Announcement

Collapse
No announcement yet.

Calxeda Claims 15x Advantage Over Intel Xeon

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Calxeda Claims 15x Advantage Over Intel Xeon

    Phoronix: Calxeda Claims 15x Advantage Over Intel Xeon

    Calxeda has put out its first benchmark of their forthcoming Calxeda ARM Server. The company is claiming a 15x performance-per-Watt advantage over a recent Intel Xeon CPU...

    http://www.phoronix.com/vr.php?view=MTEyMzc

  • #2
    Originally posted by phoronix View Post
    Phoronix: Calxeda Claims 15x Advantage Over Intel Xeon

    Calxeda has put out its first benchmark of their forthcoming Calxeda ARM Server. The company is claiming a 15x performance-per-Watt advantage over a recent Intel Xeon CPU...

    http://www.phoronix.com/vr.php?view=MTEyMzc
    Well, that's a joke. Let's see what happens when both servers are approximately the same performance.

    I fully expect ARM to win, but as badly as this test was setup to favor ARM I'm starting to have my doubts.

    Comment


    • #3
      Originally posted by smitty3268 View Post
      Well, that's a joke. Let's see what happens when both servers are approximately the same performance.

      I fully expect ARM to win, but as badly as this test was setup to favor ARM I'm starting to have my doubts.
      What exactly do you mean by same performance? Micro servers are being built for specific workloads and are showing nearly the same performance for a fraction of the power usage. A Pandaboard ES or Cotton Candy has excellent performance as a general purpose computer that runs Ubuntu all for a fraction of the cost and power. Once Cortex A15 based systems start coming out at 2Ghz in the quad and eight core varieties as well as the new ARMv8 64 bit parts there will be an even greater advantage for ARM. Also take into account that process is getting smaller as well where 22nm and within a couple of years 18nm and 16nm parts will be shipping thus increasing the performance/power ratio even greater in ARM's favor. Couple this with high end GPUs like the nVidia Kepler or Mali T6xx series and it makes an ARM laptop/desktop even more of a reality.

      Comment


      • #4
        Originally posted by smitty3268 View Post
        Well, that's a joke. Let's see what happens when both servers are approximately the same performance.

        I fully expect ARM to win, but as badly as this test was setup to favor ARM I'm starting to have my doubts.
        They're talking about performance PER WATT, which scales in a fairly linear manner once you're adding lots of CPU cores.

        Comment


        • #5
          Originally posted by Darkseider View Post
          What exactly do you mean by same performance? Micro servers are being built for specific workloads and are showing nearly the same performance for a fraction of the power usage. A Pandaboard ES or Cotton Candy has excellent performance as a general purpose computer that runs Ubuntu all for a fraction of the cost and power. Once Cortex A15 based systems start coming out at 2Ghz in the quad and eight core varieties as well as the new ARMv8 64 bit parts there will be an even greater advantage for ARM. Also take into account that process is getting smaller as well where 22nm and within a couple of years 18nm and 16nm parts will be shipping thus increasing the performance/power ratio even greater in ARM's favor. Couple this with high end GPUs like the nVidia Kepler or Mali T6xx series and it makes an ARM laptop/desktop even more of a reality.
          Hmm, i somewhat take back my comment. The performance of that server is much better than i realized.

          However, this gives me pause:

          The Intel (Sandybridge) platform is based on published TDP values for the CPU and I/O chipset, along with an estimate for DDR memory. Unfortunately, at the time of this blog post, we didn’t have a way to measure actual power consumption with the same level of fine detail.
          I don't think the standard TDP measurements are generally very accurate. At least, I know they aren't on the desktop. Maybe it's better for Xeon servers, but i doubt it.

          Also, Apache is pretty much the best case for these servers, or at least that was always true for those old servers Sun used to create that were similar (low-power/high-core), i can't remember what they were called. But since Apache is actually a decent use-case, that's valid enough.

          Comment


          • #6
            They seem to be serving static content; no, apache is far from the top in that. But if they're maxing the pipe either way...

            Comment


            • #7
              Originally posted by Darkseider View Post
              What exactly do you mean by same performance? Micro servers are being built for specific workloads and are showing nearly the same performance for a fraction of the power usage. A Pandaboard ES or Cotton Candy has excellent performance as a general purpose computer that runs Ubuntu all for a fraction of the cost and power. Once Cortex A15 based systems start coming out at 2Ghz in the quad and eight core varieties as well as the new ARMv8 64 bit parts there will be an even greater advantage for ARM. Also take into account that process is getting smaller as well where 22nm and within a couple of years 18nm and 16nm parts will be shipping thus increasing the performance/power ratio even greater in ARM's favor. Couple this with high end GPUs like the nVidia Kepler or Mali T6xx series and it makes an ARM laptop/desktop even more of a reality.
              OK, I'm not the original person calling the comparison bogus, but here are some issues:
              1. They compare a measured value against the TDP of the Xeon, not the actual usage. The Xeon was at 15%, and on top of that TDP != power draw.
              2. They have 4 GB in the ARM, 16 GB in the Xeon. Using BS TDP that extra 12GB uses 12 W extra. Unfair!
              3. They exclude the hard drive and the power supply, which is not logical.
              4. The Xeon saturated the link at 15% load. You would never use that chip in that workload with a single NIC. You would have a 4-port board at least.
              5. The V2 Xeon's are out which give performance increases and power-savings.

