Forgot your password?
typodupeerror
Intel Hardware

Intel Announced 8-Core CPUs And Iris Pro Graphics for Desktop Chips 173

Posted by Unknown Lamer
from the face-meltingly-fast dept.
MojoKid (1002251) writes "Intel used the backdrop of the Game Developers Conference in San Francisco to make a handful of interesting announcements that run the gamut from low-power technologies to ultra-high-end desktop chips. In addition to outing a number of upcoming processors—from an Anniversary Edition Pentium to a monster 8-core Haswell-E — Intel also announced a new technology dubbed Ready Mode. Intel's Ready Mode essentially allows a 4th Gen Core processor to enter a low C7 power state, while the OS and other system components remain connected and ready for action. Intel demoed the technology, and along with compatible third party applications and utilities, showed how Ready Mode can allow a mobile device to automatically sync to a PC to download and store photos. The PC could also remain in a low power state and stream media, server up files remotely, or receive VOIP calls. Also, in a move that's sure to get enthusiasts excited, Intel revealed details regarding Haswell-E. Similar to Ivy Bridge-E and Sandy Bridge-E, Haswell-E is the 'extreme' variant of the company's Haswell microarchitecture. Haswell-E Core i7-based processors will be outfitted with up to eight processor cores, which will remain largely unchanged from current Haswell-based chips. However, the new CPU will connect to high-speed DDR4 memory and will be paired to the upcoming Intel X99 chipset. Other details were scarce, but you can bet that Haswell-E will be Intel's fastest desktop processor to date when it arrives sometime in the second half of 2014. Intel also gave a quick nod to their upcoming 14nm Broadwell CPU architecture, a follow-on to Haswell. Broadwell will be the first Intel desktop processor to feature integrated Iris Pro Graphics and will also be compatible with Intel Series 9 chipsets."
This discussion has been archived. No new comments can be posted.

Intel Announced 8-Core CPUs And Iris Pro Graphics for Desktop Chips

Comments Filter:
  • 8 cores? (Score:3, Insightful)

    by chris200x9 (2591231) on Thursday March 20, 2014 @05:54PM (#46538615)
    So they finally caught up to AMD.
    • Re: (Score:2, Insightful)

      by Anonymous Coward

      No, they're well ahead of AMD in this regard. AMD's 8 "core" CPUs are actually 4 core CPUs that can process 2 integer instructions at the same time on one core. Much like Intel's current i7s are 4 core CPUs that can process an integer and a floating point instruction at the same time on one core. Basically, AMD is marketing hyper threading as being more cores.

      • Re:8 cores? (Score:4, Informative)

        by Travis Mansbridge (830557) on Thursday March 20, 2014 @06:17PM (#46538799)
        While each 2 cores on the AMDs share resources, this is different from hyperthreading, and there are indeed 8 cores. http://www.reddit.com/r/builda... [reddit.com]
      • Uhhh...no. About the only thing not shared these days in (BD-derived) AMD cores is the FPU.
        • Re:8 cores? (Score:4, Informative)

          by guacamole (24270) on Thursday March 20, 2014 @06:31PM (#46538935)

          I believe cache is shared, and is believed to be one of the bottlenecks of the current AMD CPUs.

          • The data caches are not shared. Each core has a separate data cache. The decoder is the same so they share the instruction cache. But AMD's instruction caches are 64 KB while Intel uses 32 KB sized instruction caches.

          • by Kjella (173770)

            I believe cache is shared, and is believed to be one of the bottlenecks of the current AMD CPUs.

            By far not the most significant one though, in single threaded tests the i7-4770K beats the FX-8350 by 62% in Cinebench R11.5, 73% in Cinebench R10 and 47% in POV-Ray 3.7RC6 and that's when the AMD core is not competing for resources with its sibling. With turbo the picture is a bit more complex than that but 4 Intel cores already equals 6-7 AMD cores. Then you add in cache contention, shared FPU, overhead of more threads for the last 1-2 cores of difference as in the most ideal benchmarks for AMD they're r

      • No, they're well ahead of AMD in this regard. AMD's 8 "core" CPUs are actually 4 core CPUs that can process 2 integer instructions at the same time on one core. Much like Intel's current i7s are 4 core CPUs that can process an integer and a floating point instruction at the same time on one core. Basically, AMD is marketing hyper threading as being more cores.

