Leaked specs expose next-gen Xeon lineup

The Turks over at Donanim Haber have gotten their hands on new details concerning Intel’s upcoming Sandy Bridge E processors. The site has posted purported specifications for what looks like the entire lineup of next-generation Xeons, and it’s a doozy. Here’s the full list:

Model Cores Threads Clock Speed Memory speed L3 cache TDP
Xeon E5-2603 4 4 1.8GHz DDR3-1066 10MB 80W
Xeon E5-2609 4 4 2.4GHz DDR3-1066 10MB 80W
Xeon E5-2620 6 12 2.0GHz DDR3-1333 15MB 95W
Xeon E5-2630 6 12 2.3GHz DDR3-1333 15MB 95W
Xeon E5-2630L 6 12 2.0GHz DDR3-1333 15MB 65W
Xeon E5-2637 2 4 3.0GHz DDR3-1600 5MB 80W
Xeon E5-2640 6 12 2.5GHz DDR3-1333 15MB 95W
Xeon E5-2643 6 12 3.3GHz DDR3-1600 15MB 130W
Xeon E5-2650 8 16 2.0GHz DDR3-1600 20MB 95W
Xeon E5-2650L 8 16 1.8GHz DDR3-1600 20MB 70W
Xeon E5-2660 8 16 2.2GHz DDR3-1600 20MB 95W
Xeon E5-2667 6 12 2.9GHz DDR3-1600 15MB 130W
Xeon E5-2670 8 16 2.6GHz DDR3-1600 20MB 115W
Xeon E5-2680 8 16 2.7GHz DDR3-1600 20MB 130W
Xeon E5-2687W 8 16 3.1GHz DDR3-1600 20MB 150W
Xeon E5-2690 8 16 2.9GHz DDR3-1600 20MB 135W

These details aren’t confirmed, but they do look plausible. Highlights? Check out the Xeon E5-2687W, which runs at 3.1GHz and boasts a hefty 150W thermal envelope. Intel’s 5000-series Xeon line (which the E5s look set to replace) doesn’t have anything with a TDP greater than 130W. That said, the 5000-series family tops out at six cores, three memory channels, and 12MB of L3 cache, while the E5s scale up to eight cores, four channels, and 20MB of cache.

At the very least, Sandy Bridge E appears to have a greater thirst for wattage than its Gulftown predecessor. Take the E5-2667, for example. It has the same core count and clock speed as the existing Xeon X5670. However, the X5670 has a 95W thermal envelope, while the E5-2667 has a 130W TDP. Thankfully, the E5 lineup has a number of low-power options, including an eight-core model with a 70W TDP rating.

Donanim Haber doesn’t have any new information on pricing or an expected street date for these new CPUs. I suspect we’ll see at least a couple of the new Xeon variants manifest themselves as Extreme Edition desktop chips targeted at X79 mobos, but we don’t yet have a clear indication of when those chips will arrive.

Comments closed
    • DavidC1
    • 8 years ago

    “Take the E5-2667, for example. It has the same core count and clock speed as the existing Xeon X5670. However, the X5670 has a 95W thermal envelope, while the E5-2667 has a 130W TDP.”

    Bad comparison. Compare the E5-2667 and the E5-2643. They both use 130W and both have 12 threads but the latter is clocked 400MHz higher. There’s probably something in the 2667 that makes it clocked lower.

    The 150W TDP isn’t bad as it looks like. On the Westmere Xeons, it needs the 5500 chipset and the ICH10 I/O hub. With the Sandy Bridge successor, its all a single chipset. So it saves a few watt there, even though chipset reaches TDP far less often than CPUs do.

      • axeman
      • 8 years ago

      [url<]http://www.youtube.com/watch?v=03C4tnCFEMQ[/url<]

      • Game_boy
      • 8 years ago

      The difference between those two is binning.Variance in a wafer will mean some CPUs will use less power at the same clock than others. Intel uses this to make different models in the lineup.

      TDP is not power use, it is something like the worst-reasonable-case scenario power use. Power use could be higher if a power virus was used, or much lower in typical workloads.

      • willmore
      • 8 years ago

      And it’s easier to cool one super hot chip that two cooler ones? Well, actually, thermodynamically, it’s more *efficient* to do so, but then in steps engineering and points out the practicality of doing so isn’t so hot.

