I'd like to contribute but before i choose there are some specs i'd really like to know. So far i see 3 proposed models, the Udoo basic, Advanced and Ultra all having different CPUs and RAM. Could you list the CPU models used and the difference between the 3? Can the ram for the Ultra be upgraded? The reason why the CPU model is important to me is because of the integrated GPU. I plan to run code on both the CPU and GPU but not all intel GPUs support openCL as there is a minimum hardware core and shader count for the GPU for it to run. Intel GPUs have the best openCL performance compared to nvidia and AMD GPUs based on performance per GPU core (not shader). Each GPU core for any brand of GPU consists of a bunch of shaders, units and stuff. To me the architecture is important. How will the cluster work? Will it be connected via some interconnect? Stackable on top one another? Have all the features the ultra has? I wish you could shed some more light on the details of the Udoo x86.
Intel processor specification links: Basic: http://ark.intel.com/products/92124/Intel-Atom-Processor-x5-E8000-2M-Cache-up-to-2_00-GHz Advanced: http://ark.intel.com/products/91831/Intel-Celeron-Processor-N3160-2M-Cache-up-to-2_24-GHz Ultra: http://ark.intel.com/products/91830/Intel-Pentium-Processor-N3710-2M-Cache-up-to-2_56-GHz
Thanks Seems like they all use the same architecture using intel atoms based on a recent iseries, not sure about the basic since theres no datasheet but the advanced and ultra GPUs support openCL which is nice. Will the Ultra's ram be upgradeable? How will the clustering work?
The memory is integrated into the board and not upgradeable, additionally, the UDOO X86 Ultra comes with 8 GB, which is the limit for the Intel Braswell architecture. Intel recently released beta drivers for Vulkan API that should offer improvements for power efficient CPU/GPU usage - https://software.intel.com/en-us/bl...204404-graphics-driver-for-windows-78110-1540 Intel also highlighted some of the advantages at last years GDC -
in the datasheet intel says it supports openCL 1.2 which is great. Im wondering if it would also be able to overclock the CPU or change some settings. For intel iseries CPUs theres a program called throttlestop that lets you change various parameters like overclocking the turbo frequency, changing the ratio of power between the CPU and GPU, and for some CPUs with variable TDP it also lets you change it too. So aside from RAM and CPU what exactly differentiates the 3 Udoo x86? Since this is x86 does it mean any linux compiled for x86 will run or will you port a version of linux like with the first udoo variants? Is there a picture diagram like the earlier Udoos have? I think everyone is curious to see how it will look like. Is it possible to introduce compute modules instead like with some ARM boards where you just have the RAM and CPU on a SoDIMM looking card that plugs into a board accepting multiple? Or perhaps an addon board where you plug these modules in that connects via PCIe or some other form of IO if bus to bus connection isnt possible. I did like the udoo quad in that it was quite flexible in IO that it offers and one of the few ARM boards with gigabit NIC and arduino combined with a quad core ARM A9 and 1GB of ram, more than what other similar boards had at that time.
The Braswell range of Intel Atom and Celeron CPUs don't have Turbo Boost, and instead they use a feature called Burst Mode, where the processor is put into certain states dependent on various factors like thermals and potential power savings (race to idle - decides if it's more efficient to run briefly at peak speed and stay in a deep idle state for longer - https://en.wikipedia.org/wiki/Dynamic_frequency_scaling). Apart from the CPU and RAM, there are no differentiations between the three UDOO X86 models. PCIe x2 via a M.2 socket will be added when the $500k stretch goal is reached, so spread the word about the KS campaign I'm currently in the very early stages of designing separate enclosure modules that plug into each other like Lego, so hopefully I will be able to drop in a module containing a Nvidia GTX 750 Ti. Though, an important thing to keep in mind is the PCIe interface is two-lanes and not sixteen! If an application uses a lot of texture memory and races through the graphics cards frame buffer, the application may experience problems.
Am really looking forward to that PCIe port. Are you saying that you can use mini PCIe cards and it will be 2 PCIe lanes? Is it possible to have at least 2 ports so i could attach a GPU to one and wifi card or some other card to it? I still have my egpu adapter with both expresscard and miniPCIe interfaces. im not exactly looking at using this for graphics but rather it could be part of a compute system or routing. If there is a cluster variant for the ultra and you can stack them they could make for an interesting router as you could add NICs by stacking, adding more computing power as well as bandwidth. If you could also bond the NICs in this configuration that would be great. If this is all possible there would be so many potential uses. So the cluster variant is it different or you can just buy multiple Udoos and stack them? The Arduino IOs arent fast enough for clustering that gigabit ethernet is many times faster but i could be wrong if the new arduinos are different.
The stretch goal is to route the PCIe 2x bus to one of the M.2 sockets (formerly known as the Next Generation Form Factor / NGFF) - https://en.wikipedia.org/wiki/M.2 The M.2 Key B socket will be the one connected to the PCIe bus, so providing you're able to obtain an adapter for Key B to the format you require - PCIe slot, ExpressCard, or Mini PCI, then you should be able to use your current eGPU adapter. Also, ExpressCard is PCIe x1, so performance wise, your current set-up should provide a rough idea of what an eGPU can bring to the table. The other M.2 slot is available for WiFi expansion cards, which I believe uses the UART bus to interface with the host, and there's a Gigabit Ethernet port on the UDOO X86 that's already connected to the PCIe bus.
Great, i'd like to pledge. Is there a cluster kit for 4 Ultras instead? In the mean time i need to prepare 4 PCIe adapters. If you reach the stretch goal what would be the power limits on the M.2 socket? Obviously i would be powering GPUs externally but PCIe lets you connect many things.
As it stands you'd have to work out a pledge level equivalent to 4x UDOO Ultra plus postage, and make sure the team know that at the end of the campaign from recent information in the discussion area on Kickstarter.
so you're saying i would have to contribute 4x UDOO ultra + 4x postage? I dont see a way to stack them directly above each other like shields for cluster use. So clustering the UDOO can only be done with networking?
Maybe worth putting a post on the Kickstarter page asking about a kit that allows for alternative clustering options.
Well if you could stack them the arduino bus just isnt fast enough. If you could stack them via the PCIe bus or some sort of high speed bus it would be worth it. Even usb3 stacking could work if you could achieve the full bandwidth. As long as the resources could be shared (such as combining 4 of them to get a 4 port 16 core router) would be impressive.
