Hardware Newsletter:
Email:

News Headlines
New Articles

Compare Prices

Motherboards
Abit
ASUS
Gigabyte
MSI
eVGA
Intel
Tyan
More...

Processors
AMD
Intel
More...

Memory
DDR
DDR2
DDR3
More...

Video Cards
ATI
eVGA
XFX
BFG
Sapphire
More...

search for lowest prices

send article   hardware newsletter   article comments (1)
NVIDIA GeForce 6200 TurboCache - PAGE 1
Terren Tong - Wednesday, December 15th, 2004

Introduction

With the transistor count climbing steadily upwards along with the amount of memory on a video card increasing, it is a challenge to supply manufacturers with sub 100$ cards in high volume, main stream solutions especially ones that tout the rich feature-sets of current generation cards. One of the goals that has been stressed by NVIDIA since the introduction of the NV40 family was the concept of a top to bottom architecture and they have done that very well so far. NVIDIA currently has GeForce 6 series parts at every price point except for the 100$ and below market but today they are unveiling the NVIDIA 6200 TurboCache which is an extension of the 6200 Lineup we covered briefly a couple months ago.

Impediments to the Discrete Video Card

Every manufacturing industry strives to reduce costs. In the case of PC manufacturers, a discrete video card is one of the first things to go in a low cost PC. A video card is a much harder sell to the computer layman when CPU clockspeed, memory size and hard drive size dominate the specifications sheet for most PCs. I cannot forsee any sort of cost reduction on the video card side to allow discrete cards to be a part of a 199$ Walmart PC but as the price starts creeping upwards, closer to the $800+ price point, it definitely becomes possible.

TurboCache

NVIDIA's TurboCache allows for a massive reduction of onboard memory on a video card by allowing the GPU to render directly to system memory - this is different than AGP texturing as that was a read only operation. Unlike integrated video where video memory is statically locked down (the amount of video memory can generally be set by the end user inside the BIOS), TurboCache dynamically uses and releases memory when needed.

How It Works

NVIDIA stated that TurboCache is the product of over two years of engineering. They have made some changes to the 3D Pipeline to account for the increased latency of accessing system memory rather than strictly local memory.


Left: A standard programmable pixel pipeline
Right: Turbo Cache changes the 3D pipeline a bit - note the addition of the MMU in particular

With a lot of the current generation games, there is something called a render target. This consists of data which is not displayed directly - an example of this is bump mapping. A bump map is a normal map that is represented as a texture but the bump map only serves to manipulate another texture to simulate depth. The bump map is not something that is seen directly on the model. Render targets take up a lot of memory space. TurboCache allows some of this information to be written to system memory instead of local memory freeing. When we asked ATI about HyperMemory, we were told that it was a strictly software solution as it could be applied to their high end cards if they wanted to. This is not the case for TurboCache as there is the addition of the MMU unit and the ability of the L2 Tex and second shader unit to access system memory through the MMU. The TC architecture tries to optimize the bandwidth used by allocating memory smartly depending on the data.

As with the regular 6200, the TurboCache edition loses out on color and Z-Compression. NVIDIA has stated that this will be noticeable especially when applying antialiasing but defended this decision as the target for the 6200 is the mainstream market who will likely not be fiddling around with advanced graphics solutions.

Technical Details & Configurations

There will be multiple configurations of the 6200 TurboCache, 16, 32 and 64MB versions are available. The amount of memory also affects the amount of bandwidth available to the 6200 - the more discrete pieces the faster it is. Most configurations will show up as 128MB to the system along with a 256MB solution.

The 6200 TC will also get faster as the connectivity between the chipset improves - there are more constraints for example on a 915 chipset compared to a nForce 4. This sheds some light onto the rumors of NVIDIA using HyperTransport as a inter-chip connection on their upcoming Pentium 4 - the increased speed between the Northbridge and GPU will directly benefit the TC architecture.


The peak bandwidth is the sum of each discrete bank of memory along with the total up and downstream throughput of the connection between the Core Logic and the GPU. On the left is the bandwidth available to an Intel 915 chipset while the right denotes a nForce 4 chipset.

The current bottleneck in the TC architecture is the inter-chip connectivity. According to NVIDIA, when the bandwidth of PCIe implementation approaches 6GB/s it will begin to saturate the memory bus and this is when higher specced memory will come in handy.

The numbers quoted by NVIDIA are peak, not sustained numbers so take some caution when comparing the bandwidth of the 6200 TC versus other solutions. The GPU will have to contend with other system resources to access memory so in operation, the bandwidth numbers are not quite as clear cut.

Memory allocated to TC is non-paged so there will never be an instance where TC gets pushed onto virtual memory on the HD.


The difference in memory usage between a 6200 TC and 6600GT by an application is denoted in the bottom row. Note that there is virtually no memory hit on the 2D Windows Desktop.

Conclusions

My P2-300 system from a few years back came with a RIVA128. The next few years had NVIDIA being very dominant on the OEM front with the RIVA, TNT and GeForce series. They won the war against 3dfx not from the retail space but because of their overwhelming presence in the OEM space. A quick glance at the cheapest systems from the likes of Dell and Gateway with a discrete video card currently shows an ATI and not a NVIDIA card occupying the video slot. The 6200 TC looks to change that as it gives NVIDIA a cost effective, and on paper, a more robust solution than the competing X300.

Turbo Cache is an interesting concept from NVIDIA and this is a part that will probably be very popular amongst OEMs and low cost PC manufacturers. It has pretty much all the requisite features from a marketing standpoint with DX9 and Shader Model 3 support like the rest of the NV40 family and production costs for the board vendors should pretty low since only a limited amount of memory is required when compared to competing solutions.

We cannot really comment on the 6200 TC from a performance standpoint as we do not yet have a card available for testing but if it can maintain fairly stable framerates it should provide a good compromise between integrated video and a more expensive entry level card. We emphasize the compromise not as a criticism but as an observation. A midrange card like the 6600GT is unequivocally the way to go for gamers on a budget but the 6200 TC allows the person buying an entry level PC a foot in the 3D doorway without breaking the bank.


Article Index

1.The Quick & Dirty on the 6600 and TurboCache

Submit our article to: diggDigg this! de.le.ciousdel.icio.us

Get updates when we publish new articles
Email Address:
(0.0223/d/nova)