      Plus, one has to question the idea of scaling I/O with processor capacity. Memory bandwidth and processor to processor interconnect bandwidth, sure, but device I/O? I’m not sure their architecture makes much sense in the workloads with which I’m familiar.

    • tfp
    • 8 years ago

    Does the L3 cache still run at a slower clock speed than the CPU cores?

    • LoneWolf15
    • 8 years ago

    Will all of these be planned for server sockets/LGA2011, or will there possibly be some Socket 1155 love?

    (I think I know the answer, but it doesn’t hurt to ask)

    • zdw
    • 8 years ago

    Anyone else think this is a stupidly large number of CPU’s with fairly arbitrary distinctions?

    I mean, the only thing that changes is the last 2 digits and you can go from a 2 core 4 thread model to an 8 core 16 thread model using more than twice the power?

    A better naming scheme at least would remove some confusion…

      • Farting Bob
      • 8 years ago

      Names arent as important when you are dealing with just businesses. They look beyond the name when deciding which chips to buy. You can name them CPU 101283028, CPU 101283029, CPU 101283030 etc etc and it wont make a bit of difference.

    • flip-mode
    • 8 years ago

    On its own, 150 watts for a CPU sounds high, but when I remember that is a CPU with 8 cores and 16 threads, it sounds a whole lot more reasonable.

    You could probably comfortably run 100 low-intensity virtual machines on that 150 watt processor. Heck, I wouldn’t be surprised if you could run two or three times as many. The power consumption of the RAM would end up being way higher than 150 watts.

    • OneArmedScissor
    • 8 years ago

    “Take the E5-2667, for example. It has the same core count and clock speed as the existing Xeon X5670. However, the X5670 has a 95W thermal envelope, while the E5-2667 has a 130W TDP.”

    If it’s anything like the existing Sandy Bridge, and the cache and ring bus are synced to the core clock speed, there’s your explanation.

    This really should be compared to Westmere EX. At the same TDPs, it actually has higher clock speeds across the board.

    • ronch
    • 8 years ago

    Man, I can’t wait for the next battle round between Santa Clara and Sunnyvale.

    • chuckula
    • 8 years ago

    Part of the higher power consumption is going to be due to the massive (~40) number of PCIe 3.0 lanes that are now on-die instead of in a separate north or southbridge chip as in the Westmere and earlier processors. Overall, an 8 core CPU with a 2.9Ghz clock and 135 watt TDP is not doing too bad at all when you look at the clock speeds of the earlier generation chips.

    Edit: See the list of Westmere Xeons in this article: [url<]http://www.xbitlabs.com/news/cpu/display/20110405135321_Intel_Launches_New_Xeon_Chips_with_Up_to_Ten_Cores.html[/url<] The fastest clock I saw was 2.67Ghz for a non-hyperthreaded 8-core Westmere, while these Sandy Bridge based Xeons are doing 2.9 Ghz with the extra PCIe lanes and with hyperthreading while only having 5 watts of additional TDP... not too shabby. On the low power side a 1.7 Ghz Westmere had a TDP of 105 watts while a 1.8 Ghz Sandy Bridge with more cores is at 70 watts.

      • FuturePastNow
      • 8 years ago

      “Part of the higher power consumption is going to be due to the massive (~40) number of PCIe 3.0 lanes that are now on-die instead of in a separate north or southbridge chip as in the Westmere and earlier processors.”

      And if they have good power gating, which I’m sure they will, the CPU will shut down anything that is not being used. Real-world power consumption figures will be much, much lower than the TDP.

        • Farting Bob
        • 8 years ago

        Agreed, these chips are monsters if you fully load them, and the high TDP allows them to scale higher than they would otherwise if you have the cooling to handle it. But real world usage in servers and very high end workstations will be much less than the TDP. Idle will probably be around the same as SB i7’s, maybe a bit higher thanks to the added stuff but not much so.

        • grantmeaname
        • 8 years ago

        Nope, for Intel the TDP is the average consumed power and not the thermal design power. The power consumption could peak much, much higher than 150W, and we have no reason to say it couldn’t approach that figure during real-world use.

          • chuckula
          • 8 years ago

          Can you provide any evidence to support that assertion? I have yet to see an objective review of a non-overclocked Intel chip where the actual power usage exceeds the TDP even under very high-stress benchmarks, much less “on average”.

