Asus Creates Upgradable Graphics Card


Recommended Posts

Source: http://www.tomshardware.com/news/asus-grap...-card,5324.html

Taipei (Taiwan) - The vision of upgradeable graphic cards goes back to the late 1990s, when Micron Technology was experimenting with removable sockets. In 2006, both MSI and Gigabyte showcased upgradeable graphic cards, but their concepts, which were based on GeForce Go MXM boards, never took off. Earlier this year, Asus introduced a single board with three MXM slots for ATI Mobility Radeon 3850 or 3870 cards (upgradeable with future parts), and has now unveiled its single-MXM product.

Called Splendid HD 3850M, this card doesn?t look like anything special, until you remove the dual-slot cooler. What you can see then is a MXM card with a RV670 chip and 512 MB of memory attached to the PCB that contains the Splendid HD video processor: The video processor features 12-bit gamma correction, 7-region color enhancement and dynamic contrast engine.

The Graphics chip is clocked at 668 MHz while the 512 MB GDDR3 memory operates at 828 MHz DDR (1.65 GT/s). According to Asus, this MXM card will score around 600 3DMarks (3DMark06) more than ATI?s own reference design. But what makes the different, is the fact that this product is significantly shorter than the Radeon 3850 or 3870 ATI reference design.

The Asus Trinity card has three MXM slots. The company is currently selling the card with three modules based on the Radeon HD 3850.

Thanks to a modular design, you will be able to upgrade to upcoming MXM modules, including ATI?s RV770 and RV870 chips (Radeon HD 4800, 5800 series). Interestingly, there should be no issue to put a Nvidia-GPU based MXM module onto this card, since there is no limiting logic.

Using this design, you can imagine a future where users will upgrade their graphics experience simply by buying a small module. If you would have to buy just the GPU and memory, this approach would actually lead to less money being spent, since you don?t need to buy the complete card over and over again.

This new line of products appears to be much more than an engineering exercise. We hope to see future designs incorporating HDMI-in on graphics cards too, just like on the much anticipated professional sound card, Xonar AV1 .

Asus is now on track of doing something new, something that can put them clearly ahead of the competition.

Cool.

Next step: Getting everything universal enough that you don't need a full graphics card, but can just drop a GPU on your mobo (I would assume that by this point the GPU would be able to communicate quickly enough to just use your main memory)

Great idea, Next to the CPU you drop your choice of GPU.

but can just drop a GPU on your mobo (I would assume that by this point the GPU would be able to communicate quickly enough to just use your main memory)

Sorry to nit pick, but isn't Graphics RAM a heap faster than normal RAM. :p And also a heap more expensive..

Cool.

Next step: Getting everything universal enough that you don't need a full graphics card, but can just drop a GPU on your mobo (I would assume that by this point the GPU would be able to communicate quickly enough to just use your main memory)

With memory controllers moving onto the CPU for Intel soon (already there for AMD obviously), this would require the CPU to be part of the process of giving access to the RAM. Maybe this would be a good fallback mode if someone doesn't have a graphic memory MXM module installed, but they need to offer a slot for graphic RAM too IMO.

With memory controllers moving onto the CPU for Intel soon (already there for AMD obviously), this would require the CPU to be part of the process of giving access to the RAM. Maybe this would be a good fallback mode if someone doesn't have a graphic memory MXM module installed, but they need to offer a slot for graphic RAM too IMO.

You're right. The fact that memory controllers are now integrated on the CPU hadn't even occured to me when I wrote that. You'd need to give the graphics card its own ram slot.

Still, better solution than a card.

This topic is now closed to further replies.
  • Recently Browsing   0 members

    • No registered users viewing this page.
  • Posts

    • Ahh, I didn't know that. It makes sense. I was thinking I would be fine with 50 hours, but 100 is close enough that I wouldn't feel like I am paying for more than I am using. I might check it out sometime.
    • Google's next-gen Tensor AI chips might be produced by Samsung, report says by Hamid Ganji Image via Google Google’s Tensor Processing Unit (TPU) is produced in collaboration with companies such as Broadcom and TSMC, but a recent report suggests that the search giant is in talks with Samsung to hand over part of the production to the Korean tech company. According to a report by The Information, citing people familiar with the matter, Google has begun talks with Samsung about using its chip manufacturing capacity for the next generation of Tensor AI chips. Google’s upcoming TPUs are reportedly codenamed “Icefish” and will be produced using Samsung's 2-nanometer process technology. Meanwhile, Samsung is expected to produce only a portion of the next-generation Tensor chips, with most of the production remaining at TSMC. The Information says the new Tensor chips are currently in the design stage and are scheduled to enter mass production in 2028. TSMC is generally considered one of the most reliable chip manufacturing partners and is trusted by tech giants such as Apple and Google. However, the growing demand for TSMC’s AI chips, combined with the company’s focus on meeting demand from AI data centers, has reportedly prompted Google to seek additional manufacturing partners for its next-generation TPUs. Besides Samsung and TSMC, Intel could also be assigned part of the production. The Information reported this week is that Google has begun talks with Intel to produce up to three million TPUs in 2028. However, this is not the first time Google has partnered with Samsung on chip production, as Samsung has previously manufactured Tensor chips for Pixel smartphones. Google’s Tensor Processing Unit is used in cloud data centers and competes with NVIDIA’s chips, which currently dominate the market. By relying on in-house chips, Google can not only reduce its dependence on third-party providers but also create new revenue opportunities. The company has already supplied its TPUs to the AI firm Anthropic.
    • Reminder that the Premium and Ultimate plans already come with a 100 hour per month limit, which you then have to pay more to increase in 15 hour blocks.
    • It does invite Linux because MS should have been improving their products all along these years instead chose to ignore the users now this year they’re making all of these improvements to their products where as Linux doesn’t have this kind of problems in fact it has gotten so good I can even play Windows games in Steam that I no longer ever needed Windows to carry on with my life, unlike you still use Windows and chose to try beat down down those who don't use Windows because they love their LG TV. (Surprisingly I actually own a 55” LED LG TV and it has been going strong for almost 14 years, longer than any Sony TV”)
    • As much as I love owning my own hardware, it's hard to argue with the value. I'm not a huge gamer, I'd actually be interested in a cheaper plan with limited monthly hours, or even a pay-by-the-hour plan.
  • Recent Achievements

    • Week One Done
      FBSPL earned a badge
      Week One Done
    • One Year In
      Jim Dugan earned a badge
      One Year In
    • One Month Later
      Tommi118 earned a badge
      One Month Later
    • One Month Later
      sjbousquet earned a badge
      One Month Later
    • Week One Done
      sjbousquet earned a badge
      Week One Done
  • Popular Contributors

    1. 1
      +primortal
      487
    2. 2
      PsYcHoKiLLa
      196
    3. 3
      +Edouard
      155
    4. 4
      Steven P.
      84
    5. 5
      ATLien_0
      69
  • Tell a friend

    Love Neowin? Tell a friend!