        What you describe is superscalar execution, and was the point of the original Pentium. That's Instruction-Level Parallelism not Thread-Level Parallelism. Also the Pentium Pro/Pentium 2 had three FPUs.

        It's lame that this comment is modded insightful, you're making shit up.

      • The opterons are real 16 cores that can do 4 way (4 sockets for 64 cores). However Intel have real 10 core Xeons that may as well be 20 for floating point tasks, and there are 4 socket machines that can take them (and an IBM thing that is effectively 8 sockets). The price difference between those two top ends is utterly enormous but the Intel machines are supposed to perform significantly better - whether it's worth getting one of those or four of the AMD ones for the same price depends on what you want i
    • by thesandbender (911391) on Thursday March 20, 2014 @07:03PM (#46539207)

      AMD's Bulldozer cores have Clustered Integer Core [wikipedia.org] which has two true ALU "cores" and one shared FPU. For integer instructions this is two true cores and not "hyper-threading". For FP instructions this is "hyper-threading" and why Intel has been regularly handing AMD it's arse in all benchmarks that aren't strictly ALU dependent (gaming, rendering, etc). AMD's FPU implementation, clock for clock, is a bit weaker on most instructions as well. And yes, the FPU _is_ shared on AMD processors.

      EMT64 is not "32 bits on each 1/2 of the clock cycle". That doesn't even make any sense. EMT64 is true 64 bit. x86-64 does have 32 bit addressing modes when running on non-64bit operating systems. This is part of the x86-64 standard and hits AMD, Intel and VIA.

      Hardware Queuing Support is part of the Heterogeneous System Architecture [wikipedia.org] open standard and won't even be supported in hardware until the Carizzo APU in 2015. Since this is an open standard, Intel can chose to use it.

      Both architectures have shared caches.

      WTF does nVidia's IEE-754 compliance have to do with Intel vs AMD?

      I'm not an Intel or AMD fanboy, I try to use the right one for the job. I prefer AMD for certain work loads like web servers, file servers, etc because they have the most integer-bang for the buck. If I'm doing anything that involves FP, I'm going to use an Intel Chip. Best graphics solution?... yeah, I'm not even going to go down that hole.

      • Hardware Queuing Support is part of the Heterogeneous System Architecture [wikipedia.org] open standard and won't even be supported in hardware until the Carizzo APU in 2015. Since this is an open standard, Intel can chose to use it.

        The first is not a correction of something that was "wrong in this thread" (if I was wrong in the first place - there *is* already HW for it in Kaveri, even though the implementation may change in the future) , and the second is an opinion (I really don't think that Intel will follow suit any time soon on that).

        WTF does nVidia's IEE-754 compliance have to do with Intel vs AMD?

        Well, AMD apparently takes care for the execution units to be completely interchangeable, so that code could executed on one core or the other as necessary with identical results, which is one of the

  • The non extreme / severs ones are very limited on PCI-e and even in systems like the MAC pro the pci-e limits / DMI hold it back.

    The mac pro should of had 2 SSD's but due to limits it only has one.

    • It has a pci express which is several multitudes faster as it is directly on the PCI bus. It is rated for over 700 megs a second.

      • number of lanes is to low

        • Is that really true compared to an ata port? It handles graphics cards just fine. Perhaps someone with more knowledge can eleborate on this?

          These cards are just coming out for PC's too and man they are expensive but can promise bandwidth about 1 gig/sec. I think in a few years when this is the norm the mechanical disk will finally die. AHCI will seem slow in comparison.

    • The extreme platform as you call it has had relatively affordable quad cores (i7 3830, i7 4820K) and I guess there will be a similar quad core Haswell-E for sale.
      You can go that way if you want a workstation with crap tons of RAM, I/O and PCIe slots.

  • by Stormy Dragon (800799) on Thursday March 20, 2014 @06:14PM (#46538775) Homepage

    Other details were scarce, but you can bet that Haswell-E will be Intel's fastest desktop processor to date when it arrives sometime in the second half of 2014. Intel also gave a quick nod to their upcoming 14nm Broadwell CPU architecture, a follow-on to Haswell.

    Does anyone else find it kind of weird that Intel seems to have gotten into a pattern where their supposed top of the line CPUs are perpetually a generation behind their supposed commodity CPUs in terms of technology?

    • The desktop/laptop processors are easy...single socket, relatively small number of cores.

