Intel Dedicated GPU Discussion Thread

[SteelCrysis]

https://www.techpowerup.com/241669/intel...evelopment

Intel is making progress in its development of a new discrete GPU architecture, after its failed attempt with "Larrabee" that ended up as an HPC accelerator; and ancient attempts such as the i740. This comes in the wake of the company's high-profile hiring of Raja Koduri, AMD's former Radeon Technologies Group (RTG) head. The company unveiled slides pointing to the direction in which its GPU development is headed, at the IEEE International Solid-State Circuits Conference (ISSCC) in San Francisco. That direction is essentially scaling up its existing iGPU architecture, and bolstering it with mechanisms to sustain high clock speeds better.

[SteelCrysis]

Intel downplays the original story: https://www.neowin.net/news/intel-showca...-prototype

[SteelCrysis]

Koduri is reportedly working on an Intel dedicated GPU with versions for data centers and gaming, codenamed Arctic Sound: https://www.extremetech.com/gaming/26727...ctic-sound

[SteelCrysis]

Intel dedicated GPU rumored to show up at the next CES: https://www.neowin.net/news/intels-first...-next-year

[SteelCrysis]

Linus managed to get his hands on an engineering sample of an old Intel Larrabee Knight's Corner dedicated graphics card, but sadly it doesn't work:

[SteelCrysis]

https://www.extremetech.com/gaming/27129...et-by-2020

When Intel hired Raja Koduri away from AMD and announced it was working on discrete GPU solutions, it still wasn’t clear exactly when the company would enter the market. Intel CEO Brian Krzanich has told analysts that the company intends to enter the market in 2020 — a slightly faster time frame than Nvidia or AMD may have planned on, and one with potential ramifications for both companies.

[SteelCrysis]

https://www.tomshardware.com/news/intel-...37289.html

Ryan Shrout, via MarketWatch, broke the news this morning when he reported that Intel CEO Brian Krzanich had told analysts at a private investor event that the discrete GPU would arrive in 2020. Shrout also reported that Navin Shenoy, Intel's executive vice president of the data center group, confirmed that the new discrete GPUs would target both the data center and consumer applications, like gaming and professional development.

Intel tweeted out the announcement confirming the 2020 launch date shortly after that. The company confirmed to us over email that it will offer solutions for both the data center and client markets, meaning graphics cards for desktop PCs, but it would not specify which version will come to market first.

[dmcowen674]

Hopefully it brings the prices of Graphics cards back to Earth.

Only having a Duopoly in any Industry has proven to be bad for consumers with sky high prices.

[SteelCrysis]

Elric isn't confident that Intel will succeed:

• Intel is struggling to reach 10nm
  
• He doesn't think that Intel will go for the consumer desktop GPU market
  

[SteelCrysis]

Intel has released its first teaser: https://www.neowin.net/news/intel-teases...20-release

[SteelCrysis]

https://www.techpowerup.com/248089/intel...screte-gpu

It seems that now, with their own GPU development underway, Intel has found new love for the GPU-accelerated compute space. The choice of Vulkan is also interesting as the API is available on a wide range of platforms, which could mean that Intel is trying to turn Vulkan into a CUDA killer. Of course there's still a lot of work needed to achieve that goal, since NVIDIA has had almost a decade of head start.

[SteelCrysis]

https://www.techpowerup.com/249045/darre...s-division

Darren McPhee worked 12 years for ATI and AMD. When he left AMD in 2015, he was one of the company's top marketing managers. For the last three years he has worked for various companies, but the surprise has come with Intel recruiting him to occupy the position of Product Marketing Manager in its 'Discrete Graphics' group, one of the most interesting initiatives in the recent times.

[SteelCrysis]

https://www.techpowerup.com/250032/intel...g-for-2020

According to DigiTimes, Intel's top graphics executive Raja Koduri and other senior Intel partners will be hosting a discrete GPU-focused conference this December. The conference aims to instill confidence in shareholders and customers alike in that Intel is pursuing its high-performance discrete entry into the graphics card market at a fast pace. The GPU architecture, codenamed Arctic Sound, is expected to debut by 2020, aiming for the gaming, AI, and machine learning sectors - much like any GPU solution these days. It remains to be seen which details - if any - can be gleaned from this conference, but we'll keep you up to date when those surface.