              So here's my equally bogus made up comparison, assumptions in ()'s:
              Xeon E3-L-V2 (17W at 100%) with 4 NIC board (linear scaling), 4GB RAM (4W), HD (7W), and PSU (~90% efficient)
              (6950×4)÷((4+6.7+(17×1)+7)*1.1) = 728.3 req/W

              Calxeda
              5500÷((5.26+7)*1.1) = 407.8 req/W

              Well, my-oh-my! Calxeda lose! Even if some of my assumptions and figures were worse, it would certainly be nowhere near a 15x advantage.

              Comment


              • #8
                Originally posted by droidhacker View Post
                They're talking about performance PER WATT, which scales in a fairly linear manner once you're adding lots of CPU cores.
                But not with speed per-core, which was my point. That ARM server is fast enough that i take back my original argument though.

                But my point was that if a single-core was providing 15x more speed, it's not a surprise that it would be 15x less efficient per watt.

                Edit - and again i take this back, if what the above person is saying is true. (yep, it is)
                The Sandybridge system saturated the single 1Gb NIC with less than 15% CPU utilization.
                I'll just give up now, since i don't have the time to actually look into this myself.

                But if the xeon is saturating the link without maxing out the cpu, then it seems likely that the ARM cpu really does suck compared to it performance-wise.

                So my original argument is back - cut down that Intel CPU or beef up the arm one, and things are likely to be very different. And it's just stupid to try and include full TDP for one platform when you arent' even maxing it out.
                Last edited by smitty3268; 06-20-2012, 03:13 PM.

                Comment


                • #9
                  Originally posted by kiwi_kid_aka_bod View Post
                  OK, I'm not the original person calling the comparison bogus, but here are some issues:
                  No need to go into details like you did. It's obviously such a bogus comparison that their press release just did them more harm than good. It actually gave me a chuckle.

                  I really hope Microsoft and Oracle invest heavily in their technology.

                  F

                  Comment


                  • #10
                    And now it's hit engadget! Cue the misplaced fawning over the misrepresented and bogus results.

                    As an aside, the guy at http://www.servethehome.com/ has done tests of systems with Xeon's power usage at idle and load. Typically a 70W delta! So at 15% load that chip would have been using around a quarter of the value they put into their chart.

                    Comment


                    • #11
                      this bench is such a joke. and we're not even april one.

                      Comment


                      • #12
                        Me thinks that there is more to all of this than meets the eye. HP and Dell would not be investing in ARM micro servers as heavily as they are without reason. Aside from the Intel Centerton and the new AMD Fusion APUs coming out ARM will have Cortex A15 SoCs in quad and 8 way configurations late this year/early next. If anything the Cortex A9 based units from HP and Dell may just be for product validation and testing with the intention of going full scale with Cortex A15 parts.

                        Comment


                        • #13
                          Originally posted by Darkseider View Post
                          Me thinks that there is more to all of this than meets the eye. HP and Dell would not be investing in ARM micro servers as heavily as they are without reason. Aside from the Intel Centerton and the new AMD Fusion APUs coming out ARM will have Cortex A15 SoCs in quad and 8 way configurations late this year/early next. If anything the Cortex A9 based units from HP and Dell may just be for product validation and testing with the intention of going full scale with Cortex A15 parts.
                          There are two reasons:
                          1. ARM is low power, although frankly it won't be much better than the new dual-core (and quad-core next year) ultradense Atom server blades that are coming out at the same time. As this review showed, "low power" means absolute power draw only here, and *not* performance-per-watt where the beefier Intel system is definitely in the lead despite the massive number fudging used by Calxeda to show otherwise.

                          2. ARM is *dirt freakin cheap* since you can get a full SoC that includes most of your peripherals for very little money. Part of the issue here is that you often give up a bunch of the high-speed and high-performance I/O interfaces that people take for granted on the PC. For example, something as simple as a PCIe controller is *incredibly* rare in the ARM world (I think Marvell may be the only company anywhere with a product, although some other companies may come up with solutions next year). So once again it is a tradeoff.

                          There is certainly a niche for ARM servers, but I find it hilarious when the same people who say that Intel can't ever make a dent in the smartphone market also say that ARM is going to rule the server market by next Tuesday or something. Intel has gone a much much greater distance in the direction that it needs to go to compete in mobile devices than ARM has in going to gain any real foothold in the server market. Case in point: 64 bit? Doesn't exist in the ARM world except for projections of what will happen when ARM v8 devices start to hit in 2014 (or maybe even 2015). ECC RAM support? Doesn't exist in the ARM world, they need to redesign the memory controllers for it. The list goes on from there.

                          Comment


                          • #14
                            Adjusting for those TDP issues, it looks like the ARM chip wins by about 4.5 times

                            And the test is still tilted heavily in their favor. But 4.5x seems much more realistic than 15x.