            • grantmeaname
            • 8 years ago

            Nope, because these chips haven’t been released yet, so there aren’t any reviews.

            • chuckula
            • 8 years ago

            Wait a minute… you make an over the top statement that Intel is lying and that their chips only list TDP as an “average” power. Then when I ask you to produce evidence to support this conclusion you try to turn the tables on me? What evidence do you have that ANY Intel chip ever produced now or in the future will have the TDP being an “average” power??? You have 0 evidence that these future chips will have an “average” power that exceeds the TDP. I could say that they wouldn’t use any power whatsoever and have exactly as much evidence for that stupid statement as you have.

            It looks like the AMD fanboy brigade is out in force today. Make sure John Fruhe pays you for all this drivel.

            • Antimatter
            • 8 years ago

            Since Intel defines TDP as the heat required be dissipated by the heatsink, it is possible for an Intel CPU to consume more power than its TDP rating. AMD defines TDP as the maximum possible power draw of CPU under worst case conditions.

            [url<]http://www.silentpcreview.com/article169-page3.html[/url<]

            • intangir
            • 8 years ago

            [b<]Incorrect.[/b<] [url<]http://support.amd.com/us/Processor_TechDocs/43374.pdf[/url<] page 80: ""7. The processor thermal solution should be designed to accommodate thermal design power (TDP) at Tcase,max. [b<]TDP is not the maximum power of the processor.[/b<]" Look at the tables on page 68 to see different values for Max Power and TDP.

            • MadManOriginal
            • 8 years ago

            Awesome, a reference from 2004. herpderp?

          • Voldenuit
          • 8 years ago

          I think you’ve got it backwards. It was AMD who started using ACP instead of TDP, probably because they were usually a node or so behind intel.

          From [url=http://en.wikipedia.org/wiki/Average_CPU_Power<]wiki[/url<]: [quote<]The average CPU power (ACP), is a scheme to characterize power consumption of new central processing units under "average" daily usage, especially server processors, the rating scheme is defined by Advanced Micro Devices (AMD) for use in its line of processors based on the K10 microarchitecture (Opteron 8300 and 2300 series processors).[/quote<]

            • grantmeaname
            • 8 years ago

            TDP used to be Total Dissipated Power for both manufacturers. Then Intel switched it to Thermal Design Power — a chip with an old TDP of 130 might have a new TDP of 115 or 100. In response, AMD started reporting the same number that Intel reports as TDP but under the name ACP, average consumed power.

            • chuckula
            • 8 years ago

            Please post one credible benchmark that shows the Intel CPU exceeding its stated TDP (overclocked CPUs don’t count here). Even a synthetic benchmark that stresses the CPU beyond the load that it would encounter in any real-world task would be interesting to look at. You like to make bold statements that disparage Intel, but you don’t seem to want to back up the statements with facts.

            • NeelyCam
            • 8 years ago

            ^ This. Enough with this Wiki and SPCR linking.

            Hard numbers, or it’s not real. (FWIW, I don’t remember Intel CPUs exceeding their stated TDPs for the five-some years I’ve been using them)

            EDIT: Ah, I see that the AMD fanboi Stealth Thumbdown Brigade just landed.

            • OneArmedScissor
            • 8 years ago

            Intel defines TDP in their thermal design guides as:

            “Thermal Design Power: Thermal solution should be designed to dissipate this target
            power level. TDP is not the maximum power that the processor can dissipate.”

            [url<]http://www.intel.com/content/dam/doc/guide/2nd-gen-core-lga1155-socket-guide.pdf[/url<] Section 6.1: "Analysis indicates that [b<]real applications[/b<] are unlikely to cause the processor to consume maximum power dissipation [b<]for sustained time periods.[/b<] Intel recommends that complete thermal solution designs target the Thermal Design Power (TDP), instead of the maximum processor power consumption. The Adaptive Thermal Monitor feature is intended to help protect the processor [b<]in the event that an application exceeds the TDP recommendation for a sustained time period.[/b<]" They make it very clear that it's not an absolute peak, can be exceeded by normal applications, but that the cooling doesn't need to take that into account because the processor will throttle itself if heat builds [b<]continuously[/b<]. It also describes in there exactly how [b<]turbo boost intentionally exceeds TDP.[/b<] This has been a well established fact for a long time. How would you even know that CPUs don't exceed it? Do you sit there pouring over logs of a $3,000 power meter that accurately measures every single spike? What they won't do is sit there sucking down hundreds of watts for an hour straight, as the temperature would increase until it eventually becomes dangerous, but that's completely dependent on your own cooling.