[SteelCrysis]

https://www.tomshardware.com/reviews/int...932-3.html

Intel presented a slide outlining its new Xe architecture that will come after the Gen11 graphics engine. Intel says the next generation of its graphics architecture will denote a transition from the "Gen" naming convention and will scale from integrated on-chip graphics up to discrete GPUs that will span the mid-range, enthusiast, and data center markets. That means it will scale from teraflops of performance integrated into a standard processor up to petaflops of performance with discrete cards.

This announcement certainly hints that both the integrated graphics and discrete cards will share the same underlying architecture, but Intel wouldn't answer further questions. Intel is also on track to deliver on its previously-announced timeline, saying the Xe graphics cards will debut in 2020.

[SteelCrysis]

https://www.tomshardware.com/news/intel-...38246.html

There's no doubt that Intel's entrance into the discrete graphics card space is the most exciting news the industry has received in years. We just hope it can execute on its mission this time around instead of taking unreasonable risks like it did with Larrabee. Intel has assembled a veteran team that should guide the company away from obvious missteps, though. Many of those hires hail from AMD and Intel has opened a new graphics development office in Toronto, uncomfortably close to AMD's campus, so it can do some recruiting.

We also see a new level of engagement from the company on the graphics front, as evidenced by the company's recent Ask You Anything session on Reddit where it asked enthusiasts what they would like to see from its coming products. Intel appears to be using the feedback wisely, which is a refreshing change of pace.

[SteelCrysis]

https://www.tomshardware.com/news/intel-...38631.html

Intel has confirmed that recent patches to its Linux graphics driver were related to its continued work on preparing the ecosystem for its new line of discrete graphics cards.

Phoronix reported that Intel released 42 such patches with more than 4,000 lines of code between them on February 14. The main purpose of the patches was to introduce the concept of memory regions in "preparation for upcoming devices with device local memory." (Such as, you know, discrete graphics cards.)

The integrated GPUs found in many of Intel's CPUs don't rely on this memory region setup, so all signs pointed towards the Linux drivers being part of Intel's plans to build standalone GPUs. Just in case those signs weren't obvious enough, however, Intel has confirmed the motivation behind these drivers in a tweet:

[SteelCrysis]

https://www.tomshardware.com/news/intel-...38638.html

Intel has acquired Ineda Systems, a graphics company based out of Hyderabad, India for an undisclosed all-cash sum. The deal is said to be part of Intel's efforts to develop its own discrete graphics technologies so it can compete with the likes of AMD and Nvidia.

The Times of India reported that Intel was more interested in Ineda Systems' experience than its products. Roughly 100 engineers are expected to join Intel's graphics division following the deal's closure--a significant influx of new talent for the group.

Intel responded with the following statement:

"Intel acquired engineering resources from Ineda Systems, a silicon and platform services provider based in Hyderabad. This transaction provides Intel with an experienced SOC (system on chip) team to help build a world-class discrete GPU business."

https://www.techpowerup.com/252695/intel...e-gpu-tech

Looks like this caused enough attention at chip giant Intel, who's trying to come up with a competitive design for a discrete graphics processor, that's able to take on AMD's and NVIDIA's offerings. While Ineda certainly has patents that could come in useful, it looks like Intel is more interested in the company's manpower. With around 100 engineers, the company has a lot of talent, that's experienced in chip design and how to make these chips energy efficient.
...
Ineda Systems founder Dasaradha Gude used to be managing director of AMD India, which suggests some ties with Raja Koduri, who used to be head at AMD's Radeon Graphics department and moved to Intel not long ago, to start up their discrete GPU project. This makes us speculate that Raja, who's also from Hyderabad originally, was instrumental in making this buy-out happen.

