Announcement

Collapse
No announcement yet.

System76 Galago UltraPro Haswell Ultrabook

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

  • #11
    Originally posted by nils_ View Post
    That's too bad. There's an online shop in Germany that sells BTO laptops built from Clevo barebones, but a poor Keyboard is a deal breaker. Backlit option seems to present for most models though.
    Yes really odd that some laptop makers don't focus on a great keyboard. IBM pretty much nailed it 15-20 years ago, how hard can it be to clone that design?

    Comment


    • #12
      Originally posted by guido12 View Post
      Can the 128 MB eDRAM actually function as a regular L4 cache for the CPU or is it only used by the GPU? Not sure if the CPU benchmarks presented can actually show its effect if it really can be used as an L4 cache.
      Yes it does :

      Unlike previous eDRAM implementations in game consoles, Crystalwell is true 4th level cache in the memory hierarchy. It acts as a victim buffer to the L3 cache, meaning anything evicted from L3 cache immediately goes into the L4 cache. Both CPU and GPU requests are cached. The cache can dynamically allocate its partitioning between CPU and GPU use. If you don?t use the GPU at all (e.g. discrete GPU installed), Crystalwell will still work on caching CPU requests. That?s right, Haswell CPUs equipped with Crystalwell effectively have a 128MB L4 cache.

      Comment


      • #13
        Originally posted by chrisb View Post
        Thanks. I keep reading that but I haven't seen any CPU tests to show how much relevant tasks actually benefit from it. Not sure how one would test it though. Maybe use a comparably clocked non-Iris Pro Haswell chip or somehow disable/enable the CPU from using the eDRAM as a cache. Do you know of any benchmarks specifically exercising the the eDRAM as an L4 cache?

        I'd be more interested in the Iris Pro based chips if it provides tangible benefits in CPU performance seeing as I don't play any PC games.

        Comment


        • #14
          Originally posted by hdas View Post
          It baffles me that most laptops don't get these basics right, namely reliable keyboard and touchpad input, even in 2013 and instead go for fancy features like touchscreen which nobody asked for.
          It is bizarre that the laptop makers seem so determined to produce laptops with "features" that many people find annoying. Intel insists that a laptop can't be called an "Ultrabook" unless it has a glossy touchscreen, and yet in a PC Pro survey the majority of people preferred a matte screen. Same thing in a Which? survey. Even with matte screens, I can't really understand the design decisions that some of these companies have - ie. "this laptop is matte, it's designed to appeal to buyers who don't like glossy surfaces, so let's make the bezel as reflective as possible!" - what on earth makes them think that a buyer who deliberately seeks out a non-reflective screen wants that screen to be surrounded by a highly reflective bezel? It makes no sense.

          People also criticise the lack of vertical space on 16:9 screens. 16:9 is good for watching video, or playing video games, but not so great for web browsing or other vertical-document oriented tasks. The "but it allows you to run two web browsers side by side" argument isn't accurate when a) most people don't do that (~60% browse at full screen), and b) your screen is small, and non-mobile web sites are designed for desktops with 1000+ pixel width displays. Try putting two web sites side-by-side on an ultrabook with "standard" 1366x768 - after accounting for scroll bars you now have about 650 pixels width for each, and most desktop sites will not render well at that size. The screen height will be something like 14cm - less than a portrait iPad - but with the big difference that a laptop is designed to be used at arm's length, whilst an iPad is designed to be held in front of your face (most iPad/Kindle users I've seen browsing/reading seem to hold the screen about 25cm away, laptop would be double that distance). It's amazing that laptop makers don't try to optimise their devices for vertical-oriented tasks, and ignore the success of the 4:3 tablet (3:4 in portrait) in web/book reading (tasks that many users spend most of their time doing), and just stick to 16:9. You'd think there'd be some variation beyond wide screens - the Google's Chromebook Pixel's 3:2 seems to be the only example of a maker trying something new, and it was widely praised by reviewers for the extra screen height - why did no other laptop designers notice this?