            • esterhasz
            • 8 years ago

            Thanks for taking the time to look that up.

            • intangir
            • 8 years ago

            AMD defines TDP the same way:

            [url<]http://support.amd.com/us/Processor_TechDocs/43374.pdf[/url<] page 10: "TDP. Thermal Design Power. The thermal design power is the maximum power a processor can draw for a thermally significant period while running commercially useful software." page 80: ""7. The processor thermal solution should be designed to accommodate thermal design power (TDP) at Tcase,max. [b<]TDP is not the maximum power of the processor.[/b<]" And if you look at the tables on, say, page 68, the Max Power and TDP rows are different, with Max Power significantly higher.

            • chuckula
            • 8 years ago

            Thanks for that information. P.S. –> To all the AMD fanboys, I never said that an Opteron/Phenom/Bulldozer chip would exceed its stated TDP either in practical use. I’m still interested in seeing some sort of benchmark that actually lets either an Intel or AMD CPU actually exceed its TDP in operation.

            • NeelyCam
            • 8 years ago

            [quote<]How would you even know that CPUs don't exceed it? Do you sit there pouring over logs of a $3,000 power meter that accurately measures every single spike?[/quote<] A [u<]very[/u<] fair point. Moreover, I wouldn't even see those spikes unless I measure everything after on-board regulators, as they alone filter out the fastest spikes. All I've done is low-tech Kill-A-Watt measurement between the wall and a PicoPSU brick. The brick itself levels spikes, then the regulators, and finally any off-chip capacitors. I admit - any measurement any of us can do is measurement is average by definition. (My Kill-A-Watt measures the whole system, including the drives, mobo, PSU inefficiencies etc., and with eight Prime threads active and 2600K running at 100%, I can't get KillAWatt to show anything beyond 87W... clearly an 'average' result). Realization: TDP is entirely academic. In the age of Turbo and active thermal management, it's nothing but a recommendation for the system designer (or, alternatively, a marketing term that really means next to nothing).

            • chuckula
            • 8 years ago

            [quote<]How would you even know that CPUs don't exceed it? Do you sit there pouring over logs of a $3,000 power meter that accurately measures every single spike?[/quote<] It's not that hard and it doesn't cost $3000. CPUs use DC power so all you need to remember is that Power = E*I (voltage times current) and measure from the appropriate rails AFTER the PSU converts AC to DC. Xbitlabs has a page describing one setup right here: [url<]http://www.xbitlabs.com/articles/cpu/display/sempron-3000_9.html[/url<] describing the process, and no, I don't think they spent $3000 in equipment to measure the power consumption of a Sempron. As for "how do you know power consumption every second!" that's a canard. You don't HAVE to know the power consumption every second, you just need to figure out what the CPU will draw under an artificially high load (and of course under more normal loads too for reference). I really don't care if either an AMD or Intel CPU draws 200 watts for 5 microseconds in some statistically insignificant spike. I *do* care about what the CPU will draw under a heavy and continuous load.

            • MadManOriginal
            • 8 years ago

            And under heavy and continuous load Intel CPUs have not gotten near their TDP for a long time. AMD CPUs on the other hand…

            *waits for the thumb-down brigade*

            • NeelyCam
            • 8 years ago

            [quote<] I really don't care if either an AMD or Intel CPU draws 200 watts for 5 microseconds in some statistically insignificant spike.[/quote<] Yeah - you don't, I don't, and most people don't, but purists who claim TDP numbers are BS do care. And they can't be proven wrong, either.

            • willmore
            • 8 years ago

            And if it were just a thermal issue, you’d have a good case–heat takes time to build and move. OTOH, the voltage regulators *do* care about things that just last *microseconds* and can die quite quickly when stressed way beyond their designed operating level.

        • MadManOriginal
        • 8 years ago

        Power consumption has been lower than TDP for a long time for Intel, it’s only simpletons who look at nothing other than the word ‘power’ in TDP that think otherwise.

Pin It on Pinterest

Share This