[SteelCrysis]

Intel is beginning publicity for its dedicated graphics cards, inviting gamers to get involved: https://www.techpowerup.com/252823/intel...continents

[SteelCrysis]

https://www.tomshardware.com/news/intel-...38246.html

Intel whipped the covers off its Xe graphics branding at its recent Architecture Day, but it doesn't represent the actual final brand names, like Radeon or GeForce. Instead, the Xe branding signifies Intel's full range of low- to high-power graphics solutions.
...
Intel's Xe line will come after Gen11 and replace the Gen12 branding. These graphics processors will scale from integrated graphics chips on processors up to discrete mid-range, enthusiast, and data center/AI cards. Intel says it will split these graphics solutions into two distinct architectures, with both integrated and discrete graphics cards for the consumer market (client), and discrete cards for the data center. Intel also disclosed that the cards would come wielding the 10nm process and arrive in 2020, which is in line with its previous projections.
...
Intel still has a lot of work to do on the enablement front, such as recruiting AIB partners, but it is possible the company will produce the new cards natively. It's hard to tell which path Intel will take, including which segments of the market it will target first. It's natural to expect Intel's freshman effort to target the mid-range enthusiast market with enticing price points, but true to its heritage, the company may also bring high-end cards based on enterprise designs to the enthusiast space, just like it does with its high-end desktop processors.

Only time will tell, but Intel has the deep pockets to take losses for the first few generations as it builds a customer base. In either case, Intel's new cards will force both AMD and Nvidia to become more competitive on pricing.
...
In a sign of things to come, Intel has increased the cadence of driver releases lately, with a heavy focus on zero-day releases. At its recent event, the company also reiterated its commitment to the software ecosystem in the graphics market, so we know the company is headed in the right direction on that front. Market-leader Nvidia employs more software engineers than hardware engineers, so success in this area is key.

Intel is also already working diligently on enabling its driver ecosystem, with the first Linux patches, 42 in fact, arriving in early 2019. The 4,000 lines of code enable support for device-local memory regions, which is key for supporting a discrete graphics card with its own onboard memory. Intel typically releases driver patches roughly a year ahead of the launch of new devices, so these updates indicate the project is well underway.
...
Both AMD and Nvidia also rely on outside fabs. Intel's fabs have been a liability lately as the company struggles with its 10nm process, but when used correctly, Intel's fabs are a tangible advantage that could give the company a leg up on its competitors. Chip packaging techniques are becoming the true differentiator in the waning light of Moore's Law, and its fair to say that the companies that produce AMD and Nvidia's GPUs (TSMC and GlobalFoundries) cannot compete with Intel's next-gen packaging technologies. At least for now.
...
There's no doubt that Intel needs to succeed in the GPU market, at least in the data center. We often look to the supercomputing realm to sniff out new trends, and the rise of GPUs in that space is explosive: in 2008 not one supercomputer used GPUs for computation, instead relying on the tried-and-true CPU, but now 80% of compute power in the top 500 supercomputers now comes from GPUs. It appears the pressure Nvidia's GPUs are putting on Intel's Xeon sales is finally having an impact: We would have never expected Intel to share a chart like the one above, which clearly shows that GPUs are part of the path forward to extending Moore's Law.

[SteelCrysis]

https://www.tomshardware.com/news/intel-...38851.html

Intel and the U.S. Department of Energy (DOE) announced today that Aurora, the world's first supercomputer capable of sustained exascale computing, would be delivered to the Argonne National Laboratory in 2021. Surprisingly, the disclosure includes news that Intel's not-yet-released Xe graphics architecture will be a key component of the new system, along with Intel's Optane Persistent DIMMs and a future generation of Xeon processors.

[SteelCrysis]

https://www.tomshardware.com/news/amd-sa...38901.html

Contrary to several recent reports, Intel hasn't yet displayed a render of its planned graphics card. Instead, the company recently displayed some conceptual fan art from Cristiano Siqueira at its Odyssey event at the Game Developers Conference (GDC) 2019.

Siqueira is the same student from Brazil that made a series of conceptual Intel graphics card images that swept the internet earlier this year, but Intel doesn't employ him, and his self-titled "Dragon Scale" render doesn't represent any planned designs or prototypes from Intel.

