- AGP on PCI-E, the nVidia way
As mentioned earlier, AGP and PCI Express are totally incompatible standards. In the beginning PCI and AGP were almost identical, just clock speeds separated them. This allowed the transition of PCI graphic chipsets onto the first AGP bus without modification. This time its totally different - in order to use the AGP graphic chipsets on the PCI-E bus, a bridge device is needed, this is where the High Speed Interconnect (or HSI for short) comes into play.
The HSI is a dedicated chipset that takes the role of translating the PCI Express x16 lane from the Northbridge and converting the signalling over to AGP 8x standards. This in terms has its good points and bad, which we will cover:
In the good column we have the fact that any AGP graphics chipset currently supporting the 8 x AGP standards can be placed on a PCI Express x16 graphics slot. This allows a seamless transition from AGP to PCI-E.
In the bad column, adding the HSI bridge to a graphics card requires extra PCB and electrical connection, causing a price premium. The HSI also require cooling, as its is transferring a 8GB/s bus to 2.1GB/s. This chip does heat up quite a bit, in fact it gets hotter than the GPU itself. The final fault is the speed restrictions. Since this is transferring PCI Express to AGP it is effectively reducing the 8GB/s bus that PCI Express uses into a 2.1GB/s bus that AGP uses. While this is isn't a problem with the AGP graphics chipsets as they wouldn't be able to take full advantage of the PCI-E bus, we hope this isn't implemented in future chipsets, as the PCI Express bus is simply going to waste if it continues to happen. We have been told by nVidia that future chips will natively support PCI Express.
With that now said, the HSI bus for the moment provides a great way to put nVidia AGP chipsets on the PCI-E bus and give the consumer some choices early in the PCI Express market introduction.
When it comes to memory, both cards once again have their own preferred style. The 5750 uses TSOP-II modules both on the front and rear which provides a total of 128MB of Samsung 3.6ns memory that allows the RAM to run quite comfortably at 500MHz.
The PCX5900 however, uses BGA memory. This type of RAM allows for a much cooler running system, and allows for a greater overclock in the long run. The modules surround the GPU in an "L" shape, which would allow a better designed cooler to cover both the memory and GPU. The modules of choice on the PCX5900 are Hynix 2.8ns modules.
- TV Out, the one thing in common
The one thing these cards share in common is their TV output system. Both have a built-in TV Encoder, the Silicon Image PHY is used on both cards to physically connect the cards to the TV circuitry.
For our overclocking, we used the tried and true Coolbits registry hack which does allow for a great range of core and RAM clock options, without using third party tools that can limit clock speeds and sometimes become more unstable.
Our PCX5750 from Gigabyte surprised us by hitting 500MHz on the core and 700MHz memory. Our PCX5900 managed to do 470MHz core and 852MHz memory that is just above the AGP 5900XT default speed of 400/800.
Find the lowest price on PCI Express Graphics Cards!