      It takes effort to add the bits to allow the processors to scale to 10/12 cores, huge caches, and multiple sockets. They also use more complicated memory modules, different motherboards, etc.

      Also, large companies are able to get their hands on limited quantities of these cpus well before they're generally available for large-scale ordering to allow their engineers to build products on them and test how they'll behave.

      • I would argue it the other way, that achieving high single thread performance is very complicated and requires both more design work as well as better understanding of the given process than is usually available for the initial launch where they're using projected and calculated si characteristics. A year later they have some experience with massive volume production, know to many decimal places what their yield will be, as well as have more time to do custom circuits that high freq CPUs will require.

        Adding

    • by Amtrak (2430376) on Thursday March 20, 2014 @06:52PM (#46539143)
      This is because these chips are meant for the Server and Workstation market where stability and longevity is more important than bleeding edge tech. As long as they stay the fastest chips you can buy who cares if they are a process node behind. Not the businesses actually buying them. If you want a "Kickass" gaming machine save your money and don't buy an E series Intel.
    • by rsborg (111459)

      Other details were scarce, but you can bet that Haswell-E will be Intel's fastest desktop processor to date when it arrives sometime in the second half of 2014. Intel also gave a quick nod to their upcoming 14nm Broadwell CPU architecture, a follow-on to Haswell.

      Does anyone else find it kind of weird that Intel seems to have gotten into a pattern where their supposed top of the line CPUs are perpetually a generation behind their supposed commodity CPUs in terms of technology?

      Not at all - the commodity CPU customers can do beta test for the more risk-averse enterprise server CPU customers.

    • by Kjella (173770)

      Does anyone else find it kind of weird that Intel seems to have gotten into a pattern where their supposed top of the line CPUs are perpetually a generation behind their supposed commodity CPUs in terms of technology?

      They're not really consumer CPUs, they're a spin-off of Intel's server/workstation CPUs for the enterprise. That market requires a lot of validation and is generally very conservative preferring tested and true technology so it's not unnatural for server chips to lag behind consumer chips by a generation and so the "enthusiast" processors aren't ready until the Xeons are. My guess is that most of them are "damaged goods", server CPUs with ECC, QPI, vPro, TXT or other essential server features broken, but if

    • It makes sense for a couple of reasons

      1: Intel desperately want to stop the portable computing market moving away from laptops and laptop-like tablets towards smartphone-like tablets. To do that they need to get the most power efficient technology possible into ultrabooks and ultrabook-like tablets.
      2: Making a design work properly with 2-4 cores on one chip for laptops and mainstream desktops is a lot simpler than making it work properly with 8+ cores and inter-chip links for a server part (and the high end

    • Because

      #1. For CPU heavy loads you probably have more than one CPU per board.

      #2. Most people don't use their 1U Rack-Mount Servers to play Crysis and TitanFall, they just need to handle a crap-ton of threads/ram/drives. Therefore having the latest built-in GPU features does nothing useful.

      #3. Stability > Core Speed
    • The design-side motivation is to alternate architectural changes with process shrinks so that you're not trying to debug both at the same time. Prescott tried that, and look how that turned out.

      The marketing motivation is that the buyer of the commodity part is more price sensitive and the buyer of the performance part is more feature sensitive. You use the shrunk process for commodity parts first due to the increased die per wafer, which give you both greater volume and lower cost per die so that you can s

    • by bloodhawk (813939)
      nothing weird here at all, commodity vs reliability. Stable, tested proven chips generally stay one step behind. consumer commodity chips give them a chance to weed out any problems without placing risks on chip lines that simply MUST work.
    • by mikael (484)

      Commodity CPU's can be advanced in small steps every six months, while enterprise CPU's can be advanced in large steps every few years.

  • by LoRdTAW (99712) on Thursday March 20, 2014 @06:17PM (#46538797)

    Finally! I have been waiting for next gen Iris graphics [computerhistory.org] since like forever!

  • We've been limping along with ~10% performance increases per chip generation since forever.
    • by Ken_g6 (775014)

      You asked for it, you got it! [anandtech.com] Though the downside is these two fast cores don't include AVX, AVX2, or a few other instruction sets.

    • by Salgat (1098063)
      The only reason you saw phenomenal speed increases on single cores in the past was because we were no where near the frequency barrier. Going from 200 to 400MHz was extremely easy compared to 4GHz to 8GHz, which isn't even possible except in exotic conditions.
      • This, but AMD with Athlon helped a lot (at 33Mhz a year, we might have 2Ghz CPUs now), Otherwise Intel wouldn't have had have any incentive to push clock speeds that fast. If AMD were kicking them again, I'm pretty sure those exotic conditions wouldn't be such a barrier anymore.

