POST A COMMENT

34 Comments

Back to Article

  • ddriver - Thursday, September 04, 2014 - link

    Why include gaming benches, considering this product is not intended nor optimal for the task, while at the same time prosumer and enterprise tests are completely absent? Reply
  • ATC9001 - Thursday, September 04, 2014 - link

    The gaming benches are just for good FYI, considering most of the audience are gamers/ enthusiasts. However, I do agree I would like to see more prosumer workload benchmarks! Reply
  • ddriver - Thursday, September 04, 2014 - link

    Where is 3D MAX / MAYA 3d rendering? Where is AfterEffects / Premiere / Nuke / Fusion video rendering? Where is Cubase / ProTools / Sonar audio rendering? Where are multiphysics simulations? Where are all the practical workloads that would justify the purchase of a dual socket workstation and that can actually scale up so many cores?

    Instead we got games, everyday office and multimedia applications and 2 completely impractical synthetic tests. And it is not just that review, this is becoming a common practice at reviewing workstation class mobos. Almost as ridiculous as using different JS engine implementations to measure raw CPU power in mobile devices...
    Reply
  • cen - Thursday, September 04, 2014 - link

    I completely agree. Anandtech used to have a server division but apparently this is gone now. I have noticed several times that server grade products get non-server related treatment, rendering the whole review completely pointless. Reply
  • xyzzy1954 - Friday, September 05, 2014 - link

    I'll second that. I just bought a 'gamer' setup, but for video and audio rendering. I also do a lot of OLAP analytics. I'd really enjoy a review / comparison of 97 to 99 systems for these purposes. Threads matter, RAM > 32 GB matters. I don't get all this hardware from manufacturers for review purposes...so I can't afford to do the comparisons. I feel confident that AnandTech would publish the comparisons if they had all the relevant tests 'off the shelf' (like they already do for game-based testing). Maybe we should crowd source some tests for them.... Reply
  • wintermute000 - Saturday, September 06, 2014 - link

    yeah I second that. SQL, virt, photoshop, blender, cubase etc. should be the kind of tests not games and office apps!!!! Reply
  • Kevin G - Sunday, September 07, 2014 - link

    I personally don't mind the office and gaming benchmarks as they provide continuity between reviews of consumer class hardware. It is irritating is they're the only benchmarks to be had though. A workstation motherboard needs some workstation tests! Reply
  • duttyfoot - Tuesday, September 09, 2014 - link

    i feel the same, if you review this type of system you should at least add cinebench along with all the other applications mentioned. Reply
  • MikeMurphy - Wednesday, September 10, 2014 - link

    I really enjoyed the gaming benchmarks. I've always wondered how these boards would perform in those circumstances. Reply
  • Samus - Thursday, September 04, 2014 - link

    To show this thing is a monster, except when gaming ;) Reply
  • ShieTar - Friday, September 05, 2014 - link

    Well, there is a tiny group of people who will be interested: They need a Workstation because they independently work with their PC to create content, but if possible they would also like to use the existing setup to game occasionally, instead of spend another 1k$ and put another Case next to their desk.

    Also, 10 years ago a lot of Trolls were posting "Yeah but can it run Quake" in the comment section of server hardware reviews. This kind of testing shuts them up.
    Reply
  • xxsk8er101xx - Saturday, September 06, 2014 - link

    Because the motherboard is designed for the prosumer market. Not the professional/server market. The point of the article is to basically say that this motherboard servers no purpose. It can't be used for the professional/server market because there are too few RAM slots and no mini-sas slots which the majority of the professional market has a need for. Reply
  • Kevin G - Sunday, September 07, 2014 - link

    I think this motherboard has priced itself out of the prosumer market a bit at $650 USD and then you have Xeon prices on top of that. I do see this system being a decent pick for some professional tasks though. RAM capacity is still pretty decent as this board accepts 32 GB DIMMs for 256 GB capacity. That's a bit expensive due to 32 GB DIMM pricing but I see 128 GB via 16 GB DIMMs as reasonably priced for what you get. Needing more than 128 GB is a very small niche and one that overlaps widely with the idea of just using servers to batch process the large data sets anyway.

    The PCIe setup does scream professional though. Components like capture cards and PCIe SSDs are not going to be restricted in performance. (Though it is worrisome that this board doesn't have a dedicated aux power for just the PCIe slots.) The presence of onboard audio also indicates that this certainly isn't a server board.

