Xeon E5-2600 v4 CPUs pave the way for the clouds of tomorrow

Intel released its Broadwell-EP Xeons today under the E5-2600 v4 umbrella. These chips include a number of features that virtualization-administrator readers may be interested in, including some fancy cache-management tools. The Broadwell-EP "tick" moves to a 14-nm fabrication process, and Intel has used this die shrink to increase core count while maintaining single-thread performance and holding the TDP line, according to Anandtech's review. Maximum supported DDR4 speeds also increase from 2133 MT/s to 2400 MT/s on some chips. The 22-core, 44-thread E5-2699 v4 is the king of this lineup, and it includes a massive 55MB of last-level cache. That's just one chip in the lineup, though.

Increasing core counts are amazing for virtualization density, but they aren't the sole limiting factor. In a virtualized environment, every resource is shared among guests, and the greater the granularity of control administrators have over those resources, the higher the potential for density. In modern VM environments, administrators have extensive control over most system resources, including processor frequency and core count, memory size, and disk I/O, among others.

With its Cache Monitoring and Allocation Technologies (CMT and CAT), Intel adds processor cache to the list of manageable resources. This system should prevent poorly-written or low-priority guests from taking more than their intended share of processor cache, allowing higher priority guests to make fewer memory calls and improve I/O performance.

CMT and CAT are part of Intel’s Resources Director Technology (RDT), which also adds memory bandwidth monitoring to its features today. The bulk of the new features serve as the foundation for fully-automated, software-defined infrastructures (or SDI). Intel says that the E5-2600 v4 chips are meant to allow businesses of all sizes to take advantage of SDI, not just large companies with huge public cloud infrastructures.

Broadwell-EP Xeons also get fully functional support for transactional memory using Intel's Transactional Synchronization Extensions, or TSX. Support for this feature was turned off in Haswell, Haswell-EP, and some Broadwell CPUs, thanks to an erratum that surfaced in August of 2014. Customers with Haswell-EP Xeons could use TSX for development purposes, but Intel only recommended that TSX be used on its Haswell-EX CPUs, which came to market as the Xeon E7 v3 family.

According to Anandtech, Broadwell-EP also brings some AVX performance improvements. On Haswell CPUs, mixing AVX and non-AVX workloads on the same CPU required all cores on the chip to reduce their maximum Turbo frequency to account for the increased power usage caused by AVX instructions. With Broadwell-EP, cores running AVX instructions will still clock down, but other cores will remain unaffected. Anandtech also notes that the PCLMULQDQ, or "carry-less multiplication" instruction now requires five cycles to complete rather than seven, and its throughput has been doubled. That's a boon for some cryptography-related tasks.

 

Comments closed
    • JMccovery
    • 4 years ago

    Good lord, just thinking of potential of a dual 2699 v4 system is close to madness; then, I think of the potential of a quad Broadwell EX system…

    • Krogoth
    • 4 years ago

    This also means that an i7-6xxx refresh (Socket 2011v3) is just around the corner.

    It is possible we may see the first non-Xeon 8-core SKU from Intel sold as top of the line.

    • chuckula
    • 4 years ago

    OK, this is hilarious. HotHardware managed to get its mitts on a two-socket Broadwell EP using one of the 18 core models (not even the 22 core jobs).

    They show it off in a Youtube video and then… wait for it.. run Cinebench. Hilarity ensues.

    [url<]https://www.youtube.com/watch?v=BzT6P6cVmAs[/url<]

      • Krogoth
      • 4 years ago

      Its clear that the problems with parallelism creeping up in that synthetic application and is much worse in real-world stuff.

        • Beahmont
        • 4 years ago

        Sure, but it still looks like it put up a score that was greater than 4 times the score of the next closest system. One of the systems listed was a 5960k I think. This rig has just a little more than 4 times the processor resources than the 5960k.

          • Krogoth
          • 4 years ago

          5960K doesn’t have new instruction sets or ECC support.

