Leaked slides tease Haswell-EP details

Although server-grade CPUs based on Intel’s Ivy Bridge silicon have yet to arrive, rumors are already swirling about what’s in the cards for Intel’s next, next enterprise platform. Purported details about Haswell-EP have popped up in a series of slides leaked in the Chiphell forums. The images appear to be legit, and they reveal some interesting tidbits.

Haswell is the successor to Ivy Bridge, and it will be built on the same 22-nm process. The EP version will supposedly have at least 10 cores and support four channels of DDR4 memory at up to 2133 MT/s. There will be around 2.5MB of cache per core, according to the slides, and 40 lanes of gen-three PCI Express connectivity. Hyper-Threading and Turbo Boost will be on the menu, of course.

According to the slides, Haswell-EP will be accompanied by a C610 platform hub code-named Wellsburg. This puppy promises 10 Serial ATA ports capable of 6Gbps transfer rates, an improvement upon the dual 6Gbps and eight 3GBps ports in the current C606 chipset. The C610 is supposed to include as many as six USB 3.0 ports and eight PCIe 2.0 lanes, as well.

Intel is expected to have DDR4-equipped server CPUs available in early 2014, which seems like a plausible timeframe for Haswell-EP. It will be interesting to see if enthusiasts get their own version, ala Sandy Bridge-E. Thanks to CPU-World for the tip.

Comments closed
    • ptsant
    • 7 years ago

    I really want the consumer edition of this product. Something with 10 cores and 4 channels would be perfect for my browsing and email needs.

    • bcronce
    • 7 years ago

    Mmmm.. Transactional memory commands. Threading should scale much much better for optimistic locking.

    • bcronce
    • 7 years ago

    I hope there is a nice low-power decent cost quad core version.

    • yogibbear
    • 7 years ago

    Might just make me ditch my q9450… Still going strong!

      • d34thly
      • 7 years ago

      🙂 yeah, my wife’s still rocking a q6600.

        • sweatshopking
        • 7 years ago

        i’m still using a q6600. all things considered, i really don’t see much reason to upgrade. i’m not hurting for performance.

      • Jakubgt
      • 7 years ago

      Glad to see that I’m not the only one on here still using a Q9450

        • tfp
        • 7 years ago

        Running a Q9400 here, could use a new graphics card though

        • rrr
        • 7 years ago

        Hell, I ‘m still using E8400 and in no particular rush to upgrade.

      • dashbarron
      • 7 years ago

      q9450 and I plan to run it in the ground. Penryn baby Penryn!

    • d34thly
    • 7 years ago

    Looks like I’ll be giving up my Clarksdale in 2014. Finally a hub that is dumping legacy sata and usb whilst providing enough lanes for proper PCI express3.0 SLI/Crossfire.

    • jjj
    • 7 years ago

    The “up to 35MB of LLC” would indicate max cores is 14.
    If there is a consumer versions doesn’t really matter if they price it unreasonable, like they are doing now with the E line.They are the ones loosing by not offering more cores at decent prices as an alternative to paying for a GPU we don’t use.

      • Kurotetsu
      • 7 years ago

      Frankly, I couldn’t care less about them making 10-core CPUs for consumers (which I personally find silly, but that’s me). What I want is for them to start introducing quad core CPUs at the under-$180 to $99 price range. AMD has (or had?) that with their Athlon II line. Right now, the cheapest Intel quad you can get is an i5 at $180. I’d like to be able to make a cheap platform (CPU + Mobo + RAM) with an actual quad core, not dual-core + HT, for roughly $200.

        • smilingcrow
        • 7 years ago

        But AMD’s quad cores have typically been worse than Intel’s dual core with HT so you seem overly focussed on paper specs than actual performance. Don’t read the label but read the reviews.

          • rrr
          • 7 years ago

          Not really.

          i3-2100 vs Phenom II X4 is pretty much a wash, especially since you cannot OC i3.

            • chuckula
            • 7 years ago

            Hrmm… i3 2100 product page at Intel: [url<]http://ark.intel.com/products/53422/[/url<] Lessee here: Core count [b<]2[/b<] (with hyperthreading) Phenom II X4: 4 cores (sorry, no product page link AMD wants these chips to go down the memory hole since they make BD look bad). I think you just agreed with smilingcrow there even though you didn't mean to....

