows non-linearly with frequency; the precise point where things get ugly will vary according to the details of the fab process and transistor library involved, but it always happens [url=http://i272.photobucket.com/albums/jj163/idontcare_photo_bucket/Intel%20Core%20i7-2600K/i7-2600KPower-ConsumptionwithH100NT-H1.png<]somewhere around 5 GHz[/url<] (from [url=http://forums.anandtech.com/showthread.php?t=2195927<]this interesting thread[/url<] at AT), So that's where the frequency race effectively ended. You'd like to see more raw speed improvements. So would a multi-billion dollar industry full of very smart people. Unfortunately, physical reality -- given silicon transistors and our current knowledge -- disagrees. I'm underwhelmed by the lack of hoverboards, light sabers, and faster-than-light travel too. This is a server CPU, and some server tasks can make full use of the 240 threads (2 threads/core x 15 cores x 8 sockets) this enables -- assuming you can keep them fed from cache/memory.

    • ronch
    • 9 years ago

    I just Googled it up and linked to the first video that popped up. Sorry, I didn’t bother to watch it.

    • JdL
    • 9 years ago

    I’d like to see more raw speed, not just more cores. Sure this is more power-dense and all that, but what about actually speeding up the cores to 4 GHz or more? What ever happened to the GHz race? IMHO, multithreading peaked at 16 threads on a CPU. Time for some raw speed improvements.

    Underwhelmed.

    • UberGerbil
    • 9 years ago

    [url<]http://en.wikipedia.org/wiki/Xenon_(processor)[/url<]

    • Neutronbeam
    • 9 years ago

    No, no he won’t.

    • MadManOriginal
    • 9 years ago

    Intel fifteen core!

    p.s. Why did you link a Spanish subtitled version of that vdeio?

    • MadManOriginal
    • 9 years ago

    Considering Intel is synonmous with reliability (whether that’s right or wrong is another issue but ‘no one ever got fired for buying Intel’ is still a thing), I’d say that European cars are not what they should be shooting for.

    • NeelyCam
    • 9 years ago

    No; it won’t be accomplished until the decision is being made between a CPU and a real (=european) car

    • ronch
    • 9 years ago

    Hey, [url=http://www.youtube.com/watch?v=Uwz8Tsvh5oU<]sexy lady!![/url<]

    • ronch
    • 9 years ago

    Ok guys, amidst all the progress and amazing tech Intel has on offer these days, I say let’s not forget little ol’ AMD. Remember, AMD has roughly 10% the R&D budget of Intel and it’s still impressive how they can manage to keep themselves in the picture. Those 16-core Opterons probably aren’t gonna run circles around these 15-core Intel chips (more like the other way around) but they’re still pretty damn good, IMHO. For such a thrifty tech firm, AMD’s big core chips are still mighty impressive. Apart from IBM or perhaps Fujitsu, the rest either just license ARM cores or create custom ARM cores that are still way below the complexity of Bulldozer and its iterations. AMD has a legion of engineers capable of matching any other firm’s engineers out there.

    • MadManOriginal
    • 9 years ago

    Intel’s mission has been accomplished: moving from mere car-CPU analogies to making car-CPU an actual purchase decision.

    • ronch
    • 9 years ago

    I’m having trouble deciding between buying one of these 30-thread monsters or a Toyota Camry.

    • ronch
    • 9 years ago

    [u<][b<]THIRTY[/u<][/b<] threads!!! Even Krogoth will be impressed!

    • ronch
    • 9 years ago

    Yeah, I was lucky enough to unlock my X3 720 and run it stable enough at 3.3GHz (stock = 2.8) on my MSI 785GM-E65 with the stock cooler. A terrific deal considering I got it for less than a Core 2 Duo E8400 which I was originally eyeing back in the day.

    • Srsly_Bro
    • 9 years ago

    GO BACK TO BED WE DIDNT WAKE YOU YET!

    • UnfriendlyFire
    • 9 years ago

    Triple-core CPUs existed because AMD decided instead of turning a quad core into a dual core just because of one bad core, it would go with triple-core.

    And the triple-core could be unlocked into a quad-core, if you’re lucky.

    • Chrispy_
    • 9 years ago

    Physics simulation test? You mean like a test to make sure that the simulation accurately represents physics?

    I guess you could put the Xeon and the Kaveri on a table and then pour sand over them under a 1000fps camera. If we’re making bets I would say that Kaveri wins because it gets more sand grains stuck in the pins.

    • Deanjo
    • 9 years ago

    19 2/3 CORES OR GTFO!

