Intel is back in the great game of dedicated graphics cards, with the blue team suggesting that they launch their first Intel Xe graphics solutions in 2020 through a wide selection of platforms, environments and prices. This is Intel's first new discrete graphics card in more than twenty years.
The technology behind this GPU and its potential performance remain shrouded in mystery, but as time goes by, the details come to light. If it proves to be a viable alternative to Nvidia and AMD, this is one of the most important events in the graphics card industry in this century.
This is what we know so far.
Prices and availability
Intel has stated that we can expect a debut in the summer of 2020 and that it pointed to a wide selection of graphic options, some aimed at consumer electronics and others for high performance. Intel revealed the first of its business-focused GPUs in November 2019. Intel has said that the Ponte Vecchio GPU will not launch until 2021, as it will be based on a second generation of Xe graphics. WCCFTech reported that it will first be used on the Aurora Supercomputer.
In mid-2019, Intel executive Raja Koduri tweets a photo of his Tesla, with the name "ThinkXE" on the board and with a label showing the month of June and 2020. This could be an early announcement of a launch in June 2020 for the Xe graphics card, perhaps in time for Computex.
In December 2019, however, rumors began to circulate that the development of Xe graphics was not progressing fast enough. A user of the Chiphell forums (via WCCFTech) stated that the development of Xe graphics did not work well and that we may not see any Intel GPU before the end of 2020. They also claimed that this will delay Ponte Vecchio in 2022 as soon.
We will set our graphics free. # SIGGRAPH2018 pic.twitter.com/vAoSe4WgZX
– Intel Graphics (@IntelGraphics) August 15, 2018
If true, this will be problematic for a number of reasons, because the Xe graphics are scheduled to appear on the next Intel Tiger Lake processors, currently scheduled for a 2020 version. This is likely to be a priority for Intel, for which could cause a shortage of dedicated GPUs if they are launched at a similar time.
There is also the rumor that Intel will be the only company that sells its GPUs at launch, as it has had trouble defining new relationships with card partners that include them to create custom variants and with overclock.
A leak of information on the release of Intel video drivers in the summer of 2019 referred to four different discrete graphics cards. That suggests that for the ?gamers?And the fans of?hardware?, There will be a relatively wide selection of graphics cards to choose from.
Architecture and performance
When Intel made its official announcement about the new graphics card technology it was developing, the company made it clear that it was building a dedicated graphics card. While that suggests that he was building something other than his existing built-in GPUs, these cards will be based on the same generation 12 architecture at the core of their integrated graphics solutions for the next processor generations like Tiger Lake, but properly scaled.
Intel previously announced that there will be three different microarchitectures as part of the Xe range, known as Xe-LP, Xe-HP and Xe-HPC. These will cover basic level graphics chips, enthusiast level graphics cards and GPUs based on data centers for rendering and supercomputing. However, in each case, Intel take advantage of the ?Xe? brand for all future graphics projects, whether integrated or discrete, although there will be large performance differences between them.
The best view of the architecture behind these new cards came in our own report on a slide obtained from Intel that indicated the TDP (Power of thermal design, for its acronym in English), and the architecture "in mosaic" of these first cards Xe.
According to an internal presentation of Intel from the beginning of 2019, these execution units (UE) are divided into ?mosaics?, potentially similar to the chiplets we have seen used in AMD Ryzen 3000 Zen 2 processors. When aligned with a leak of information from some controllers in July 2019, it was deduced that these mosaics probably contain 128 UEs and could be combined with the Intel Embedded Multiple Die Interconnection Bridge (EMIB).
In another part of the presentation, Intel claimed to be developing graphics cards across the price and performance spectrum, until the end. Designs ranging from a TDP of 75W to 500W are included. At the top end, that's twice the TDP of the most powerful gaming graphics card in Nvidia, the 2080 Ti. With the previous load confirming that the GPU use an input voltage of 48V, that makes it almost certainly a high-performance card.
It is possible that this 500W, four-tile card is the Ponte Vecchio GPU that Intel announced in November 2019. Although Ponte Vecchio did not debut until 2021, Intel called it the ?first GPU exascale? and declared that it would use a Compute eXpress Link interface ( CXL) that runs on a PCIexpress 5.0 connection.
According to the reports, the other single and double mosaic GPUs will have TDP of 75W, 150W and 300W. While the latter is still high, these are much more typical TDPs that may well be equivalent to gaming graphics cards for conventional desktops. Comparable cards that already exist at those levels include the GTX 1650 at 75W, the RX 5600 XT at 160W and the Titan RTX at 280W.