    I have yet to encounter a good reason for SAS in the professional space. 6 Gbit SATA provides enough bandwidth and SSD's can saturate that. SATA hard drives are adequate for bulk storage locally. (There is definitely a niche for SAS in server space with multi path IO for redundancy.) I can see a need for RAID5/6 in the prosumer and professional space but SAS isn't a requirement for it (though incidentally most RAID5/6 cards are SAS based). Even then, in most cases software RAID1 is 'good enough' for redundancy to prevent downtime in the professional space.
    Reply
  • AnnihilatorX - Sunday, September 07, 2014 - link

    Kindly read the article!: "While gaming is not a focus of motherboards like the GIGABYTE GA-7PESH3, the system may be in use by content developers relying on an accurate representation with what they are making. " Reply
  • gchernis - Thursday, September 04, 2014 - link

    Nice write-up! Do you have any data on total CPU utilization with multithreaded tasks? What about kernel-time vs. user mode time split? Reply
  • JellyRoll - Thursday, September 04, 2014 - link

    "Server motherboards historically have done rather poor in this test,"....perhaps because it is a server motherboard? are you testing a server workload on it? Reply
  • Samus - Thursday, September 04, 2014 - link

    Anand used to do SQL benchmarks. This is what we need for this review to be complete. Reply
  • mavere - Thursday, September 04, 2014 - link

    The Handbrake results are a bit disappointing because data-dependency breakpoints occur both naturally (at scene boundaries) and manually (at regular ~10s key-frame intervals). Conceptually, the transcoder can easily reuse the source's chapter-timing metadata and keyframes to outsource chunks of work to each socket with minimal communication overhead.

    I'm guessing that NUMA performance was never a pressing concern for any of Handbrake's and x264's contributors.
    Reply
  • JDG1980 - Thursday, September 04, 2014 - link

    I just can't see a good reason to buy into Ivy Bridge at this time, when Haswell-EP is just around the corner. Personally, I'm seriously considering the E5-1650 v3, assuming it comes in at a similar price point to its predecessor (and assuming that there are some decent workstation boards with ECC support available).

    And it's especially hard for AnandTech to justify the time reviewing this soon-to-be-obsolescent board when they still haven't gotten around to Tonga. I know it's not the most groundbreaking new GPU design (and AMD picked about the worst place in their lineup they possibly could to slot it), but still, it's a new GPU and we haven't seen a review despite every other major site doing so 2 days ago. What's going on with this?
    Reply
  • ruthan - Thursday, September 04, 2014 - link

    I would buy such machine for gaming if there would be performance benefit, so its good to know and mainstream everytime rules.. Its like SPARC or Power architecture, they cant compete because of mainstream x86 market is huge so, research revenue there are also huge, users requirements are bigger, more universal. This Sparc, Ithanium or Power market exist only because of stupid company policies, like everything from one company, support only for this HW, lies about super duper optimalisation for such HW, and especially license per CPU socket (here realy doesnt matter about price of CPU because, SW licence per socket is astronomical, so you are forced to buy Sparc or Power machine..). Reply
  • tuxRoller - Thursday, September 04, 2014 - link

    So, you've got a workstation class board but are running Windows 8...
    Is it so difficult to throw centos on there for the benefit of those working at Pixar?
    Reply
  • MrSpadge - Friday, September 05, 2014 - link

    Those working at Pixar likely won't need an Ivy-EP review. Apart from that.. I'm sure Ian is glad not to have to deal with CentOS :p Reply
  • tuxRoller - Friday, September 05, 2014 - link

    Centos is easy to setup, and the new release is quite nice.
    My point with Pixar was that there are many serious creators who use Linux workstations, and not just for simulations. Unless you work in the area it's possible you may not know the prevalence of Linux among workstations, especially relative to windows 8, in my experience.
    That aside, seeing the comparative performance numbers of the two OS's is useful to get an idea of what the hardware itself can do.
    Reply
  • mpbrede - Friday, September 05, 2014 - link

    I really hope that this crappy grammar is just an indication of haste and not of lower editorial standards now that Anand has left the room.

    "As a result, gaming often sees a hit in performance, as well as basic tasks." -> ...result, gaming, as well as basic tasks, sees a hit...

    "Interestingly GIGABYTE does not supply any extra power connectors for these PCIe slots, indicating that not each port might not be able to provide 75W when all are populated (e.g. seven graphics cards)." -> read it slowly, how many "nots" should there be in a single sentence?
    Reply
  • aryonoco - Friday, September 05, 2014 - link

    A server/workstation motherboard requires its own benchmarks.

    For anyone who is in the target market of this product, this review is absolutely worthless. Gaming benchmarks? Really?!
    Reply
  • colonelclaw - Friday, September 05, 2014 - link

    Thanks for the review. In the future please consider including the following benchmarks:
    3DSMax + Vray, Maya + VRay/Arnold/Renderman, Adobe Media Encoder, Adobe After Effects, Adobe Premiere Pro. Some VRay RT and Octane multi-GPU benchmarks would be a nice addition, too.
    The one thing I find slightly strange about this motherboard is the low number of memory slots. All our DP motherboards have twice as many per processor (and many are full). I wonder why Gigabyte decided on just eight?
    Reply
  • eanazag - Friday, September 05, 2014 - link

    I think the point of making the board where the CPUs line up was to allow better airflow to the add-on cards (GPUs). Lining up the CPUs didn't allow for the normal # of mem. slots.