          It is an apples to oranges comparison at best.

      • JMccovery
      • 4 years ago

      Holy crap… That’s pretty much all I can muster on that.

    • GrimDanfango
    • 4 years ago

    As an indie CG artist, I was rather concerned a few years back investing the money in a handful of 16-core (2x 8-core) Sandy Bridge workstations – given the usual tendency for computer kit to become horribly obsolete and depreciate to nothing in a couple of years… but I’m rather glad that 3 generations on, my E5-2687w v1 chips are still holding their own perfectly well.

    Still surprises me how much things have failed to really move on in that time. The core counts certainly go up, but then the frequencies get chipped away seemingly to compensate. The fact that 3 generations on, the “workstation” chip is only really hanging somewhere around the 50% improvement mark over the Sandy Bridge parts seems a bit underwhelming.

    Glad I can wring at least a couple more years of value out of these beasts, but it’d be nice to see a decent leap forward *sometime* soon…

    Either that, or I guess CGI simulation and rendering as an industry will all start to finally move over to OpenCL/CUDA in search of extra performance.

      • Krogoth
      • 4 years ago

      The laws of physics and economics are at fault here. It is no longer easy to crank up the megahertz or transistor budget without turning the silicon into a blast furnace when loaded.

      The only viable way to increase computing prowess has been adding more cores. The problem with that is there are a limited number of applications that take advantage of those cores.

        • GrimDanfango
        • 4 years ago

        Yeah, I understand the reasoning… it’s just not as exciting as the old days 😛

        At least it forces us to work at wringing more performance out of what we’ve got instead of just relying on brute-force power.

        • Anonymous Coward
        • 4 years ago

        This is where “the cloud” gets neat. They [i<]can[/i<] find things for cores to do... no matter how many are per chip, per socket, per server. You'll share the CPU with an unknown number of others, each of you using how every many cores you have opted to rent. Did you know Amazon has a highest-bid market for otherwise unused cloud server capacity? By booting Linux to avoid licensing costs, a person can get their hands on lots of cores for [i<]cheap[/i<].

        • the
        • 4 years ago

        Indeed. Even with a 160W power budget, Intel hasn’t gone beyond 3.6 Ghz max for several generations of Xeon now. A low core count, >4.0 Ghz Xeon would be desirable for many users, especially in the context of software licensing costs. Programs are licensed per core or per socket but thankfully not by clock speed.

          • Krogoth
          • 4 years ago

          That’s a vanishing small market that isn’t worth bothering making SKUs for.

          The crowd who are still up on the megahurtz = better are using desktop-tier stuff a.k.a X and K series desktop chips or their rebranded “Xeon” counterparts.

            • the
            • 4 years ago

            I would disagree.

            Hardware costs pale in comparison to software licensing costs. At this juncture, doubling the number of cores in a system from say 8 to 16 increases performance by ~75% but it also doubles the software licensing costs. However, doubling clock speeds from 2.0 Ghz to 4.0 Ghz can yield similar if not >75% performance gains but keeps software licensing fees the same. That is where the demand stems from. It isn’t that they can’t get the same performance increase from core count but rather businesses don’t want speed as much on software licensing for the same performance.

            This market is also the niche where RAS matters to maintain uptime. They want the Xeon E5/E7 feature set over the desktop offerings but also want the high clock speeds to help lower software licensing costs.

            The big niche for high core count in servers are virtual machines that need the RAS and embarrassingly parallel HPC workloads. Databases are starting to swing toward needing more clock speed to reduce overall transaction latency as even the ‘small’ number of cores in a system today provides enough parallelism for numerous in-flight transactions.

    • flip-mode
    • 4 years ago

    MHz per dollar is kinda disappointing. $996 for a 3.5 GHz quad core….

    Cores per dollar aint bad though.

      • Krogoth
      • 4 years ago

      You miss an important item.

      They are dual-socket capability chips and that always carried a hefty premium. Dual-socket 2011v3 boards aren’t cheap either.

