|New to 3D art?|
Want to get started?
Check my tutorials!
I work with 3D graphics, so from time to time I have to build a new PC to keep up with the ever growing hardware demands. I usually build PCs meant to last for at least 5-7 years while remaining usable. It's important to know that this is no small feat, since most computers become obsolete in about 2 years. My current system (built in 2009) is an Intel X58, which means SATA2, USB2 and DDR3 RAM. What does that mean in practical terms?
Consider the HDD (hard disk drive) is the slowest part of any given computer. It's a mechanical disk that spins at fixed speed, with a mechanical arm with read/write head that moves around the surface to access data stored on it. If you remember the ancient gramophone, that's the same mechanics. The only things that differ is that the disc spins faster (typically at 7200 RPM) and the needle is magnetic.
Basically all current HDDs and SSDs are built for SATA3 (6GB/s), so mine is running at half speed (SATA2 = 3GB/s). That alone can slow me down to half speed, no matter how fast the rest of the system may be. The same goes for USB2, where USB3 transfer speeds are many times faster. I could add USB3 support through an expansion card, but there is no way to make a SATA2 system run at SATA3 speeds.
What about DDR3? This means my RAM has 3 direct physical pathways to communicate with the system bus, while the more common DDR2 RAM has only 2. You can imagine this as information highways, where the more lanes they have, the more traffic can go through it at any given time. This is very important when it comes to 3D graphics because a scene can move a lot of data, not only from 3D models but also from high-res texture maps.
It's worth mentioning that DDR3 systems are rare because they are expensive. The vast majority of people will have DDR2 systems. How do you know which one your computer has? It's simple - DDR2 RAM comes in even amounts, while DDR3 comes in odd numbers (multiples of 3). If your RAM comes in pairs of sticks, it's DDR2. If it comes in sets of 3, then it's DDR3, and you will notice that in higher pricing.
But when it comes to 3D rendering with GPUs, the star of the show is the video card. So far CUDA and OpenCL have dominated the market, so only nVidia cards could be used. More recently technology is coming to a point where basically any video card could be used for GPU rendering, assuming they have the necessary hardware. This is measured in number of "streaming processors", or SPs for short. I have an nVidia GTX 580 with 512 SPs, and I have incredibly fast Octane GPU rendering speeds with only these few. In comparison, the GTX 580 with only 512 SPs has the computing power 10X faster than the Intel i7 quad core processor.
This week I have replaced my old GTX580 video card with a new one - the almighty GTX 980 Ti "Super Clocked" with ACX 2.0 cooler and a full metal backplate. Let's not confuse this one with the regular GTX 980 card. Note the "Ti" after the name, which indicates a "Titan"-based GPU. Nvidia has created a supercomputer based on this processor architecture, and the GeForce Titan was the very first video card to punch that much power on a single GPU. Imagine the computing power of a supercomputer on a video card inside your PC. But the Titan price tag ($1000) was out of reach for most of us.
There were other "Ti" cards before, like the GTX 780 Ti, but none of these would get even close to the performance of a Titan. Coming from that, the GTX 980 Ti was the very first consumer video cards to actually match, and even beat the Titan in benchmarks, this time for $690 bucks, which is a little more than what I've paid for my old GTX 580 about 6 years ago. You may think this is still very expensive, but remember I am thinking on long term investment. The new PC should still be significant 5-7 years from now, and this is no small feat - each decision should count.
There are some misleading factors concerning what makes a video card feature better for 3D rendering with GPU. First of all, there doesn't seem to be a direct relation between rendering speed and the number of SPs in a given card. For example, the brand new GTX 980 Ti ships with 2816 SPs, which is 5.5 times what we had with the old GTX 580, but that doesn't mean it should render 5.5 times faster. As a matter of fact, it doesn't.
The amount of video RAM (VRAM) has doubled from the GTX 580 to the GTX 980 Ti, but that won't affect rendering speeds. What this do is to allow for better handling of larger 3D scenes, or scenes with very large textures. Octane and some other GPU renderers nowadays allow using system memory in addition, or instead of VRAM for cases where the scene is larger than would fit inside VRAM alone. This may sound great, but the catch is a direct hit on rendering speeds. Scenes that fit entirely inside the VRAM alone render much faster than using system RAM.
