Part 1: When One GPU Wasn’t Enough
There was a time when the ultimate enthusiast PC did not necessarily have the fastest graphics card.
It had two.
Sometimes three.
And, for a brief and particularly ambitious period, even four graphics processors could find their way into a single desktop.
Today, the idea of putting multiple graphics cards into a PC can seem excessive. Modern GPUs are enormously powerful on their own, and for most applications there is little reason to divide a workload between several of them. But there was a period when SLI and CrossFire represented something genuinely exciting: the possibility that graphics performance could scale not just by making one processor faster, but by putting more processors to work.
That idea is what this collection is about.
More Than a Faster Graphics Card
There was a particular feeling that came from seeing two graphics cards installed side by side.
It wasn’t just about the benchmark result.
Two identical cards, connected together and working on the same workload, made the architecture of the computer visible. The system was no longer simply a machine with a powerful graphics processor. It was an experiment in parallel computing sitting on the motherboard.
And that made multi-GPU systems interesting even when they were not the fastest machines available.
SLI and CrossFire were products of a particular era in PC hardware — an era when enthusiasts were willing to experiment with motherboards offering multiple full-length PCI Express slots, bridge connectors, increasingly powerful power supplies and carefully matched CPUs.
The computer itself became an expression of the idea that one processor did not have to do everything alone.
A Collection Rather Than a Single Build
Our multi-GPU systems span several generations of this idea.
They don’t all represent the same approach, and that is precisely why they belong together.
The collection begins with Hybrid CrossFire, where an ECS A780GMA platform can combine its integrated graphics with an external Radeon HD 3450.
From there, we move into the era of serious multi-GPU enthusiast hardware, including a CrossFireX system built around an ASUS motherboard and four Radeon GPUs: two HD 3870 ICEQ cards and two single-slot Sapphire HD 3850s.
We also have more conventional two-card systems, including two Palit GeForce 9600 GTs in SLI, installed on an ASUS M2N32-SLI Deluxe with an AMD Phenom X2 550.
Then comes the GTS 450 SLI system — a configuration that has a more personal story, because these are two cards that ended up together through the way the hardware was actually acquired rather than because we set out to build a historically perfect pair.
And further along the AMD side, we have a 990FX platform with an FX-8350 and Radeon R9 280 CrossFire.
Taken together, these machines show something more interesting than a collection of old graphics cards.
They show the evolution of an idea.
Matching the Whole Machine
One of the things we care about with these systems is that the GPUs are not considered in isolation.
A multi-GPU system is a platform.
The motherboard matters because its PCI Express layout determines what can actually be built. The CPU matters because it needs to make sense for the graphics generation. Memory, chipset, cooling and power delivery all belong to the same equation.
This is why the Phenom X2 550 belongs with the 9600 GT SLI system rather than simply replacing it with a much newer processor because it happens to be compatible.
The same principle applies to the 990FX and FX-8350 system.
We are not trying to create the fastest possible retro machine by mixing components from unrelated generations. We are trying to reconstruct the balance that made sense at the time.
That balance is part of the history.
When Four GPUs Became the Point
The four-card CrossFireX system is perhaps the clearest expression of this philosophy.
Two Radeon HD 3870 ICEQ cards sit alongside two single-slot Sapphire HD 3850s, on an ASUS motherboard designed for an enthusiast platform of the period.
The motherboard itself is part of the story. Its unusual approach to memory cooling is the kind of detail that makes these old enthusiast boards fascinating: the platform was designed with serious performance hardware in mind, not simply as a generic motherboard with several expansion slots.
The resulting system feels almost like a physical demonstration of the CrossFireX concept.
And that is especially interesting because its configuration is remarkably close in spirit to the multi-GPU systems AMD was demonstrating during the CrossFireX era.
Four GPUs in one desktop was never going to be the normal solution.
That was the point.
It represented the enthusiast tendency to ask not “What is enough?” but “What happens if we go further?”
The Power Question
There is another detail that matters to us.
We are not building these systems around enormous 1000-watt-class power supplies simply because four GPUs can theoretically consume enormous amounts of power.
The goal is not to create a laboratory monster.
The goal is to build believable systems.
Period-appropriate hardware. Sensible CPUs. Appropriate motherboards. Reasonable power requirements.
That makes the machines more interesting because they remain recognisable as PCs that an enthusiast could actually have built and used.
The spectacle comes from the architecture, not from throwing unlimited electricity at it.
From Experiment to Enthusiast Standard
The story becomes particularly interesting when we compare the different generations.
Hybrid CrossFire represents the beginning of the idea in a relatively modest form: use the graphics resources already present in the motherboard and combine them with a discrete GPU.
CrossFireX takes the concept much further.
SLI demonstrates another interpretation of the same basic philosophy, from two relatively mainstream GeForce cards working together to more ambitious enthusiast configurations.
The 990FX/R9 280 system represents a later and much more mature version of the idea, where multi-GPU enthusiast platforms were already well established.
But the story doesn’t end with increasingly powerful multi-GPU systems.
It ends when the idea itself begins to lose its importance.
When One GPU Became Enough
As individual GPUs became more powerful, the reason for installing two of them became increasingly difficult to justify.
Multi-GPU performance depended heavily on software support, workload characteristics and scaling. A second GPU could provide impressive gains in the right circumstances, but it could also provide much less than expected.
Eventually, a single high-end GPU became the simpler and often better solution.
That is an important part of this collection.
We are not trying to argue that SLI or CrossFire was always better.
It wasn’t.
What makes the technology worth preserving is that it represents a different way of thinking about computing.
For a while, the answer to “How do we make this faster?” was not necessarily “Build a faster processor.”
Sometimes the answer was:
Add another one.
Why Keep These Systems?
There are many faster computers available today.
That isn’t the reason we keep these.
These machines capture a particular moment in PC history when parallel graphics processing was something enthusiasts could see, touch and experiment with.
The bridges, the multiple PCIe slots, the cooling arrangements, the power connectors and the physical presence of several graphics cards all made the architecture tangible.
You could open the case and see the concept.
And that is what makes this collection interesting to us.
It is not a museum of the fastest GPUs.
It is a collection of machines built around a question:
What happens when one GPU isn’t enough?