    • MadManOriginal
    • 9 years ago

    Or in a 3P server, if you have 3 of these and add the 1+5 for each socket, it’s a 666 server!

    • MadManOriginal
    • 9 years ago

    Dats a lotta cores.

    /Italian stereotype voice

    • OneArmedScissor
    • 9 years ago

    Nice find. It’s like a sequel to the bizarre Westmere EP Xeon X5698, a dual-core that ran 4.4 GHz, but still had all 12MB of L3 and 3 memory channels:

    [url<]http://en.wikipedia.org/wiki/List_of_Intel_Xeon_microprocessors#.22Westmere-EP.22_.2832_nm.29_Efficient_Performance[/url<] Lo and behold, 6 core Ivy Bridge EX can still use 37.5MB of L3 and 8 memory channels: [url<]http://en.wikipedia.org/wiki/List_of_Intel_Xeon_microprocessors#.22Ivy_Bridge-EX.22_.2822_nm.29_Expandable_3[/url<]

    • ronch
    • 9 years ago

    It’s not the first time a CPU with an ‘odd’ number of cores came out. To be fair, triple-core AMD chips as well as 6-cores (to a lesser degree) seem odd too.

    • UnfriendlyFire
    • 9 years ago

    It’s a design compromise. Intel used a 5×3 config instead of 8×2 so latency doesn’t kill the CPU’s performance.

    • danny e.
    • 9 years ago

    15 cores seems like it might be the devils work.
    If you take 15 and add 666 and then subtract 666, you get 15!

    • smilingcrow
    • 9 years ago

    There’s a large range and they go well beyond 4P.

    • UberGerbil
    • 9 years ago

    If you look at the block diagrams and floorplan at [url=http://images.anandtech.com/doci/7753/Core.png<]Anandtech[/url<], it's actually a native 15 core organized as three groups of 5 cores, with three ring interconnects handling each of the three permutations of paired groups of 5 (which satisfyingly clever). It also appears Intel plans to produce "chopped" 10 and 6 core versions.

    • keltor
    • 9 years ago

    E7s are the 4P processors.

    • The_ENLIGHTENED_One
    • 9 years ago

    How would this bad boy fare against a Kaveri in a physics simulation test that makes use of Kaveri’s GPU?
    What’s the SP rating on the Xeon E7 v2 btw?

    • Farting Bob
    • 9 years ago

    15 cores seems like an odd number. Or are they all originally a 16 core chip with just 1 disabled that failed to meet Q&A standards?

    • Farting Bob
    • 9 years ago

    Nowhere near as much as 1 or 2 GPU’s that would cost 1/50th of the total price?

    • Beelzebubba9
    • 9 years ago

    You might already know this, but CPUs like this Xeon E7’s tend to be used in systems that require a lot more RAM than 2P-4P systems can typically provide.

    • brucethemoose
    • 9 years ago

    What kind of PPD would one get folding on 8 of these monsters?

    • Bauxite
    • 9 years ago

    Oh yeah, data protection technology…thats enterprise buzz-speak for the same AES-NI instructions they put in consumer products and 1-4P servers back in 2011.

    I am very curious about the “FCLGA2011” socket these E7 are listed as using, I wonder how close the pinout is to the current one (also listed as “FCLGA2011”). Sure would be fun to flash my bios and buy a cheap 15 core on ebay years from now.

    • Prestige Worldwide
    • 9 years ago

    ding ding ding!

    • Bauxite
    • 9 years ago

    You jest, but these have no meaningful advantage over E5 v2 for 2 and 4 socket servers except slightly more [i<]possible[/i<] cores per socket at quite a hefty price premium. Same 4 channel ddr3 ram config, same 3.0 x40 lanes per socket. (edit: according to [url=http://ark.intel.com/products/family/78584/Intel-Xeon-Processor-E7-v2-Family/server<]ark[/url<] its actually only 32) Not sure why they even bother having 2P and 4P limited SKUs, if you are core hungry its usually going to be cheaper and better to get more sockets.

    • LostCat
    • 9 years ago

    I think he might be joking about SSD prices. I dunno.

    • Srsly_Bro
    • 9 years ago

    You got it. Now we’re all going to be upset if you come back tomorrow and bother us before we wake you,

    • uni-mitation
    • 9 years ago

    Sexy lady!

    • Prestige Worldwide
    • 9 years ago

    Let me be the FIRST to yawn… wake me up when it’s under 50 cents / core

Pin It on Pinterest

Share This

Share this post with your friends!