I have ran a benchmark to compare what I get from the GPU upgrade alone. I ran the default Octane 2.4 benchmark scene in 3 different ways: Direct Light, Path Tracing and PMC using my current X58 system. Real world performance gains can be seen on the last column on the right.
What are these 3 options?
* Direct light is the only "biased" rendering method from the bunch. It greatly simplifies the math calculations, which speeds up the render at the cost of realism. In my benchmark, the GTX 980 Ti gets me render times 2.2 times faster than the GTX 580, even with 5.5 times more SPs. I am not complaining, since this is over twice the rendering speed.
* Path Tracing: this is the default Octane rendering method, and also the one I use the most. It works great for scenes where there isn't much of reflective or refractive materials. It creates reasonable realism at reasonable speeds, but certainly slower than Direct Light because light calculations are physically correct (unbiased).
* PMC: also known as "brute-force Monte Carlo", this method uses classic ray-tracing to calculate all light interactions in a physically correct way. In other words, it ray-traces the entire scene, pretty much like POV-Ray would do, but a little more optimized. Ray-tracing creates the most realistic results at the cost of the longest rendering times. It has very few parameters to configure, making it the easiest method to set up. Even though this is the slowest from the bunch, it can potentially outperform Path Tracing when it comes to scenes with lots of transparency and/or translucency, so choose wisely.
From the benchmark results, we can see that the new GTX 980 Ti can beat the GTX 580 with 90% speed gains in unbiased rendering, and over twice the speed in biased mode. Until now, this kind of performance could only be reached with a $1000 Titan, but now we can get it for $690. And when it comes to pricing, it is worth mentioning that my aging 6 years old GTX 580 was still being sold for the same price up to the point where it was no longer available in the market years later. Flagship video card prices don't come down for some reason, as they do for everything else in computer hardware.
One striking thing I've noticed during rendering is that the GTX 580 sounds like a jet plane is about to take off inside your computer, and temperatures can reach 70C during load. The GTX 980 Ti is absolutely silent during full load, and temperatures don't exceed 55C at normal 20C room temperature, and consider I had my computer case opened during tests. I assume this is thanks to the second generation 2 fans ACX cooler with the open box design, which allows air to be blown out in all directions instead of tunneled to the back as in the GTX 580 and the stock GTX 980.
With the GTX 580 design, air intake would push the air through the entire video card, and expelled outside the case through a narrow vent in the back. This would keep all the hot air outside the computer case. With the open ACX 2.0 design, air is blown out inside the case in all directions, which is both a good and a bad thing. It's good because the hot air doesn't get trapped inside a closed design, so the fans have less work to cool the card. This uses less energy and makes the card more silent. Conversely, it's never a good idea to warm up the inside of a computer case because it stresses everything around it. This open ACX 2.0 is only recommended for computer cases with optimal air flow, like this one I have - the Cooler Master HAF 922 with built-in cable management.
When it comes to power usage, the GTX 580 takes 3 power cables from the PSU: two 8-pins and one 6-pins. It's a power hog, no doubt about it. The GTX 980 Ti takes only 2 cables: one 8-pins and one 6-pins, and runs cooler at steady 55C when in full load, and half the temperature of the GTX 580 when idle (24C). This alone will cause less stress to the PSU, which translates into a more stable system that will last longer.
After these many years, I already got used to how noisy the GTX 580 used to be, and I always knew when it had finished a render (gets quiet). Coming from that, it's somehow disconcerting how silent the GTX 980 Ti is, for I can't tell when it's doing something based solely on noise - it doesn't make any. All this while crunching numbers at twice the speed of my older card, which I believe justifies the price. I claim this based on the fact that most next-gen flagship cards only bring an average of 20-30% speed gains, where this card goes way beyond that range (90-200%).
This concludes my personal review of the EVGA GTX 980 Ti 6GB SC+ ACX 2.0 video card for GPU render with Octane. Needless to say, the video card design itself is gorgeous, and the full metal backplate gives it a professional look with the usual top-notch finish I expect from EVGA flagship cards. The backplate is not only for protecting the electronics below it, but it's also part of the ACX 2.0 cooling system, serving as an extra radiator surface, and keeping the PBC from warping under its own weight. A well-thought design.
I consider this an out-of-the-ordinary release from nVidia, with a fine craftsmanship from EVGA, and the unusually high performance comparing to anything that was previously released before in this price range. Not to mention it runs cooler and silent. What else could I ask for? Highly recommended!