            • rrr
            • 7 years ago

            Reading comprehension requiring some polish? I stated it’s a wash, he stated AMD’s quad cores are worse. How is that agreeing?

      • OneArmedScissor
      • 7 years ago

      [quote<]They are the ones loosing by not offering more cores at decent prices as an alternative to paying for a GPU we don't use.[/quote<] "We" wouldn't be using that for anything but 28 e-pene meters in the Windows task manager. This myth that the GPU could be swapped for more cores needs to die in a fire. Look at Sandy Bridge: Quad-core w/ GPU - 216 mm2 SB-E quad-core - 294 mm2 Despite the fact that the quad-core version of SB-E is made with its own die, not a cut down version of the 8 core chip, it's *edit* some particular amount larger than my tiny brain can quantify *edit.* There is a lot more that has to go into that than just the silicon for a core itself, whereas a GPU can be cut or expanded to practically any size. The GPU would be plenty useful if they'd just standardize switchable graphics on desktops.

        • chuckula
        • 7 years ago

        [quote<] it's still nearly twice the size[/quote<] A bit of an exaggeration.. it's still nearly 40% bigger though, and most of that is due to the fact that the added I/O for SB-E takes up a bunch of real estate even though it doesn't add a huge number of transistors (relatively speaking).

          • OneArmedScissor
          • 7 years ago

          *edit* No math wizardry necessary. My bad! *edit*

          [quote<]added I/O for SB-E takes up a bunch of real estate [/quote<] Which is still going to be the case in a quad-channel, 40 lane chip with even more cores yet. There's not some alternative scenario here where you just plug 14 CPU cores into the standard desktop socket, with no way to feed them.

            • chuckula
            • 7 years ago

            [Edit: Erased after OneArmed noticed the unit notatation issue]

            • OneArmedScissor
            • 7 years ago

            [quote<]The big-mama SB-E die *is* nearly twice the size of the consumer SB chip, so maybe that is the source of your confusion.[/quote<] Oops, that's the problem. Sorry, end of the week brain fart. I somehow just forgot that the 4 and 6/8 core have the same package, so that picture I posted is totally irrelevant and that really threw me off hahaha. *M-M-M-M-MONSTER face palm* I'm not contesting the point that the extra transistors for more cores are potentially quite small, though. Just the idea that you'd somehow keep the same size as the the GPU version and get a few more cores out of it. What you'd have to do is start with SB-E and then add more cores, rather than take away the regular SB GPU and swap it out. It's just going to have to be bigger and more expensive to get anything out of even more cores yet.

            • chuckula
            • 7 years ago

            I agree with your general idea 100%. One problem in stripping the GPU is that Intel never devotes all that much space to the GPU in the first place (one reason why AMD’s IGPs tend to be faster). Additionally, the GPU is very dense (denser than a standard CPU core and a LOT denser than I/O like memory controllers and PCIe). To get a 1:1 swap ratio for stripping the GPU, about the only other feature that really makes sense is more cache, which is also quite dense.

            • OneArmedScissor
            • 7 years ago

            [quote<]To get a 1:1 swap ratio for stripping the GPU, about the only other feature that really makes sense is more cache, which is also quite dense.[/quote<] We've seen that to actually be counter-productive for PCs, though, as the latency just goes up, while no use seems to be made of the increased capacity. Per core, it's 1.5MB for i5s, 2MB for SB i7s, and 2.5MB for SB-E i7s, but sometimes even the i5 actually wins. There just doesn't seem to be anything left to copy and paste in PC CPUs but the GPU. Even though Intel had kept the GPU smaller than AMD, the mainstream version of Haswell looks to go down the same road. Maybe the Phenom II X6 made people think otherwise? AMD just tacked two more cores onto the quad-core Phenom II die, without even adding more L3. However, those weren't hyperthreaded cores, which did less work than Intel's. Even so, the 8 core Opteron was dual-die, and ran all four memory channels. Then came Bulldozer. There is a lesson to be learned in here somewhere!

            • sweatshopking
            • 7 years ago

            STOP THIS GENIAL DISCUSSION RIGHT NOW. THERE IS NO ROOM FOR IT. PLEASE POST FANBOY COMMENTS ONLY.

Pin It on Pinterest

Share This