Why do GPUs Transmit “Upgrade Fever?” [Tech Talk]

I’m going to go out on a limb here and say that pretty much any serious PC gamer has likely experienced “upgrade fever” before. With the rapidly developing tech of today’s digital age, it’s hard not to succumb to it. After all, pretty much every year the two GPU giants Radeon (AMD) and NVidia will release a brand new line of “flagship” graphics cards, as well as very respectable mid-level and higher-end cards for consumers of all different types. Newer cards may boast higher stock clock speeds, more shader cores (CUDA or Stream), and of course more Video RAM to allow for much larger calculations to be done.

Last year, in fact, Radeon introduced the PC gaming world to the Fury and Fury X cards, which rather than using GDDR5 like most cards on the market, used something called High-Bandwidth Memory which due to its location on the die allows for considerably faster access to the card’s RAM. While HBM has a lot of potential for the future of gaming due to these faster access speeds, 2015 seems to have been the test year for it because the Fury X only boasted 4GB HBM, and while many claim that the lower VRAM is countered by the faster access speeds (e.g. memory volume is less of a bottleneck because data is processed faster), the supposed 2016 line of cards should feature at least 8GB of HBM or more, which would make them worth waiting for potentially.

Meanwhile, NVidia has a very strong offering with the GTX 980 and the 980 Ti, which while using GDDR5 technology still have very high clock speeds with the 980 Ti featuring 6GB DDR5. Radeon, on the other hand, has the 390 and 390X, both of which feature 8GB GDDR5 and very fast clocks as well.

So, the question I always face each year when I see these bevies of new cards come to the market is whether or not it’s worth upgrading or if it’s better to see what the next line of cards has first.

First, I think it’s always a good idea to take a step back and consider the options as well as your reasons for even thinking about upgrading in the first place. Do you really need a new graphics card? What’s wrong with your current one? What features would a new card have that your current card lacks? Do you need to upgrade right now, or could you wait to see what the next line of cards has to offer first?

The problem with waiting around for the next line of graphics cards to release goes back to the same old notion of “chasing the next big thing.” If you always hold out for the next product to drop, you’ll never get to enjoy the current ones – you’ll always be holding out for the upgrade. Also, even if you do pick up something that’s in the current generation, you’re likely going to be less satisfied and just using it to bide your time for the next one to come, and that’s a very expensive proposition. Basically, you just have to ask yourself where this all ends?

Not too long ago, I was in this situation, being tempted by the newer lines of graphics cards. I had a very respectable R9 290 Windforce Edition card (4GB GDDR5) that I paid a pretty penny for back when the card was new. It’s actually even today a pretty solid card. Sure, it did run rather hot compared to others (despite being the “windforce” card), but it could handle most any game thrown at it at 1920×1080 or even ultra-wide with settings maxed out and the frame-rate very high. It could also even record at very high settings and not really have much of an issue. The problem was that the card would run hot, would experience artificial “throttling” spikes when this occurred in order to lower the temps, and as newer games released that I wanted to record at high settings (often my YouTube videos are as high as 1440p60 these days), I realized that the GPU wasn’t keeping up quite as much. Being able to play games very smoothly is very important to me, and having very nice video quality is also very important. Moreover, I don’t mind further future-proofing a rig so that upgrades will be less necessary.

When I was thinking over my options, I was enticed at the time by two strong contenders: Getting one of the new R9 390X cards (more shader cores than the 290, 8GB DDR5, much better stock cooling, and slightly lower power draw) or maybe getting a Fury or Fury X. I really didn’t want to break the bank, though, and felt that $400 would be a good spending point for me at the time, and it was only after a great deal of research that I settled on Gigabyte’s R9 390X G1 Gaming Edition (8GB) card.

While poking around, doing a bit of research for this article, though, I became aware of another new technology that we may very well be seeing this year in upcoming GPUs: GDDR5X memory. This newer memory format is based on the current GDDR5 standard, but has much faster prefetch technology which, in layman’s terms, translates to effectively doubling the memory bandwidth of the card. Depending on the amount of GDDR5X memory present, this could result in performance that is very close to HBM1 or HBM2 standards, but on cheaper technology. There’s really little doubt though that High-Bandwidth Memory will likely be the eventual future of graphics performance (and even CPUs at some point, if the rumors of AMD’s Zen line are to be believed), but upgrade branches of older technology, like GDDR5X, will continue to breathe life into existing architecture and provide developmental alternatives.

One rumor I read suggested that the Fury X II (codename, of course) will use HBM 2.0, but there’s a strong likelihood that the R9 490X (unconfirmed, though likely) would make use of GGDR5X and likely have more physical memory available to it. The Fury X II would cost more, of course, but the performance would likely be pretty close in most regards to the 490X. The 490X in this case would have the edge of having more memory to play with while benefiting from much faster clock speeds on the RAM than the 290 and 390/X series had, but the trade-off would be greater power-consumption due to the lower usage of HBM.

Some say, though, that the new GDDR5X technology may slow the adoption of HBM. After all, in one example of using 384-bit memory with it, the data rate would be around 660 GB/s, while at 512-bit it would be around 800 GB/s. This would put GDDR5X faster than HBM 1.0 and very close to HBM 2.0, so depending on the development and production costs, it might be more enticing for card manufacturers to prefer GDDR5X over HBM all things considered.

But, these are things we don’t need to worry about for a while,  yet. More importantly, when whatever the new line of cards is comes out, the cost of the current generation will go down. If you don’t care to get the 2016 model, you may decide it’s better to get a higher-end version of the 2015 line of cards. The great thing about all the third-party manufacturers is that they all work hard to eek out the absolute most they can from the cards they work with, and they are always trying to one-up each other. By the time the summer rolls around, you may see variations of the 390 and 390X cards with more memory or better cooling. On NVidia’s side of things, the same is likely the case.

So, keep some of these things in mind when you are toying with the idea of upgrading. It’s not always the better idea on your wallet or even for your sanity to chase the newest thing. Besides, money aside, newer technology always comes with “early adopter issues” that would be ironed out later on. Then again…Well, if you waited for the 2016 cards and then decide to wait for the 2017 ones which may feature upgrades to the architecture, you’ll be waiting a long time. By 2017 you’ll hear about newer tech coming in 2018, and you’ll want to wait even more 😛

Get out there and enjoy what you have and consider all the great cards on the market. If you do want to upgrade, upgrade for you! Have fun, and enjoy the great things that are out there and try not to get too bogged down in all the insanity. 🙂