    • tsk
    • 4 years ago

    I’m gonna donate my lowly 5820k to charity and get some of this goodness!

      • Beahmont
      • 4 years ago

      I’m a charity case. I’m on a Core2 Quad. I can do puppy eyes and sit-up and beg for a 5820k.

      • chuckula
      • 4 years ago

      I vote you send it to the Starving-Orphans-but-Really-Chuckula Relief Fund!

      I know a guy there who will take it for free, all you have to do is pay the shipping!

    • blastdoor
    • 4 years ago

    Pretty awesome lineup.

    The 2687W looks particularly interesting to me.

    The only downside here is the sky-high prices. For the time being, though, Intel is the only game in town so they can charge whatever they want. I’m hoping AMD can offer something similar in performance to the 2687W for half the price, but perhaps more watts — maybe a 16 core zen with the same overall performance and maybe 180 watts?

      • jihadjoe
      • 4 years ago

      Two cynical and pessimistic points:

      1) Enterprise stuff will always be expensive.
      2) A competitive AMD will only mean AMD will charge higher prices for its own chips (as opposed to forcing Intel to lower theirs)

      I remember the Opteron 800 was $2650+ for two cores back when AMD was the undisputed CPU king.

        • f0d
        • 4 years ago

        [quote<]2) A competitive AMD will only mean AMD will charge higher prices for its own chips (as opposed to forcing Intel to lower theirs)[/quote<] people seem to forget how expensive the fast amd cpu's were when they were beating intel yes we do need competition from amd but that wont bring lower prices - if anything it will just mean amd will sell cpus for high prices again not enterprise cpus but still valid examples [url<]https://techreport.com/review/8295/amd-athlon-64-x2-processors[/url<] Athlon 64 X2 4200+ 2.2GHz 512KB $537 Athlon 64 X2 4400+ 2.2GHz 1024KB $581 Athlon 64 X2 4600+ 2.4GHz 512KB $803 Athlon 64 X2 4800+ 2.4GHz 1024KB $1001 AMD Athlon 750 MHz [url<]http://www.cpu-world.com/CPUs/K7/AMD-Athlon%20750%20-%20AMD-K7750MTR52B%20A.html[/url<] Price at introduction $799 AMD Athlon 1 GHz [url<]http://www.cpu-world.com/CPUs/K7/AMD-Athlon%201000%20-%20AMD-K7100MNR53B%20A.html[/url<] Price at introduction $1299 AMD Athlon XP 3000+ [url<]http://www.cpu-world.com/CPUs/K7/AMD-Athlon%20XP%203000%2B%20-%20AXDA3000DKV4D.html[/url<] Price at introduction $588

          • ronch
          • 4 years ago

          Yup. So true. When did CPU prices *ever* come down because AMD was competitive? AMD CPUs become affordable only when they’re behind. K6-2, late Athlon XP period, late K8 days, K10 days (prices didn’t exactly tumble but AMD did offer better perf/$$$), and when Vishera came out (AMD tried to price high when FX first came out, and also with their insanely conceived ‘Centurion’ Vishera SKUs… they WILL try to price high if they could).

          So expect Zen to be out of reach if it turns out to be competitive. And then people who don’t want to spend $350+ on a CPU will just settle for an FX-8350 or FX-6300. Come to think of it, maybe that’s why Gigabyte etc. are coming out with ‘new’ AM3+ boards.

          Edit – Zen’s higher end SKUs will be out of reach for most folks but AMD will surely fill in all price points with it where they can.

            • Anonymous Coward
            • 4 years ago

            You’re being ridiculous. The mere existence of options gives pressure to offer a better product and/or better price, and AMD will certainly offer Zen at all price points, as soon as they are able.

