Graphics and Memory Controller Hub

At the heart of what most would like to identify as the North Bridge of the i810 chipset (in reality its the GMCH hub, there are no North/South Bridges anymore) is the Graphics and Memory Controller Hub, or GMCH for short. The i82810 GMCH supports uni-processor systems only, and is compatible with Socket-370 and SC242 (Slot-1) CPUs.

As you may be able to guess, there are two parts to the i82810 GMCH, the graphics and the memory controllers, although they are both packaged together. The graphics portion of the GMCH is essentially an i740 accelerator integrated into the chipset hub. The beauty of the integrated 2D/3D core, as we noticed in AnandTech's investigation of VIA's MVP4 chipset, is that it benefits from direct access to the system's main memory via the memory bus.

dfi_sm.jpg (12256 bytes)
Click to Enlarge

The memory bus can be explained the same way we've been addressing all of the buses in this discussion so far, a tunnel of a set width (32-bits) able to transfer only a certain amount of traffic at once (32-bits * 100MHz operating speed of the FSB * 2X AGP Operating Mode = 6400Mbits/s = 800MB/s). The memory bus is the connection between the GMCH and the system memory, and with the video integrated into the GMCH, the video core is capable of accessing the system memory at 800MB/s, much faster than the 533MB/s transfer rate of the AGP 2X specification, however less than the upcoming AGP 4X specification. The tradeoff is made with the fact that the video core in the GMCH has no local memory, meaning all of the graphics data must be stored in the system memory via the memory bus. This makes the "incredible" 800MB/s transfer rate look quite wimpy in comparison to what the big boys are capable of, making the integrated video in the GMCH not a solution for a hard core gamer.

As briefly alluded to by the above paragraph, the GMCH uses the system memory for its frame buffer and its z-buffer. It is capable of transferring between the GMCH and system memory via the memory bus at 100MHz (800MB/s) which is a general drop in performance from what we're all normally used to with AGP 2X graphics accelerators. However when AGP 2X accelerators run out of local memory to play in and run to the system memory, the playing field is essentially leveled and tipped somewhat in favor of the GMCH because of its increased transfer rates.

ms-6182_sm.jpg (12480 bytes)
Click to Enlarge

Accelerated Hub Architecture 2 flavors of GMCH
POST A COMMENT

1 Comments

View All Comments

  • xrror - Monday, December 8, 2014 - link

    It's amusing in retrospect how difficult it was for Intel to obsolete the older BX chipset. One thing people forget is the venerable 440BX was actually the hi-end server chipset - it wasn't intended to be the bread and butter Slot 1 mainstream chipset. But all the mobo makers migrated to BX since the "mainstream" LX, MX, ZX chipsets were just gimped too much.

    But if you ever wondered why BX had things like support for 1GB (!) of RAM, dual-processor, ACPI, etc that we take for granted now - that's why. It was supposed to be a server chipset ;p
    Reply

Log in

Don't have an account? Sign up now