                            Comment


                            • #15
                              Originally posted by chuckula View Post
                              There are two reasons:
                              1. ARM is low power, although frankly it won't be much better than the new dual-core (and quad-core next year) ultradense Atom server blades that are coming out at the same time. As this review showed, "low power" means absolute power draw only here, and *not* performance-per-watt where the beefier Intel system is definitely in the lead despite the massive number fudging used by Calxeda to show otherwise.

                              2. ARM is *dirt freakin cheap* since you can get a full SoC that includes most of your peripherals for very little money. Part of the issue here is that you often give up a bunch of the high-speed and high-performance I/O interfaces that people take for granted on the PC. For example, something as simple as a PCIe controller is *incredibly* rare in the ARM world (I think Marvell may be the only company anywhere with a product, although some other companies may come up with solutions next year). So once again it is a tradeoff.

                              There is certainly a niche for ARM servers, but I find it hilarious when the same people who say that Intel can't ever make a dent in the smartphone market also say that ARM is going to rule the server market by next Tuesday or something. Intel has gone a much much greater distance in the direction that it needs to go to compete in mobile devices than ARM has in going to gain any real foothold in the server market. Case in point: 64 bit? Doesn't exist in the ARM world except for projections of what will happen when ARM v8 devices start to hit in 2014 (or maybe even 2015). ECC RAM support? Doesn't exist in the ARM world, they need to redesign the memory controllers for it. The list goes on from there.
                              did you even read the publicly available spec http://www.calxeda.com/wp-content/up...-Brief-612.pdf it says right there
                              "Key Features"
                              "..Integrated high-performance interfaces such as memory
                              controllers with full ECC support
                              and I/O subsystems for local SATA
                              2.0 ports and PCIe 2.0 support"

                              did you blink and miss all these options for ARM you say are missing but in fact are NOT missing at all , in fact some ARM blocks are at the for front of improving speed for everyone even in the x86 space with scalable native PCIe to SSD SOC
                              http://denalimemoryreport.wordpress....d-ssd-designs/

                              http://www.marvell.com/storage/syste...rollers-WP.pdf

                              ECC RAM support hardware blocks are everywhere just because YOUR current ARM Mobile phone SOC doesn't have it specified doesn't mean it doesn't exist perhaps you might try the ARM database first before wrongly assuming next time http://infocenter.arm.com/help/index...h09s04s02.html its even got its own entry on the latest High speed ARM AMBA 4 AXI bus OC

                              Cortex-R5 Technical Reference Manual Revision: r1p2
                              Home > Level Two Interface > AXI slave interface > TCM ECC support

                              9.4.2. TCM ECC support


                              The TCMs can support ECC, as described in TCM internal error detection and correction. If a write transaction is issued to the AXI slave, the slave interface calculates the required ECC bits to store to the TCM. If the write data width is smaller than the ECC chunk size then a read-modify-write sequence is automatically performed by the AXI slave."



                              what else, oh 64bit samples exist and are being worked on right now http://lists.linux-foundation.org/pi...ne/000551.html
                              "Will Deacon arm.com The majority of work and discussion around the ARM port is currently focussed on platform code and soc support, as shown by the git traffic generated from the new arm-soc tree.

                              However, there are still a few of us working on the core architecturewhich now has support for large physical addresses, virtualisation and other exciting bits and pieces which require work in the core port. There's also a 64-bit architecture on the way...." and when he says on the way he means generally
                              availability OC ,not it doesn't exist yet, its already with partners.
                              http://lists.linux-foundation.org/pi...ne/000574.html
                              ".... PCI breakout session. One of the next major pieces for ARM
                              consolidation and DeviceTree support is PCI support. More ARM platformsare appearing with PCIe. I think there's opportunity for consolidation with PowerPC DT PCI code."

                              Samsung sell a 512bit memory bus SOC block and chips etc ....you get the idea, Google an ARM block of your choice and it will probably already exist today, and if so it will be in someones commercial SOC and being used right now somewhere you never even considered... will it be available in ARM YOUR Phone next Tuesday for you to use OC not, but it may already be in that SOC and not even used OC so you would never know, for instance.

                              http://lists.phcomp.co.uk/pipermail/...ne/004373.html
                              "Henrik Nordström wrote:> tor 2012-06-14 klockan 19:45 +0800 skrev Tom Cubie:
                              >
                              >
                              >> All hardware? That will be interesting. There are some peripherals in
                              >> A10 are never used, nor there are linux drivers for it,even very few
                              >> people know A10 has that function, like can bus controller, memery
                              >> stick conroller...
                              >
                              > Is the memory stick controller available to be configured on any pins?
                              > Have only seen references to there being a ms controller but no pin
                              > information at all. Or any driver...


                              information on memory stick is subject to NDAs. E.g. TI always
                              omitted MS docs for their chipsets and only provided them
                              to MS NDA customers..
                              "
                              Last edited by popper; 06-25-2012, 01:13 AM.

                              Comment

                              Working...
                              X