            • ronch
            • 4 years ago

            Yes Zen will most likely span all price points from Celeron-class to Core i7-class and Xeon-class. Point is, prices will NOT bottom out because of a strong AMD. Zen will only allow AMD to play upmarket and charge prices similar to how Intel asks for their i7s and Xeons (and of course, downmarket where they’ve always done so). As an example, during the K8 days AMD DID have products spanning all the way from cheap Semprons to Athlon FXs to Opterons. Prices were high wherever AMD could price high. CPUs didn’t become cheaper. AMD didn’t give you the same performance for half the price. Maybe 10-20% cheaper for the same performance was historically more like it. I got my Phenom II X3 720 for a bit less than a Core 2 Duo E8400 back in Dec. 2009 and the 4th core could be unlocked with no glitches. That was a fantastic deal. 20% higher aggregate stock performance at a lower price. But you don’t find that kind of perf/$$$ across the board.

            Do not mistake the pace of innovation in the x86 space solely due to competition either. The increase in performance, while partly due to competition, was also because until 2008 or so, there were still a lot of low hanging fruits to be picked. Not so today. Don’t expect performance to take leaps and bounds again if Zen turns out able to match Intel (which is a big IF). Intel knows this and that’s why they’ve officially said Tick-Tock is over. Does AMD know something Intel doesn’t?

            • Anonymous Coward
            • 4 years ago

            [quote<]CPUs didn't become cheaper. AMD didn't give you the same performance for half the price. Maybe 10-20% cheaper for the same performance was historically more like it.[/quote<] A situation which has not yet occurred, but is not impossible, would be for (1) AMD to offer a competitive product, while (2) AMD can manufacture in huge volumes. If they can do those things at the same time, they will be able to flood the market and drive prices down while still increasing their earnings. If they can not do both of those things, prices will remain high. If they can do #1 but not #2, they can have a good profit. If they can do #2 but not #1, they will struggle. They have never been able to do [b<]both[/b<] of those things before, so prices have never had any sane reason to do down. [quote<]Do not mistake the pace of innovation in the x86 space solely due to competition either.[/quote<] Yes I am aware of the physical limitations that CPU manufacture is up against. I don't expect you to pay attention to what I have been saying before, but certainly I have many times predicted on this forum that Zen will not beat Intel.

            • ronch
            • 4 years ago

            I don’t think anyone but the most fervent and delusional AMD fanbois expect Zen to outperform Intel. Maybe they could and that’s a nice surprise, but the chances are pretty slim.

            • w76
            • 4 years ago

            Ya’ll don’t get the whole “economics” thing. Competition does lead to lower prices, but that competition has to be meaningful as well. It’s well known AMD was capacity constrained when it had the performance crown. Markets are about supply and demand, so it with lack of supply it wouldn’t make logical sense to engage in a price war. If they lowered prices too much, they’d simply sell out at a lower price level than what they otherwise could’ve moved their limited supply at.

            Likewise, if Uber was just 1 dude with a cellphone, Uber wouldn’t be disrupting the entire legacy cab market in major cities, even if Uber was free. That’s an extreme example, but the relevant point is changes in quantity demanded along demand curves and AMD’s own supply curve at the time.

            In markets with meaningful competition, you do see prices fall. Even when options are limited: thanks to the competition between trucking and rail, I can buy inventory from Georgia (by truck) and California (by rail) at essentially the same freight cost. Rail wouldn’t pass on that savings inherent to its technology if they weren’t forced to by the market.

            • Anonymous Coward
            • 4 years ago

            I love those up/down thumbs.

            • blastdoor
            • 4 years ago

            The key is what would have happened in the absence of competition.

            Thinking back to the 2000-2006 era, Intel’s plan was to push Rambus plus Netburst for the desktop (IA-32) and Itanium (IA-64) for server/workstation.

            AMD totally disrupted that plan. AMD forced intel to extent x86 to 64 bits, they forced Intel to go dual core with x86, and they forced Intel to drop Rambus.

            The effects of that competition were huge and very positive for everyone who buys CPUs.