[SteelCrysis]

https://www.techpowerup.com/254720/intel...re-xe-gpus

Intel's Xe discrete GPU project head Raja Koduri recently visited a Samsung Electronics silicon fabrication facility in Korea at the backdrop of the company's major 5 nm EUV announcement. This sparks speculation that Koduri could be exploring Samsung's portfolio of sub-10 nm contract-manufacturing offerings to mass-produce Xe discrete GPUs.

[SteelCrysis]

https://www.tomshardware.com/news/intel-...39224.html

Intel published a news byte today outlining its announcements at the FMX graphics trade show taking place in Germany this Week. It includes the tasty tidbit that the company's forthcoming data center specific Xe graphics architecture will support hardware-based ray tracing acceleration.
...
As a quick refresher, Intel's Xe graphics architecture is Intel's forthcoming range of low- to high-power graphics solutions. These graphics processors will scale from integrated graphics chips on CPUs, up to discrete mid-range, enthusiast and data center/AI cards. Intel said it will split these graphics solutions into two distinct architectures, with both integrated and discrete graphics cards for the consumer market (client) and discrete cards for the data center. The cards will come wielding the 10nm process and should arrive in 2020.

[SteelCrysis]

https://www.tomshardware.com/news/intel-...39298.html

Intel also announced that an XE GP-GPU (General-Purpose Graphics Processing Unit) will be the first 7nm product to come to market, arriving in 2021. Intel will use its EMIB (Embedded Multi-Die Interconnect Bridge) and 3D Foveros technologies for the Xe Graphics Architecture. Intel also revealed that the new GP-GPU design will power the exascale Aurora supercomputer. We have extensive coverage of Aurora here.

[SteelCrysis]

https://www.techpowerup.com/256960/intel...phics-card

Intel China through its Weibo (Twitter-equivalent) handle put out more official-looking renders of its Xe graphics card. The Weibo post doesn't cite an author, leading us to speculate that the company's industrial design team is close to finalizing a product-design for at least the client-segment derivative of Xe.

[SteelCrysis]

https://www.tomshardware.com/news/intel-...40029.html

Going forward, Intel will unify all future graphics products under the new Xe branding, whether they be integrated or discrete solutions. The chipmaker has big plans to launch discrete graphics cards to compete against Nvidia and AMD in the mid-range and enthusiast market. Intel's graphics cards will be built with the 10nm manufacturing process, and hopefully, arrive next year.

The Intel driver refers the DG1 and DG2 codenames, whereby DG reportedly stands for discrete graphics. The DG1 and DG2 products carry the LP and HP designations, respectively. The acronyms likely stand for low-powered and high-powered, which might be Intel's way of differentiating the high-end offerings from the low-end ones.

We're not entirely sure what the numbers at the end of the codenames mean at this time. If we had to make an educated guess, the numbers could represent the number of execution units (EUs) for each product. In that case, the DG2 family would have up to three members with 128, 256 and 512 EUs each. Judging by the number of EUs, the DG2 products are probably aimed at the mid-range segment.

[SteelCrysis]

https://www.techpowerup.com/257911/intel...at-usd-200

During interview with Russian YouTube channel called PRO Hi-Tech, Raja Koduri, Intel's chief architect and senior vice president of architecture, software and graphics, talked about his career, why he left AMD, and where Intel is going with its discrete GPU attempts. However, one of the most notable things Mr Koduri said was regarding upcoming GPU lineup code-named Arctic Sound. He noted that Intel plans to release first GPU as a mid-range model at a price of $200, while enterprise solutions that utilize HBM memory will follow that.

Koduri said that he wants to replicate AMD's strategy of capturing high-volume price-points, such as the $199 Radeon RX 480. The plan here is to bring an affordable, good performing GPU to the masses - "GPUs for everyone" as he calls them. Additionally, he states that Intel's current strategy revolves around price, not performance, providing best possible value to consumers. Intel's approach for the next two or three years is to launch a complete lineup of GPUs, with a common architecture being used for everything from iGPUs found inside consumer CPUs to data-center GPUs.