    • by Sloppy (14984)

      Ok, but aside from the n% increase over the n% increase over the n% increase over the n% increase, what has Intel done for us?

      Intel makes the 2 fast core processor right now, today, and it'll cost you a staggering $120 to $150. It's called the Haswell Core i3 and each of its cores is faster than any of the cores in your $5000 machine from 2007. It will run Dwarf Fortress faster than anyone would have imagined back then.

      Of course there's no limit to what you'd like, but if you have a problem with the amazi

  • Intel Announced 8-Core CPUs And Iris Pro Graphics for Desktop Chips

    Okay, I know that strictly speaking it did happen in the past, but that's not how headlines are usually written.

  • by sshir (623215)
    Am I is the only one concerned about amount of RAM?

    I mean, everybody is so excited about DDR4... But do people understand that instead of 8 dimm slots we'll get only 4 (1 dimm per channel instead of 2-3)? So while keeping costs on this side of reasonable, we're getting only half the amount of memory?

    WTF?!!
    • I mean, everybody is so excited about DDR4... But do people understand that instead of 8 dimm slots we'll get only 4

      No...not everyone. Going from DDR2 to DDR3 netted fractional gains in real world applications and indications are that the same will be true going from DDR3 to DDR4.

      Also plenty of consumer level boards only have 4 DIMM slots now. Which has always been plenty for most people, ever since we moved up from DDR1 boards and their crappy 2GB limit per stick.

      • by Rockoon (1252108)

        No...not everyone. Going from DDR2 to DDR3 netted fractional gains in real world applications and indications are that the same will be true going from DDR3 to DDR4.

        To put a bullseye on this, its because latencies havent really changed. Its a rare workload that isnt either CPU limited or RAM latency limited, rather than RAM bandwidth limited. DDR4 isnt going to change that.

    • You'll have 16GB unbuffered DIMMs so you aren't losing anything. With Haswell-EP using LR-DIMMs allows 3 per channel for 768GB per CPU.
    • So while keeping costs on this side of reasonable, we're getting only half the amount of memory?

      I suspect it will be a pain when the platform first comes out but in time 16GB desktop DDR4 modules will become affordable while I doubt 16GB desktop DDR3 modules ever will (if the boards even support them)

      • Fun story, AMD supports 16GB unregistered DDR3 DIMMs, but Intel CPUs don't, except the 8-core Atom and presumably Broadwell. If those 16GB DIMMs ever get affordable and readily available, it would probably be in 2015 when there are Broadwell desktops/laptops around.

      • 16GB DDR3 modules are only $165 now, which is roughly the same price as 2x8GB of DDR3.
        • 16GB regsitered ECC DDR3 server modules are only $158 according to newegg but at least on the intel side you need a server board and CPU to use them.

          16GB unregsitered non-ecc desktop DDR3 modules are another matter. Afaict only one specialist manufacturer has announced that they are making them and when I google the part number they list I don't find anywhere actually selling it. Also from what I have read the standard init code that intel gives to bios manufacturers doesn't support 16GB modules and it is u

  • by edxwelch (600979) on Thursday March 20, 2014 @06:44PM (#46539075)

    The eDRAM simply makes the chip way too expensive.
    If you look at the price of i7 core 4770R: $358. It's an i7 but has only has 6 MB of cache (compared to the 8mb of the regular i7 4770). So basically, it's about the same value as a i5-4670K which cost $243. With the price difference you could buy a Radeon R7 260X, which will trash Iris Pro in performance.

  • Unless they have a refresh of the Pro when the chip launches or soon after the Pro is back to being too expensive for the performance.

  • I don't think the poster knows what hardware "enthusiasts" really mean. No one is looking forward to the "e" EXTREME series CPU. Perhaps some people with more money than brains. Enthusiasts want CHEAP hardware that they can then fiddle with to gain big results. Haswell while a decent CPU only really offered better power efficiency and a few more instruction sets that might be potentially useful in a few years. Not exactly a ringing endorsement.

    This sort of thing costs $1000+ dollars, so unless you are a ric

Our informal mission is to improve the love life of operators worldwide. -- Peter Behrendt, president of Exabyte

Working...