          Comment


          • #15
            Originally posted by guido12 View Post
            Thanks. I keep reading that but I haven't seen any CPU tests to show how much relevant tasks actually benefit from it. Not sure how one would test it though. Maybe use a comparably clocked non-Iris Pro Haswell chip or somehow disable/enable the CPU from using the eDRAM as a cache. Do you know of any benchmarks specifically exercising the the eDRAM as an L4 cache?

            I'd be more interested in the Iris Pro based chips if it provides tangible benefits in CPU performance seeing as I don't play any PC games.
            I've only read the Anandtech review, in particular look at p17 for OpenCL and p18 for generic CPU. Iris Pro beats the 4770k in OpenCL performance, sometimes doubling the performance, which is very good for a mobile part. For CPU see the graph with row Crystalwell Advantage - the estimated gain seems to be anywhere from -4.5 to 9.5 but he also says "Intel claims that with the right workload, you could see huge double digit gains". I haven't seen any benchmarks showing that, though, I'd be interested if there are any.

            Comment


            • #16
              Originally posted by chrisb View Post
              I've only read the Anandtech review, in particular look at p17 for OpenCL and p18 for generic CPU. Iris Pro beats the 4770k in OpenCL performance, sometimes doubling the performance, which is very good for a mobile part. For CPU see the graph with row Crystalwell Advantage - the estimated gain seems to be anywhere from -4.5 to 9.5 but he also says "Intel claims that with the right workload, you could see huge double digit gains". I haven't seen any benchmarks showing that, though, I'd be interested if there are any.
              Those tests don't seem to be very promising. I'm far from an expert on how CPU cache works and which tasks highly benefit from them but I wonder if someone more knowledgeable can test and/or specify what kinds of tasks would highly benefit (ie. highly utilized file server, compression, signal processing like FFTs, convolution and other mathematical operations, etc.) from a large L4 cache.

              Comment


              • #17
                Originally posted by guido12 View Post
                Those tests don't seem to be very promising. I'm far from an expert on how CPU cache works and which tasks highly benefit from them but I wonder if someone more knowledgeable can test and/or specify what kinds of tasks would highly benefit (ie. highly utilized file server, compression, signal processing like FFTs, convolution and other mathematical operations, etc.) from a large L4 cache.
                Functionally it is much the same as having higher memory bandwidth with lower latency, so apps that will perform better are the ones that are currently memory bound. The GPU is the obvious winner since it is loading textures all the time, or streaming data for GPGPU apps. The extra bandwidth is more important to the GPU when you have combined CPU+GPU because it no longer has dedicated memory, and the memory it does have is DDR3 rather than the DDR5 you get on midrange graphics cards. On the CPU side, memory bound apps are things like media transcoders, though admittedly that's not a great example any more since it's now more efficient to do transcoding on the GPU. Apps that do compression and encryption can also be memory bound, depending on exactly what is happening - for some algorithms encryption will be CPU bound, but on CPUs which can do particular algorithms in hardware (eg. AES-NI for Intel) then it's might well be memory bound. There are some benchmarks of different memory modules here - as you see going from DDR3-1333 to DDR3-2133 results in negligible gains for many tasks, the only big winner there is WinRAR (+25%), media coding is only +5% which is a surprise. At the end of the day it's difficult to predict which particular apps will benefit, that's why you have to benchmark.

                Comment


                • #18
                  Just found a huge thread covering this laptop here. The extreme keyboard problems (missing key presses etc) sound like it may have just been a bad initial batch.

                  Comment


                  • #19
                    VT-D?

                    Does anyone know if the ultrapro supports VT-D?

                    I guess the CPU and chipset both support it, but the bios doesn't have the option I guess (I don't own this laptop yet). I opened the latest CLEVO W740SU bios file (since the galago ultrapro is a branded W740SU) with AMIBCP and under the menu option of "VT-D" it said enabled under "Optimal", not entirely sure if that is definitive. Yet after asking System76, they said it didn't support it, though I'm not sure if their answer is entirely correct because they said it doesn't have an iommu... So I've been looking around online getting mixed answers everywhere yet no answers from anyone who owns one. So, help would be appreciated. Thanks.

                    Comment

                    Working...
                    X