[SteelCrysis]

https://www.tomshardware.com/news/intel-...40083.html

Russian YouTube channel Pro Hi-Tech posted a video claiming that Raja Koduri, Intel's chief architect and senior vice president of Architecture, Software, and Graphics, said that the company would launch its first discrete graphics card at the $200 price point. After a bit of fact-checking, we found that the comments attributed to Koduri are not accurate.
...
Pro Hi-Tech is a Russian outlet, so it isn't surprising that the sit-down interview with Koduri is in Russian, but the channel chose to dub over Koduri's comments with a translation that renders Koduri's actual comments, spoken in English, undecipherable.
...
We followed up with Intel to see if Koduri actually made the statements attributed to him, and the company responded that Koduri's comments "had been misconstrued and/or lost in translation." Intel provided us with audio of Koduri's actual comments, which we've transcribed here for reference:

“Not everybody will buy a $500-$600 card, but there are enough people buying those too – so that’s a great market.

So the strategy we’re taking is we’re not really worried about the performance range, the cost range and all because eventually our architecture as I’ve publicly said, has to hit from mainstream, which starts even around $100, all the way to Data Center-class graphics with HBM memories and all, which will be expensive.

We have to hit everything; it’s just a matter of where do you start? The First one? The Second one? The Third one? And the strategy that we have within a period of roughly – let’s call it 2-3 years – to have the full stack." - Raja Koduri

[SteelCrysis]

https://www.tomshardware.com/news/intel-...40562.html

In a conspicuous Twitter post on Friday, Intel’s chief architect Raja Koduri seemingly hinted at a June 2020 announcement or release of the company’s forthcoming Xe graphics cards.
...
The picture shows the backside of a Tesla Model S, whose custom license plate unambiguously reads ‘THINKXE.’ ("Think Xe.") Given Intel’s previous disclosures, also seeing ‘2020’ on the plate does not come as a surprise, and indeed gets one to think.

Most conspicuous is the June month on license plate. The June 2020 date seems too specific to be a coincidence. Intel might have an announcement, and perhaps even a release, planned for Computex in June 2020.

https://www.tomshardware.com/news/intel-...40589.html

Today brings us two new bits of information. For starters, the integrated Mobility Xe iGPUs should offer twice the performance of Gen11 iGPUs, with the goal to deliver 1080p gaming at 60 fps for many titles.

This was detailed by Intel’s Director of Tech Kenichiru Yasu at IDC in Tokyo when showcasing some benchmarks for the Iris Plus GPUs, which didn’t quite do the trick yet to hit the 60 fps mark. Do note that the statement was phrased as 60 fps as a target where Iris currently hits 30 fps. Given that this industry is renown for over-promising, we’re keeping our fingers crossed for the time being about the level of significance we can expect from this performance bump.

Moreover, however, now that Nvidia has been working hard on marketing Ray tracing, it appears Intel will also be joining the force. It shouldn’t come as a surprise, given that AMD has also been working on its own ray-tracing solutions. Ray tracing for Intel’s Xe GPUs is said to be limited to the discrete graphics cards.

[SteelCrysis]

https://www.techpowerup.com/260236/intel...n-mid-2020

According to the sources close to DigiTimes, Intel will unveil its first discrete 10 nm graphics cards named "Xe" very soon, with the first wave of Xe GPUs expected to arrive some time in 2020. Said to launch mid year, around July or August, Intel will start selling initial Xe GPU models of the long awaited product to consumers, in hope of gaining a share in the massive market using GPU for acceleration of all kinds of tasks.

Perhaps one of the most interesting notes DigiTimes reported is that "... Intel's GPUs have already received support from the upstream supply chain and has already been integrated into Intel's CPUs to be used in the datacenter and AI fields.", meaning that AIB partners already have access to first 10 nm graphics chips that are ready for system integration. First generation of Xe graphics cards will cover almost whole GPU market, including PC, datacenter, and AI applications where NVIDIA currently holds the top spot.

[SteelCrysis]

https://www.tomshardware.com/news/intel-...xe-support

Intel’s latest driver work for Linux 5.5 has shown that it is working on multi-GPU capabilities for its discrete Xe and Gen 12 graphics solutions. The capabilities would include integrated and discrete graphics working concurrently.
...
Such multi-GPU capabilities are further noteworthy because there are rumors that Intel’s Arctic Sound discrete GPU will consist of a multi-chip package (MCP) of possibly four graphics dies, although it is not known if this multi-GPU work is related.