            • Anonymous Coward
            • 4 years ago

            Yeah AMD did a good job there, also generally in offering an alternative to those mis-guided P4’s.

            • Anonymous Coward
            • 4 years ago

            Also lets not forget that AMD basically killed Itanium and that whole mess Intel was trying to get everyone stuck in.

            • Anonymous Coward
            • 4 years ago

            The chances are zero.

          • blastdoor
          • 4 years ago

          The fact that you always have to pay a premium for the best product doesn’t mean that there is no value to competition.

          Back then, the Athlon wasn’t just competitive, it was superior. As a result, Intel had to cut prices. If AMD hadn’t been offering a superior product, Intel would have charged much higher prices for their second-rate product (because it wouldn’t have been a second-rate product — it would have been the only game in town).

          Suppose Zen comes out and is a big improvement over AMD’s current lineup, but still lags Intel (I think this is the most likely scenario). That means AMD will have to charge less than Intel in order to sell the product. AMD might charge more than they are now, but it will have to still be less than Intel. And it will have to be a price that represents a better price/performance deal, at least for some markets, than Intel. This is unambiguously good for people who buy CPUs.

            • f0d
            • 4 years ago

            the thing is it diddnt make intel lower their prices at all

            they still had $500 >$1k cpu’s even though they were slower and they had prices all over the price range like they do now also

            the top end p4 was $1k and there was cpu’s all the way down to $100 like they do now, if anything intel cpu’s are cheaper now than when they had a slower product that amd

            the only thing that happened was that amd were able to charge high prices for their cpu’s instead of the budget prices they were selling their k6-2’s at

            athlon introduced 1999

            $800 intel 2000
            [url<]http://www.cpu-world.com/CPUs/Pentium_4/Intel-Pentium%204%201.5%20GHz%20-%2080528PC021G0K%20%28BX80528JK150G%29.html[/url<] $600 intel 2000 [url<]http://www.cpu-world.com/CPUs/Pentium_4/Intel-Pentium%204%201.4%20GHz%20-%2080528PC017G0K%20%28BX80528JK140G%29.html[/url<] $350 intel 2000 [url<]http://www.cpu-world.com/CPUs/Pentium_4/Intel-Pentium%204%201.7%20GHz%20-%20RN80528PC029G0K%20%28BX80528JK170G%29.html[/url<] $1k intel 2003 [url<]http://www.cpu-world.com/CPUs/Pentium_4/Intel-Pentium%204%20Extreme%20Edition%203.2%20GHz%20-%20RK80532PG0882M%20%28BX80532PG3200F%29.html[/url<] $500 2002 [url<]http://www.cpu-world.com/CPUs/Pentium_4/Intel-Pentium%204%202.2%20GHz%20-%20RK80532PC049512%20%28BX80532PC2200D%29.html[/url<] $350 2002 [url<]http://www.cpu-world.com/CPUs/Pentium_4/Intel-Pentium%204%202%20GHz%20-%20RK80532PC041512%20%28BX80532PC2000D%29.html[/url<] $1k intel 2005 [url<]http://www.cpu-world.com/CPUs/Pentium_4/Intel-Pentium%204%20Extreme%20Edition%203.73%20GHz%20-%20JM80547PH1092MM%20%28BX80547PH3730F%29.html[/url<] $800 intel 2005 [url<]http://www.cpu-world.com/CPUs/Pentium_4/Intel-Pentium%204%20670%203.8%20GHz%20-%20JM80547PG1122MM%20-%20HH80547PG1122MM%20%28BX80547PG3800F%29.html[/url<] $600 intel 2005 [url<]http://www.cpu-world.com/CPUs/Pentium_4/Intel-Pentium%204%20672%203.8%20GHz%20-%20HH80547PG1122MH.html[/url<] $400 intel 2005 [url<]http://www.cpu-world.com/CPUs/Pentium_4/Intel-Pentium%204%20662%203.6%20GHz%20-%20HH80547PG1042MH.html[/url<]

            • blastdoor
            • 4 years ago

            This is more consistent with my memory:

            [url<]http://www.extremetech.com/extreme/75294-intels-pentium-d-price-half-that-of-amds-x2[/url<]

      • DrCR
      • 4 years ago

      2687W, wow, that’s alarmingly high wattage. That’s more than an order of magnitude above AMD’s offerings.

