ATI HyperMemory Technology For Graphics - Preview



Back in December 2004, NVIDIA introduced their latest evolution in the budget graphics card market with the release of TurboCache technology.  The focus with TurboCache was both system resource utilization and economics, in an effort to deliver a quality business-class/casual gaming video card at the lowest possible price point.  NVIDIA wanted to achieve this while still offering many of the features of the higher-end 6600 and 6800 series cards but at a much lower price tag.  Through the use of software and hardware, a new TurboCache designed video card could allocate a chunk of system memory to be used concurrently with the video card's local memory.  This meant that even though the card had only 16MB or 32MB of on-board memory, it could allocate a full 128MB Frame Buffer with a peak bandwidth of 13.6GB/s.  Thanks to the 8GB/sec (4GB bi-directional) bandwidth offered by PCI Express Graphics links to system resources, new architectural options were now available that could not fully be exploited with the aging AGP bus.

Today, ATI has released their counter-attack to TurboCache with the release of their HyperMemory Technology.  On paper, HyperMemory is similar in design to TurboCache, using a combination of local memory and system memory to offer advanced performance in a business-class card.  However of course, ATi is claiming to one-up their competition on all fronts, including price, performance and features.  Also, unlike NVIDIA's TurboCache launch, today's HyperMemory technology announcment is what we affectionately like to refer to as a "paper launch".  This means we haven't had the luxury of working directly with HyperMemory, nonetheless, we have been provided a fair amount of information to give us a feel for what to expect when the cards arrive to market.

Details of ATI's HyperMemory Configurations
What We Know So Far

ATI's initial offerings consist of two graphics cards, the Radeon X300 SE 128MB HyperMemory and the Radeon X300 SE 256MB HyperMemory.  Each utilize the RV370 core clocked at 325MHz and comes with local memory compliments of 32MB and 128MB.  Additionally, each card will have a 64-Bit memory interface with local memory clocked at 300MHz.  This in turn, results in both models topping out equally at 4.8GB/s bandwidth with local memory and 12.8GB/s when combined with system memory.  Both models are PCI Express 16X native and of course are not available in AGP variants obviously.

When we compare the Radeon X300 SE HyperMemory cards to the latest TurboCache offerings, we see that ATI is bringing a seemingly more balanced approach to its speed, feeds and feature set.  For example, each card comes with a 64-Bit memory bus, where the TurboCache models come in 32-Bit and 64-Bit flavors.  Additionally, there is more consistency in the clockspeeds, meaning the peak memory bandwidth with each card is the same whether you select the 128MB or 256MB model Radeon.  Each TurboCache ready GeForce 6200 has either varying clockspeeds or memory bus, so each card offers a different peak memory bandwidth.

Lastly, the GeForce 6200 TurboCache models come with a recommended minimum requirement of 512MB of system memory, where the Radeon HyperMemory models require 256MB minimum.  Granted, most of us agree that 512MB should be the minimum memory allotment for most modern systems, but in a workstation environment, it is slightly more likely to find systems running 256MB comfortably. 

You could look at the above offerings from both camps from a couple of perspectives.  ATi has fewer build-out options of their technology that they are bringing to market or NVIDIA's approach is slightly more confusing with respect to performance expectations in a given feature set or price point.  In either case, it's easy to see the similarities as well as the differences in each company's approach to shared memory graphics solutions over PCIe.

Next, let us turn our focus to ATi's HyperMemory technology and highlight its features and benefits compared to traditional graphics card configurations.

Related content