[SteelCrysis]

https://www.tomshardware.com/news/intel-...u-power-on

Swan announced during the call that Intel had powered on the company's DG1 GPU, its first discrete graphics card based on the 10nm process and Xe architecture, for the first time during the quarter, which "is an important milestone." The 10nm DG1 is expected to launch next year, and Swan also revealed that the company is on track to deliver its 7nm discrete GPU in the fourth quarter of 2021. That indicates that Raja Koduri's graphics group is on schedule with a regular cadence of products in the pipeline.

[SteelCrysis]

https://www.techpowerup.com/260524/intel...y-shipping

Additionally, the developer kit for the "DG1" graphics card is supposedly being sent to various developers over the world, according to European Economy Commission listings. Called the "Discrete Graphics DG1 External FRD1 Accessory Kit (Alpha) Developer Kit" this bundle is marked as a prototype in the alpha stage, meaning that the launch of discrete Xe GPUs is only a few months away. This confirming previous rumor that Xe GPUs will launch in 2020 sometime mid-year, possibly in July/August time frame.

[SteelCrysis]

https://www.tomshardware.com/news/intel-...soc-design

Intel has hired Masooma Bhaiwala, a former AMD executive. She will lead Intel’s discrete GPU efforts, the first of which will debut in 2020.

Masooma Bhaiwala confirmed her position to CRN as vice president, discrete GPU SoCs within Intel's Graphics and Throughput Computing Hardware Engineering organization. That organization, in turn, is part of the Intel Architecture, Graphics and Software group led by Raja Koduri, also a former AMD executive. Koduri joined Intel two years ago to lead the company’s discrete graphics ambitious to develop a full stack of discrete GPU hardware across all segments.

[SteelCrysis]

https://www.extremetech.com/gaming/30383...d-graphics

Ever since Intel announced Xe, its next-generation graphics processor, there’s been speculation and discussion about what kind of GPUs the company would bring to market. Ice Lake’s new GPU is an important step along this path, with its substantial performance uplift and Gen 11 graphics. There have been rumors that Intel’s next-generation 10nm CPU, codenamed Tiger Lake, will have 96 EUs — and a new bit of information suggests at least one model of Intel’s upcoming dGPU will feature that many EUs as well.

Hot Hardware spotted this data via Twitter user Komachi. Komachi found the following document on the European Economic Community (EEC) website:

The “96EU” remark implies that this is a Tiger Lake-style configuration, while the “Alpha” points to the hardware still being very early in development. I’m not going to try to speculate too much on how much performance moving from 64 to 96 EUs will get Intel, though we’d expect that kind of shift to boost performance by 1.5x on paper, assuming no other significant bottlenecks in the design prevent scaling out. Since we’d be dealing with a standalone card, we can probably assume clocks equal-to or higher than an Ice Lake laptop part. With 8 threads per EU, we’d consider this a 768-core configuration (though AMD, NV, and Intel GPUs all perform a different amount of work per-core, so GPUs can’t be compared directly on core count).
...
The takeaway is this: Even if Intel launches a low-end card with a similar configuration to its iGPU, overall dGPU performance is very likely to be higher — but we can’t really judge by how much. There’s no way to perform this kind of comparison with an Nvidia card, and we have older AMD results that point in one direction and newer results that imply a smaller gap for some configurations. The RX 550 is almost never slower than the 3400G, despite running at lower clocks. If Intel’s goal is to challenge from the low and midrange markets first before it makes a play for the high-end, bringing in a lower-tier part first makes sense. Intel may be looking for a chip that can let it challenge Nvidia’s lower-end parts in laptops and the occasional desktop more than it wants to bring a huge-die product to market for gamers. Every dollar of OEM laptop spend dedicated to a non-Intel GPU is a dollar of profit that Intel isn’t capturing. Intel’s CEO, Bob Swan, has openly stated that he intends to focus on being a company with 30 percent market share in a huge range of markets rather than laser-focusing on 90 percent market share in the CPU space. Taking more space in critical consumer markets is key to doing that.

