ASUS and Intel are putting together a webcast that they've invited me to attend. The topic of discussion? Sandy Bridge. The webcast will air after Intel's official announcement of Sandy Bridge at 9AM PST on January 5, 2011 at CES.

The discussion will be a conversation between myself, Gary Key (former AT Motherboard Editor, current ASUS Technical Marketing Manager), and Michael Lavacot, an Intel Consumer Field Application Engineer. 

If you have any questions you'd like to see me answer on air or that you'd like me to grill ASUS and Intel on, leave them in the comments to this post and I'll do my best to get them addressed.

Of course we will also have our full review of Sandy Bridge around the same time. 

Update: Intel posted some of the videos from this webcast on its YouTube channel. I tried to answer as many of the big questions you guys asked as I could in the video or in our Sandy Bridge review

I'll add links here for more videos as they get posted:

POST A COMMENT

223 Comments

View All Comments

  • tipoo - Wednesday, December 08, 2010 - link

    What percentage of the transistors on die are dedicated to the GPU? And how much of the TDP comes from it? Reply
  • tipoo - Wednesday, December 08, 2010 - link

    Oh, and can we ever expect OpenCL on SB graphics? Reply
  • HibyPrime1 - Wednesday, December 08, 2010 - link

    OpenCL is definitely an important one.

    As a more broad question with the same intentions, do they plan to put a more comprehensive focus on GPU drivers now that they have a focus on GPU performance?
    Reply
  • ltcommanderdata - Thursday, December 09, 2010 - link

    As well, what degree of benefit do they foresee in OpenCL due to the IGP and CPU sharing L3 cache which should greatly reduce the latency and increase the bandwidth of data transfers between the CPU and IGP compared to copying data back and forth using a crossbar as in AMD's Fusion or over FSB or PCIe as in the case for nVidia IGPs or for discrete GPUs?

    Since Intel is working on OpenCL 1.1 drivers for both CPU and IGP, will we be able to see dynamic loading, where the drivers will automatically load balance between the CPU and IGP? Again the shared L3 cache should help here. And will this dynamic load balancing also extend to any discrete GPUs that are attached?

    Can the IGP stay active even if a discrete GPU is connected? The ideal use-case for a game would be DirectX/OpenGL on a discrete GPU, OpenCL for say physics on the IGP, and the CPU doing it's standard processing and maybe helping out with OpenCL as needed. Rather than the IGP being permanently disabled if a discrete GPU is connected.

    And finally, will OpenCL drivers also be made available for Arrandale's IGP? It was once reported that DirectCompute drivers would be coming for Arrandale's IGP this year, which seems increasingly unlikely now.
    Reply
  • billythefisherman - Wednesday, January 05, 2011 - link

    Ok so the chap said we can't use the IGP at the same time as a discreet graphics card but he said in the future we will be able to. Does anybody have any idea whether we're talking about Ivy Bridge or new motherboards or motherboard firmware updates here?

    Surely the motherboard has to signal to SB to turn off the IGP so if you wanted to use it purely for physics the motherboard simply tells SB it hasn't got one plugged in.

    Ok so I'm assuming its not this striaght forward but looking at Intels block diagrams the IGP looks like its plugged directly into the ring bus so surely it must be able to access memory through L3 and be able to process that memory at the same time as the other cores including the discrete graphics?

    This would seem to be a *massive* missed oportunity if this can't be achieved through a motherboard firmware update as it really could of provided massive processing power to support the CPU - who cares about AVX when you have a GPGPU on a ring bus connected to you!
    Reply
  • mlavacot - Wednesday, January 19, 2011 - link

    Hi Everyone - Sorry for the delay on this post. I actually started with some later posts so please look through those for a lot more details on various topics. Here is a quick update since the broadcast.

    We do not support OpenCL acceleration on the graphics portion of the Processor for Sandy Bridge. OpenCL would be handled by the CPU.

    The Sandy Bridge parts release so far are not intended to replace the existing X58, i7-9xx desktop platforms, but you can assume we will introduce a platform that will.

    To overclock the CPU of a ‘K’ SKU part, you must use the P67 PCH. To overclock the graphics portion of the ‘K’ SKUs, you must use the H67 PCH.

    You can use both the integrated processor graphics and discrete card graphics at the same time as multi-monitor. In this configuration, you can also take advantage of the Quick Sync capability. It you only plug in one monitor to the discrete card, Quick Sync will not be available. We might be able to change this behavior so stay tuned.

    There are no “switchable” desktop graphics solutions today; however, some solutions are in the works. The current trend for desktop switchable is to have graphics data from the add in card transfer via the PCIe bus to the processor and then out the processor graphics port. You might be able to do a poor man’s switchable solution today by just plugging both graphics outputs to the same monitor (two cables to two different input ports of the monitor). Then you use the monitor input button to switch between solutions depending on the app that you are running.

    Thanks for the questions and watching the webcast. Mike
    Reply
  • talevski - Thursday, January 06, 2011 - link

    i think that amd 880g mainbord with cpu araound 90 dolars plus some 55xx series gpu can do better in terms of encoding decoding video playback games etc. and all that without alot of money spend on inetl new socekets wich you have to trow away when they make the next cpu.So please corect me if i am wrong

    to anandtech&co
    Reply
  • ltcommanderdata - Thursday, January 06, 2011 - link

    I just wanted to thank Anand and Michael for taking a crack at answering some of my questions. The OpenCL answer seemed vague on whether OpenCL will be accelerated by the IGP or just through a x86 CPU driver so hopefully that will be clarified in a blog post as promised.

    In terms of switchable graphics on desktop, Michael did bring up some very practical considerations about transferring the display signal. I do wonder more on the ability to operate the IGP as a pure GPGPU computational device leaving the discrete GPU to handle all display/graphics related functionality. This puts the IGP to use without consideration of switch-ability since both the IGP and discrete GPU will be operating simultaneously. This presumably can be achieved in Sandy Bridge with appropriate BIOS/EFI updates to not disable the IGP when a discrete GPU is plugged in assuming the IGP has sufficient GPGPU programmability and performance to make it worthwhile.
    Reply
  • tipoo - Wednesday, January 05, 2011 - link

    I got so excited when you said my username and question! But come on, your Indian, you should know its pronounced "Tea - Pooh", lol. Awesome interview though! Reply
  • Gary Key - Wednesday, December 08, 2010 - link

    1. Can I call in for the webcast and skip the HD taping session? I am not camera friendly. Reply

Log in

Don't have an account? Sign up now