Tuesday, 5 April 2016

Nvidia's extra-powerful Pascal cards expected in May

Nvidia GeForce GTX

Rumours about Nvidia's next series of graphics cards are hotting up, with reports suggesting the firm’s new consumer kit will be launching at Computex at the end of May. Here's what you need to know about the GPUs.
We don’t know the names of the new graphics cards themselves, but the code name for the next generation as a whole is Pascal. It’s expected that Nvidia will replace its top-tier GTX 970GTX 980 and Titan X cards in this round of updates (according to Nvidia hawks WFFCTech), with lower-end cards coming later, as the company did with the GTX 960 and GTX 950.


Pascal cards will have a much smaller fabrication process, down to 16nm (nanometres) from Maxwell GPUs' 28nm.
This isn’t unexpected; each generation of CPU and graphics hardware often gets a reduction in process size. What it does mean is that the new top-end Nvidia cards will be able to include the same number transistors while occupying less space and consuming significantly less power. Less power usually means less heat too. It’s not clear how much less power this will be, as it depends entirely on how many transistors and cores Nvidia chooses to include on its new cards.
Speaking specifically about the GTX 980 replacement, it’s expected to occupy fewer ATX power connectors, down from 12 (6+6) to eight.
image: http://static.trustedreviews.com/94/0000354f2/4656_orh616w616/NVIDIA-GeForce-GTX-950-PCB-635x489.png
Nvidia GTX 950 The reduction in process size means you can expect even more gaming laptops to feature desktop-specification graphics cards crammed inside ever-thinner chassis. Don't expect them to be any cheaper, though.

In terms of memory, the new card is expected (although not guaranteed) to get 8GB of GDDR5X memory versus the GTX 980’s 4GB of GDDR5. While the bigger number looks attractive, note the ‘X’ at the end of GDDR5X. This in itself is important, because GDDR5X memory is a significant improvement over GDDR5. It’s more power efficient, allows for more compact arrangements of memory modules and has a higher maximum throughput than GDDR5.
For those looking for something a bit meatier, there are also some rumours about the replacement to Nvidia’s flagship enthusiast card, the Titan X. Again, it’ll be manufactured on the 16nm process, and may be the first consumer-level graphics card to use HBM2 memory. This is a step up from HBM, which was seen on the AMD R9 Fury X. HBM2 is capable of a huge 256Gb/s bandwidth while also accommodating more memory in less space with lower power consumption. It’ll be pricey, though.


At this stage it’s impossible to say. Those who just shelled out for a GTX 970 or 980 probably won’t feel the extreme desire to go out and spend the same amount of money again, but those who haven’t upgraded in a few years will likely have some very good reasons to consider the latest cards.
Expect a full round-up of Nvidia's Pascal technology at Nvidia's GPU Technology Conference (GTC) this week, although actual card announcements are expected at Computex.

X80 OR 1080 – WHAT’S IN A NAME?

One of the less important uncertainties about Nvidia’s new chips is the naming convention. Up until now we’ve seen chips numbered in the thousands (8600 GT, 7800 GTX) and then the hundreds (GTX 960, GT 750). It’s obvious that Nvidia has run out of numbers in the hundreds, having reached the 9xx series, so where it goes from here is open for debate. Recently 'leaked' heat shrouds would point to Nvidia heading back into the 1000s, but with the tech taking such a big leap, you might expect Nvidia to re-brand their cards to X80, X70 instead of the slightly confusing 1080 and 1070 shown above.