HBM and PCIexpress 4.0
In an interview (already eliminated) with Intel's head of graphic development, and Raja Koduri, long-time leader of Radeon Technology Group, suggested that these new graphics cards may not use the most typical GDDR6 for memory. Instead, it was suggested that these GPUs may use higher bandwidth memory (HBM), which is much more expensive.
That was minimized and discredited mostly by Intel, but the Ponte Vecchio GPU announcement showed the HBM memory used in the design. The internal documents that Digital Trends obtained in February 2020 also pointed to the use of high bandwidth memory in its Xe GPUs, specifically HBM2e, which is the most efficient and fastest generation of HBM. According to the reports, it is stacked in 3D directly on the GPU using Intel Foveros technology.
That does not mean that we will see such an expensive memory used in consumer graphics, but it is an interesting note that whatever Intel is working on should be able to take advantage of HBM. It is rumored that the next-generation AMD ?Big Navi? GPU has an HBM memory option along with GDDR6, so it is possible that Intel offers something similar with its GPUs, which offers the high-end HBM options and the options of GDDR6 low range.
Most consumer graphics cards adhere to GDDR6, including all high-end RTX 2080 Ti from Nvidia. The only recent exception has been Fury-X from AMD, Vega 56, 64 and Radeon VII. His recent RX 5700 ms graphics cards left HBM2 in favor of GDDR6.
Another note in the internal Intel documents we have seen was a mention of PCIexpress 4.0 support. Although current graphics cards don't even come close to the saturated PCIexpress 3.0 x16 slots, PCIe 4.0 support alludes to the performance potential of the new Intel cards. They might also be better optimized to require fewer lanes, enabling more lanes for storage units and other additional cards.
Nvidia has been dedicated to real-time ray tracing, making it a key feature of its current RTX graphics cards. Despite its slow start, technology has the potential to become the most important new feature in computer graphics in the coming years. The problem is that increasing lighting and realistic shadows can be costly in terms of performance. AMD has doubted more about diving into the world of ray tracing for that exact reason, although it has plans to support it in the future, especially on consoles such as the Playstation 5.
Long before the launch of Xe, Intel has already left the support door for lightning tracking on its future GPUs. Jim Jeffers, senior senior engineer at Intel (and senior director of advanced rendering and visualization), made the following statement: roadmap Intel Xe architecture for optimized data center rendering includes lightning path hardware acceleration support for the Intel Rendering Framework family of APIs and libraries. We still do not know what that declaration means for ray tracing in games, but, if hardware acceleration has been implemented, we would be surprised that Intel also did not give it to players.
Drivers and software
Both Nvidia and AMD have their respective driver software packages that do more than help the GPU communicate correctly with the system in general. Features such as image sharpness, shading filters, game recording, low latency inputs and dynamic resolution adjustment have improved the offerings of leading GPU manufacturers. Intel wants to do something similar when it launches its Xe graphics cards and has already begun to lay the groundwork for that.
In March 2019, Intel presented its Graphics Command Center. At this time, it only works with Intel's integrated graphics solutions, but includes options for launching and optimizing games and the ability to modify global GPU options in all applications. It's pretty basic for now and offers basic functionality for integrated Intel GPU users, but the foundations for a more complete Intel GPU software package have been laid when Xe debuts in the future.
Along with the hardware development, Intel is reported to be devoting a lot of time and energy to its driver development, and WCCFTech reports that they need great optimization before seeing the light of day.
AMD students are helping to achieve it
Intel has not released a discrete graphics card in 20 years. He developed what became a coprocessor, Larabee, in the late 2000s, but it turned out to be far from competitive with modern graphics cards, even some intriguing use cases were found in its own right. To develop its graphic architecture on something worthy of a dedicated graphics card, Intel hired some industry experts, especially Raja Koduri. He was hired directly from AMD, where he spent several years as chief architect of the Radeon Technology Group, directing the development of AMD Vega and Navi architectures.
He has been there for over a year and even in mid-2018 he was joined by Jim Keller, the principal architect of AMD's Zen architecture. It directs the development of Intel silicon and, according to Intel, help "change the way Intel builds silicon." That could be considered additional evidence of Intel's drive towards a viable 10 nm production.
Other former AMD employees that Intel chose over the past year include former AMD global product marketing director Chris Hook, who spent 17 years working at the company, and Darren McPhee, who now directs Intel product marketing for discrete graphics.