AMD made an interesting announcement today. They are claiming to have the world’s first hardware virtualized GPU. Dubbed the FirePro S7150 and S7150 x2, these two server GPUs are not intended as direct output devices, but are to be used to power graphics for virtualized solutions. From the announcement AMD is diving into the cloud gaming, GPU assisted cloud computing and also in GPU accelerated VDI applications.
It is said that nature abhors a vacuum and that is certainly true. Something will come along to fill the void if we let nature take its course. Unfortunately this law is a little mutated in the consumer electronics market and especially in the PC component world. Here is reads; the market cannot stand not having an “It” technology, so we much create one. It seems that the last few years we have been watching this happen.
On the 19th of January Samsung announced that they had begun mass production of their 4GB HBM 2.0 3D memory. This announcement was the starting gun for the next big GPU race. As we know both AMD and NVIDIA are racing to get viable products to the market in time for Oculus and HTC to launch their consumer version VR headsets. Up until now we have really only seen the developers’ kits and while these have been impressive they are not what most are hoping for in the final product.
AMD as a corporate entity is facing some rough times. As of their last earnings call we saw that they are still losing money and really do not have a product ready to combat this. The Rage Fury line of GPUs is doing ok in terms of sales, but as of this writing AMD has not been able to take a significant amount of market share from NVIDIA or even Intel. This is not to say that Radeon Graphics, or AMD’s APUs are bad products, it is just that they are not performing as well as the competition. In terms of the APU AMD still cannot compete with the compute power of Intel’s Core series even though the GPU side of the APU is a much better product.
AMD says it’s a VR thing now. Well ok, not really, but AMD is leveraging the increased memory bandwidth in their high-end R9 Fury cards to push both 4k and VR. They showed off the R9 Fury X dual-GPU reference design working for the first time at VRLA (Virtual Reality Los Angeles). This card will feature two 28nm Fiji GPUs plus an estimated 8GB of 2.5D HBM 1.0. The memory would be split between the two GPUs at 4GB each.
AMD is delaying the launch of their new Zen based CPUs until the end of 2016. The news comes on the heels of a rather subdued, if optimistic, earnings report in which Lisa Su, AMD CEO, focused on AMD’s commitment to innovation and highlighted improved R&D spending. It is true that AMD’s operating loss has dropped massively when compared to Q4 2014, but much of this is due to cut backs in operating costs over actual revenue. This is despite claims that AMD’s GPU and Semi-Custom sales were to thank. During 2015 we saw AMD start to pull back from other products (and product lines) to conserve money and focus on creating a compelling product. Zen is the hope for AMD’s future along with moving to a 14nm process.
Around 2013, AMD entered into an extended partnership with a group of companies to create the Heterogeneous Systems Architecture Foundation. These companies mostly ARM licensees and included Samsung, MediaTek, Texas Instruments (Ti), AMD, Imagination, Qualcomm, and even ARM themselves. The group was similar in nature to the one that AMD had with Motorola and Ti back before the Athlon processor came into existence. The partners were all working on technology and resource sharing to make programing for devices simpler. We also saw it as a chance for AMD to offset R&D costs and potentially enter into some beneficial agreements.
The average GPU is a pretty powerful computational device. The highly parallel design and efficient memory structure means that you can execute operations at a rate that puts most CPUs to shame. With the advent of Cuda and OpenCL the door was opened for developers to push workloads to the GPU and get back some pretty nice returns. Microsoft and many others joined in and began making access to the GPU simpler starting with DirectX 10.
As most people are aware, AMD dropped the first GPUs to utilize HBM (High Bandwidth Memory). These GPUs use a form of HBM called 2.5D which requires the use of an interproser layer than both the memory and the GPU sit on. This is opposed to the 3D stack in which the memory sits on top of the processor that owns it. The traditional stacking of 3D Memory provides significant performance benefits, but would require a different chip for every memory density you plan on releasing. In the GPU world this can be a big problem and is why both AMD and NVIDIA have opted for the 2.5D method.
In September of 2015 (that would be this year) AMD announced that they were splitting off their GPU business into the Radeon Technologies Group (RTG). This move had a number of reasons (most good) and would serve to distance the graphics group from the CPU business. After talking to a few investors they were very optimistic about this move and would consider investing in RTG where they might not have done so in AMD. The split was a long time coming and is actually how AMD should have handled the ATi buy back in 2006 (Merger with separate business units and not a complete buy out).