    In actuality I would think most buyers of this board would just go with one CPU. It allows for the video compute cards to have more juice from the ATX 24 pin. I have bought 2P boards for one proc because of the peripheral items like slots, networking, RAID, and built-in network management. When you look at non-server boards and add the pricing of add-on management it starts to make you consider the server/workstation board.
    Reply
  • The Von Matrices - Friday, September 05, 2014 - link

    The reason for this is because Gigabyte chose to use square ILM sockets.

    For those who do not know, LGA2011 has two sockets - square ILM and narrow ILM. Most consumer motherboards use the square ILM sockets; you only see the narrow ILM sockets in servers. The narrow ILM socket is obviously narrower, which allows more memory slots on the same size motherboard. The disadvantage of the narrow ILM socket is that the memory slots are so close to the processor that large and quiet tower heatsinks can't be used since they cover the closest memory slots.

    Basically, when designing a dual LGA2011 SSI CEB motherboard, you have to choose two of the three following features:

    16 DIMM slots
    7 PCIe slots
    Square ILM sockets

    If you use 16 DIMMs and 7 PCIe slots, you have to use narrow ILM sockets, which eliminates the use of tower heatsinks.

    If you use 7 PCIe slots and square ILM sockets (like Gigabyte did) you now can use tower heatsinks. However, then there is only enough space on the motherboard for 8 or 12 DIMM slots.

    If you use 16 DIMMs and square ILM sockets, there's only enough space on the motherboard for 6 PCIe slots.
    Reply
  • joannecdinkins - Friday, September 05, 2014 - link

    just as Larry answered I didnt even know that people able to get paid $6104 in a few weeks on the internet .
    go to this site>>>>> paygazette.ℭOM
    Reply
  • KAlmquist - Friday, September 05, 2014 - link

    One thing that's not addressed in this review is whether it makes sense to run Windows 7 on a multi-socket system. The benchmarks showing dual E5-2687W processors running significantly slower than a single E5-2687W suggest that Windows 7 is making bad scheduling decisions. Would Windows Server or Linux do better?

    Let me explain my point about scheduling decisions in a bit more detail. If you run a piece of software on a system with a single processor, the operating system has to make decisions about when to run each thread, and which core to run it on. Now switch to a system which is similar to the first system except that it has two processors. If the operating system simply ignores the second processor, and makes the same scheduling decisions as it made on the first system, then the performance of the two systems should be quite close. I'm not suggesting that this approach to scheduling on a dual processors system makes sense, but it is a possible approach. So if the approach to scheduling used by Windows 7 results in even worse performance, we know that Windows 7 could do better. In fairness to Microsoft, Windows 7 is not marketed as a server OS, so maybe it shouldn't be expected to perform well on server hardware.
    Reply
  • tuxRoller - Saturday, September 06, 2014 - link

    I found a reference on serverfault that says windows ultimate 7 supports up to two sockets. How well it is supported is another matter...than anandtech seems uninterested in investigating. Reply
  • Kevin G - Sunday, September 07, 2014 - link

    The scheduler is the same between Windows 7 Pro and Windows 2008R2. The difference is in the configuration defaults.

    The more defining factor between a single and dual socket system in Windows is how it handles NUMA. Modern chips all have their own memory controller and things get complicated when a thread running on one processor requests data off the memory in another socket. That'll cause a slow down due to the additional latencies involved. Ideally the scheduler will run a thread on the same socket as the memory it accesses but for large data sets, this isn't possible or practical to continually context switch were a thread runs. This is why having a system with the same clock speed, the same number of cores and the same amount of memory will generally be faster on a single socket than a dual.
    Reply
  • Filiprino - Sunday, September 07, 2014 - link

    What about pinning a game to a socket?
    You should get the same performance as the equivalent single socket CPU, same frequency.
    Reply
  • Gonemad - Monday, September 08, 2014 - link

    I believe the point of the reviewer was to prove this motherboard is worthless to gaming enthusiasts, that will get better bang for buck elsewhere.

    It could be a tremendous gaming platform back in the single-core day, as it was the case. I knew a couple of people that built gaming machines out of server parts exactly to get the benefits of a dual-core system, even when the applications were not built for it, but just forcing Windows to run on the 2nd chip, while the game would run on the 1st. At least, that was the consensus back then, however wrong that may be.

    Today, it is proved to be a moot point, since you get more performance out of a single 8-core chip, with shared cache and optmizations, than relying on true dual-chip design to get your cores.
    Applications that already relied on multiple cores just took a leap ahead, having so many of them to populate, which was evident on the compression software's benchs.

    I believe that this whole benchmark is far from useless; it proves that this use case is totally incorrect for the purpose of gaming. On the other hand, the server platform is an excellent number cruncher, with the reliability to match it, but it can take some gaming on the side dish without effort. It won't take 'best bang for buck' lists by storm, but it won't force the buyer to build a machine just for gaming.
    Reply

Log in

Don't have an account? Sign up now