      [sub<]Edit for clarity[/sub<]

        • Krogoth
        • 4 years ago

        You are cramming 12-cores at 3.0Ghz on a Socket 2011 package. The 14nm process is designed with low-power in mind not super-high clockspeeds. Just take a look at higher-core SKUs. They are ultra-conservative with their clockspeeds in order to keep TDP somewhat sane.

        It reminds me of 6-core Gulftowns back in their heyday. They were hot chips when loaded but blew the competition out of the water. AMD only had Istanbul-based Phenoms/Opterons at the time with Bulldozer on the horizon.

      • ronch
      • 4 years ago

      Same performance at half the price? Poor AMD just can’t catch a break now, could they? But that won’t happen. For example, even today, a typical AMD CPU that delivers half the performance of a typical Intel CPU is priced at half what the Intel CPU sells for. So AMD holds on to their price/$$$ relative to Intel. Bulldozer and it’s lineage are so far behind that AMD simply can’t price any lower so they can’t offer better overall performance/$$$. They can probably offer better perf/$$$ if Zen turns out able to command higher prices but we’ll see.

      So half the price for similar performance is quite unlikely unless AMD wants to start a bloody price war it can’t win. I’d settle for 10-15% cheaper. Intel will keep their prices up and let AMD coast along a few dollars below just like they always have prior to Bulldozer. Let’s keep AMD alive.

    • chuckula
    • 4 years ago

    Interestly tidbits from Anand’s article:
    1. The full-size 22 core version has a die size of about 455 mm^2.

    2. The cut-down die size that supports up to 15 cores has a die size of 306 mm^2. For those of you who remember back a few years, our pals Bulldozer & Piledriver each come on 315 mm^2 dies.

      • tipoo
      • 4 years ago

      455 mm^2 on the worlds most advanced 14nm fab…Nerd boner. One wonders what a GPU could do with that.

        • chuckula
        • 4 years ago

        Quite a bit considering that GPUs generally use denser metallization layers & transistor layouts since they don’t have the same speed-path requirements of high-end CPUs.

        • biffzinker
        • 4 years ago

        So tipoo, 22 Cores/44Threads, DDR4 2400 tied to quad memory controllers pushing 77 GB/s of bandwidth, and 55 MB L3 Cache doesn’t give you a…Nerd boner?

          • tipoo
          • 4 years ago

          Well yeah, but I can’t feasibly use that 😛

          • the
          • 4 years ago

          It doesn’t for me. I’ve been spoiled by the 72 core, six channel DDR-2400 plus HMC chip in the next Xeon Phi.

          Also SkyLake-EP looks to be a solid jump over Haswell–EP/Broadwell-EP next year. 3D Xpoint based DIMMs are going to be very, very nice.

        • USAFTW
        • 4 years ago

        That’s interesting since the largest Haswell-E 22nm part was 646 mm^2. But even so, getting a 455 die to yield on 14nm is unprecedented and probably took some alien-level ingenuity to do.

          • Waco
          • 4 years ago

          There’s a reason it costs $4k in bulk. 🙂

        • Anonymous Coward
        • 4 years ago

        Meh, one huge graphics card against another. The game you play will be the same…

        • ImSpartacus
        • 4 years ago

        Probably not as much as you’d think.

        I expect that Intel optimizes its process for the needs of a cpu, not a gpu (i.e. frequency vs density).

        I remember that was a discussion point for amd apus since either the cpu portion or gpu portion compromised as they must share a process that’s inherently imperfect for at least one of them. I can’t recall the source. Probably a llano or trinity review.

