Remembering Intel 740

With Intel attempting to get into 3D gaming graphics again, Custom PC’s Ben Hardwidge looks at the time it failed to take on 3dfx in the late 1990s.

Back in the late 1990s, I worked at a computer shop in Derby, where we sold components over the counter, while pointing to a sign that said ‘components are sold on the basis that the customer is competent to fit it themselves’. There were often compatibility issues between components, but there were two cards I’d always try to steer customers away from, as they nearly always came back to the shop, accompanied by a tired, angry face and colourful vocabulary.

One was a PCI soft modem that required an MMX CPU and refused to cooperate with Freeserve, Dixons’ free ISP that was taking the UK by storm. The other was Express 3D graphics card, based on Intel’s 740 gaming chip.

This was before Nvidia had coined the term ‘GPU’ for its first GeForce cards, which could take the burden of transform and lighting calculations away from the CPU. The CPU was still expected to do a fair bit of work in the 3D pipeline, but you bought a 3D card to speed up the process and make games look much smoother than software rendering. 

However, unlike the 3dfx Voodoo and VideoLogic PowerVR cards at the time, which required a 2D card to output to a monitor, the i740 wasn’t a sole 3D card – it could function as a 2D and a 3D card in one unit, and at £30 it was also cheap. You can see why people were drawn to it.

Another factor in its popularity was being made by Intel; thanks to the company’s relentless marketing campaigns, this meant people assumed it would just work without problems. It also used the brand-new Accelerated Graphics Port (AGP) interface, which people often assumed meant it would be faster than the PCI-based 3D accelerator cards.

The problem for us was that people who wanted cheap graphics cards usually also wanted cheap CPUs and motherboards, which meant going for an AMD K6 or Cyrix 6×86 CPU and a non-Intel motherboard chipset. The i740 didn’t like the AGP implementation on non-Intel chipsets very much, and it particularly didn’t like the ALi Aladdin chipset on which our most popular Super Socket 7 motherboards were based. 

If you wanted the i740 to run properly, you really needed a Pentium II CPU and Intel 440LX or 440BX motherboard, and they were expensive. Then, once you’d paired your cheap graphics card with your expensive foundation gear, the i740 wasn’t actually that great, with comparably poor performance and still a load of compatibility issues. However, it had some interesting tech and history behind it that’s worth revisiting.

Aerospace beginnings

Intel didn’t have much in the way of graphics tech in the 1990s, but it had spotted a big market for 3D acceleration. The ATX motherboards for its latest Pentium II CPUs also came with an AGP slot, and a 3D AGP graphics card could potentially encourage people to upgrade (more on this later).

With little 3D accelerator expertise in house, Intel teamed up with US aerospace company Lockheed Martin to develop a consumer graphics card. That might seem a bit left field, but Lockheed Martin had acquired a variety of assets through various mergers and takeovers. In 1993, GE Aerospace was sold to Martin Marietta, and in 1995, Martin Marietta merged with Lockheed to form Lockheed Martin. 

GE Aerospace was a division of General Electric, and its main business was providing systems and electronic gear to the aerospace and military industries, including simulators. In 1994, it started to branch out, working with Sega to produce the hardware for its Model 2 arcade machines, including 3D graphics tech for texture-mapped polygons and texture filtering. It was used for titles such as Daytona USA and Virtua Fighter 2. 

In 1995, Lockheed Martin created a spin-off dedicated to consumer 3D graphics tech called Real3D, mostly using employees from GE Aerospace. Real3D worked with Sega on the 3D graphics hardware in its Model 3 cabinet, which was released in 1996, and then later began working with Intel to produce a consumer 3D graphics card, codenamed ‘Auburn’, which would become the 740.

An AGP showcase?

Intel had clear aims for the i740 when it was released in 1998 – it needed to be cheap and it needed to showcase the new AGP interface featured on the latest Pentium II motherboards. AGP had huge potential. 

Although AGP was mainly based on the existing PCI interface, it had a direct connection to the CPU, as opposed to sharing the PCI bus with other cards. This not only freed up bandwidth, but also meant the AGP bus could run at a higher clock speed than the PCI bus.

Real3D’s PCI i740 card was often faster than the AGP equivalent, as it didn’t rely on system memory. Photo credit: vgamuseum.ru

Another one of its benefits was sideband addressing via a dedicated bus, meaning that all the usual address/data lines could be used solely for data throughput rather than both addressing and data functions, with the sideband bus handling address requests. 

This massively increased the speed at which an AGP card could read from system memory compared with a PCI card, and meant an AGP card could practically use system memory as well as its on-board memory. You may remember the ‘AGP aperture’ setting in old motherboard BIOS screens – that was the amount of system memory you could allocate to your graphics card.

Most 3D cards didn’t rely on this feature, instead being piled with fast on-board memory to maximise performance, but Intel decided to go all out on it with the i740. The result was a card that only used its on-board memory as a frame buffer, with textures being stored in system memory. 

This meant Intel could save money on memory (the cheapest i740 cards only came with 2MB compared to 8MB on the cheapest Voodoo2 cards), while also ensuring the cards required the new AGP interface.

The first problem, of course, was that using system memory and its interface wasn’t anywhere near as fast as using on-board graphics memory. The other problem was that the need for the graphics card to constantly access system memory ended up starving the CPU of memory bandwidth. 

That was a big problem at a time when the CPU was still doing a fair bit of the work in the 3D pipeline. The growing use of larger textures in 3D games to improve detail made the situation even worse. What’s more, as I mentioned earlier, the AGP implementations on most Super Socket 7 motherboards just weren’t designed with a card such as the i740 in mind.

It also didn’t help that some board makers (including Real3D under the Starfighter brand) started making PCI versions of the i740 with a bridge chip and more on-board memory, and these cards were usually faster than the AGP equivalents, as they didn’t rely on system memory for texture storage. 

Curtains for the i740

What seems bizarre now is that, at the time, I remember a lot of discussion before the launch about how Intel’s work with Real3D was going to result in Intel having a monopoly on 3D graphics, and putting the likes of ATi, 3dfx and VideoLogic out of business. 

Intel had access to huge silicon manufacturing facilities, it had a massive research and development budget, and it had the proven expertise of Real3D at its disposal. In reality, the i740 was soon cancelled and almost completely forgotten by the end of 1999.

Custom PC #218 out NOW!

Get your hands on the latest issue of Custom PC at your usual outlet, or online from the Raspberry Pi Press store.

You can also download a PDF of Custom PC #218 for the bargain price of £0.00.

from Hacker News https://ift.tt/3pGHi3T