I've just sat down at my desk in the GPU lab here in Australia, after an exhausting but awesome Computex 2017, and one of the first things I see is that AMD has released the die shot for the Vega GPU. STAHP.
The new Vega GPU die shot teases that the new GPU is divided into 8 clusters, each packing 8 Next Generation Compute Units. Each of these new NCUs features 64 stream processors, providing a total of 4096 FP32 cores. The difference between the previous-gen compute units and the new Vega NCU ones is that Vega packs 16-bit processing with doubled throughput.
Isn't it beautiful? Radeon RX Vega launches at SIGGRAPH 2017 which takes place in California between July 30 and August 3.
We're still two months away from the launch of Radeon RX Vega, with AMD announcing at Computex 2017 that it will launch Radeon RX Vega at SIGGRAPH 2017 which takes place between July 30 and August 3.
But the bigger story for AMD right now is that their Radeon RX 400 and RX 500 series graphics cards are flying off the shelves... but why? Cryptocurrency mining, that's what. The cards are either selling out, are sold out, or they're being sold for ridiculous prices in the second hand market.
It feels like it's 2013 all over again, but between then and now ASIC hardware dedicated to cryptocurrency mining went crazy. Efficiency was king, beating consumer graphics cards... until Ethereum, a new cryptocurrency that can be mined liked a Bitcoin.
AMD is set to launch its HBM2-based Radeon RX Vega graphics card sometime in August, but Micron is now promising mass production of GDDR6 memory in early 2018, primed and ready for what I think will be a refreshed GeForce GTX series from NVIDIA.
Micron has confirmed it will be entering mass production of GDDR6 in early 2018, with NVIDIA set to use it on new graphics cards expected in 2018. Better yet, Micron added that it hit 16Gbps on GDDR5X, up from the 11Gbps found on the GDDR5X on the GeForce GTX 1080 Ti.
Micron added: "I am excited to announce that our Graphics design team in Munich has achieved 16Gbps data rates in our high speed test environment-another first for memory industry. The left picture shows the data eye opening at 16Gbps based on a critical PRBS pattern sequence, with great timing and voltage margin. The right image below shows stable data timing margin (horizontally) versus data rate (vertically), from our base sort speed of 10Gbps up to an unprecedented 16Gbps. This result is based on measurements on a meaningful sampling size of our mass production G5X silicon - not theoretical simulation data.
We strongly believe that our expertise and experience running ultra-high data rates on G5X is going to be a big advantage for driving performance in GDDR6".
AMD was slowly but surely clawing back discrete GPU market share from NVIDIA over the last 6-12 months, but according to the latest report from Jon Peddie Research, graphics board shipments dropped 29.8% quarter-over-quarter, and 19.2% year-over-year.
NVIDIA now sits with 72.5% of the discrete GPU market to itself, their highest in months. NVIDIA has saturated the market with their GeForce GTX 10 series products, as they continue to release cards - and I'm sure they still have more right around the corner as a response to Radeon RX Vega from AMD.
JPR explains: "The attach rate of AIBs to desktop PCs has declined from a high of 63% in Q1 2008 to 42% this quarter, a decrease of 9.0% from last quarter which was negative. Compared to this quarter last year it decreased 12.9% which was negative. The overall GPU shipments (integrated and discrete) is greater than desktop PC shipments due double-attach-the adding of a second (or third) AIB to a system with integrated processor graphics-and to a lesser extent, dual AIBs in performance desktop machines using either AMD's CrossFire or NVIDIA's SLI technology improved attach rate".
Computex 2017 - AMD had a tiny little whimper of Radeon RX Vega at Computex this year, revealing that it would detail Radeon RX Vega at Siggraph 2017 - which takes place between July 30 and August 3. For our entire Radeon RX Vega coverage, I have a big article on it here.
But since I'm on the ground in Taipei, I talked with some friends in the industry and we had an interesting discussion to say the least. First off, I exclusively revealed that AMD will have less than 16,000 units of Radeon RX Vega in the months after its launch, and this seems to be something that is forming into a solid fact.
Then there's the fact that AMD promised a Q2 2017 launch of Vega, and while they met it by making Radeon Vega Frontier Edition available on June 27 - that's the professional card, not the consumer RX Vega. On top of that, the enthusiast level cards will launch in Q3 - while the mainstream cards are supposedly being pushed back to Q4 2017 at the earliest, but it could spill into Q1 2018.
Computex 2017 - AMD took the stage at Computex 2017 to announce that Radeon RX Vega will launch at Siggraph 2017, which takes place July 30 to August 3.
AMD showed off Radeon RX Vega running Prey on Ultra settings at 4K, with two Radeon RX Vega graphics cards in CrossFire... all powered on their new ThreadRipper 16C/32T processor. We don't know anything other than the release date, but it seems AMD has missed the Q2 launch with June wrapping up in 30 days, but July 30 is only two months from now.
Will you be buying an AMD Radeon RX Vega graphics card?
Computex 2017 - This is the last of our GALAX graphics card unleashing their own single-slot GeForce GTX 1070 Katana graphics card.
In all its glory, the GALAX GeForce GTX 1070 Katana - the one and only single-slot GTX 1070 graphics card.
As you can see, the single-slot card is just that - thin.
The modified PCB from GALAX is short as well.
Computex 2017 - GALAX had its super awesome GeForce GTX 1080 Ti HOF OC Lab Edition graphics card at Computex 2017, but there was also the new GTX 1080 Ti HOF Limited Edition - you know, the one with the LCD display.
GALAX has clocked the GPU at 1645/1759MHz for base/boost, respectively.
As you can see on the left NVIDIA has forced its GEFORCE GTX branding, while the middle is graced with the LCD display.
GALAX pushes the power consumption boundaries, requiring 3 x 8-pin PCIe power connectors.
Computex 2017 - Something that surprised me at Computex this year was that GALAX showed off their new GeForce GTX 1080 Ti HOF OC Lab Edition graphics card.
All that we know about the elusive GALAX GTX 1080 Ti HOF OC Lab Edition is that it has base/boost GPU clocks of 1569/1683MHz, respectively - and that it requires triple 8-pin PCIe power connectors.
Computex 2017 - The very first graphics card that I ran towards at Computex this year was MSI's new GeForce GTX 1080 Ti Lightning Z, something we got some more hands-on time now that it was in a PC and turned on.
The beautiful Mystic Light RGB goodness comes out to play, and it really does look freakin' slick in person.
MSI requires 8+8+8 pin PCIe power connectors, but will the additional 8-pin PCIe power connector help GPU clocks that much? I doubt it.
MSI has also doubled down on the heat sink for the GTX 1080 Ti Lightning Z, too.