      • Srsly_Bro
      • 4 years ago

      They weren’t at 14 nm. Perspective, bro. 🙂

        • chuckula
        • 4 years ago

        Part of AMD’s problem is that Zen won’t be at 14nm (at least not Intel’s version of 14nm) either.

        I’m laying out some perspective on physical parameters that both Intel & AMD have to work with over the next year or so at least until 10nm arrives in the later part of 2017.

          • tsk
          • 4 years ago

          It will arrive at the very end of Q4 2017, in the form of a Core M part, Is my educated guess.

            • the
            • 4 years ago

            Intel never produces large dies aimed for servers on a new process node. The Core M and various mobile parts are ideal for a new process node in many ways.

          • brucethemoose
          • 4 years ago

          TSMC 16nm is as close as AMD has been in a long time. Inferior, sure, but it’s not a full node behind Intel (or worse) like it has been in the past.

          Also, seeing how hard physics has pushed every node back, I doubt 10nm will be on schedule.

          • ImSpartacus
          • 4 years ago

          I honestly haven’t followed zen. Is it rumored to not use a modern finfet process?

        • JumpingJack
        • 4 years ago

        That is kinda the point.

      • Kougar
      • 4 years ago

      As if 22 cores at 455mm2 wasn’t mouth-watering enough already… The full-size version is actually 24 cores, Intel simply doesn’t ship any models with all of them enabled yet.

        • the
        • 4 years ago

        Those are probably reserved for Broadwell-EX. The 22 core EP should be the same die as the 24 EX die based upon previous Intel segmentation.

      • the
      • 4 years ago

      455 mm^2 is smaller than the 18 core Haswell-EP/EX which came in at 662 mm^2. Intel has plenty of room to grow for SkyLake-EP.

    • NTMBK
    • 4 years ago

    Yawwwn. Wake me up when you can buy Knight’s Landing.

      • Waco
      • 4 years ago

      They’re neat for HPC…but why would [i<]you[/i<] want one? They won't be cheap, they won't use standard boards, and they won't be good for most user (even power user) workloads.

        • NTMBK
        • 4 years ago

        Not everyone is just interested in this stuff for gaming, you know 🙂 Don’t worry, I can find good use for it.

          • biffzinker
          • 4 years ago

          What use for it, I’m curious.

            • NTMBK
            • 4 years ago

            Work stuff that I can’t really discuss 🙂 Lots of floating point, lots of memory bandwidth. We use GPUs at the moment, but I want to experiment with Phi.

            • Waco
            • 4 years ago

            I didn’t mean to imply you thought you’d game on it – I was wondering what kind of workloads you might have.

            I generally only deal with simulations, so I understand their use there (our next machine is going to have almost 10,000 of them).

            Are you able to disclose what type of application you’d be running?

            • tipoo
            • 4 years ago

            Found the FBI iPhone cracker!

            • the
            • 4 years ago

            While Xeon Phi has been targeted toward HPC, I’ve been curious to see how they’d work as nodes for mundane tasks like web servers. As long as a logical core can provide the minimum latency/throughput necessary, the raw number of cores should provide decent performance through parallel requests.

            While I wouldn’t expect great performance, I am curious how Xeon Phi would perform on mundane DB workloads. With 16 GB of HMC on the new Xeon Phis, I wonder how an in-memory DB would perform given its large bandwidth.

        • Beahmont
        • 4 years ago

        Because of the Siren Call of “MOARRR KORES!”

        72 cores (Even modified Airmont Cores) with HT won’t be enough of course. It never is. But c’est la vi.

          • Waco
          • 4 years ago

          More like…MOAR memory bandwidth!

        • Klimax
        • 4 years ago

        There is some work on BOINC to support it and it could be very interesting for some video tasks. (Would beat rewriting DX 9-based GPU FFT filtering)

        ETA: And some other fun stuff…

        • the
        • 4 years ago

        Ego.

Pin It on Pinterest

Share This