[SteelCrysis]

https://www.extremetech.com/gaming/30452...ng-to-isvs

While Intel didn’t launch any hardware at CES this week, the company did show off its upcoming Tiger Lake CPUs and the DG1 discrete GPU in a brief gaming demo. What Intel didn’t show, however, was an independent graphics board. It turns out that the company does have GPUs ready to ship out for early software validation. Units are already shipping to ISVs.
...
There’s no sign of a PCIe connector on this card, implying that the power draw is 75W or less. That would make sense for a chip that appears to be debuting as a mobile part. Intel may be planning to sell DG1 as an EMIB-attached solution alongside a Tiger Lake CPU, or it could be prepping a discrete mobile card for ISVs to include. Either way, Xe is going to ship in three specific families:

Xe LP is ultra-mobile PCs, entry-level graphics, and midrange graphics, with TDP’s expected to be in the 5W-20W range and the ability to scale up to 50W. Xe HP would cover the 75W – 250W segments, delivering (in theory) gaming performance that could compete with cards from Nvidia and AMD. Above 250W would be the Xe HPC, intended for HPC/exascale systems, deep learning and training, and cloud graphics.

It’s entirely possible that Intel would choose to commercialize its Xe HPC silicon for gaming as well, if the demand was there and if the characteristics of the GPU lent themselves to this kind of endeavor. Nvidia and AMD have both designed high-end GPUs for the workstation and server markets before bringing them over to the consumer space. The DG1 is expected to have 96 EUs and 768 shader units in total (8 threads per EU), but there are rumors of higher-specced DG2 cards, with 128, 256, and 512 EUs. That last, if true, would put Intel’s largest GPUs on approximate core parity with what AMD and Nvidia ship at the high end of the market — but we can’t really compare GPUsSEEAMAZON_ET_135 See Amazon ET commerce based on core count, because the amount of work done per core can vary significantly between different architectures.

One thing to keep in mind is that DG1 performance may improve over the GPU’s first few years of life. While Xe is based on previous GPU architectures, not all developers make games that target Intel IGPs in the first place. There’ll likely be a learning curve for both Intel and its software partners to negotiate.

[SteelCrysis]

https://techreport.com/news/3467997/inte...fo-at-gdc/

Intel is getting ready to make a big splash with its Xe graphics cards, but the company has mostly spoken in broad terms about how it’ll scale from exascale needs down to integrated mobile chips and everything in between. Now, though, Intel is ready to talk in detail about Xe at GDC 2020 this March.

Antoine Cohade, Intel’s Senior Developer Relations Engineer, will give a presentation called a “Primer on Intel Graphics Xe Architecture” at the Game Developers’ Conference.

[SteelCrysis]

https://www.tomshardware.com/news/intel-...aked-specs

Digital Trends has reportedly gotten its hands on an internal presentation from Intel’s Data Center Group pointing to specifications for Intel's Xe, codenamed Arctic Sound, graphics cards. The presentation is from early 2019, however, so it's possible that some things could have changed since then, such as final TDP numbers.
...
According to the presentation, Intel’s high-end Arctic Sound graphics cards will feature up to one, two or four slices. This refers to the number of chiplets the graphics card contains, and they're also called tiles because they are identical.

The four-tile configuration has a 400W or 500W TDP and 48V input. That shows that Intel is serious about delivering high performance, while making unlikely that it's intended for individual consumers.

The card with two tiles is presented with a 300W TDP and 12V input. That's as high a TDP as Nvidia’s Tesla V100 and slightly higher than the Nvidia Titan RTX workstation GPU.

The single-tile card reportedly has a TDP of 75W or 150W. One of the slides suggests that the single-tile configuration is intended for client use.

[SteelCrysis]

https://www.techpowerup.com/264637/intel...enerations

Intel is rumored to be aligning its future-generation Xe GPU development with TSMC's node development cycle, with the company reportedly negotiating with the Taiwanese foundry for 6 nm and 3 nm allocation for its large Xe GPUs. Intel's first Xe discrete GPUs for the market, however, are reportedly built on the company's own 10 nm+ silicon fabrication process.

While Intel's fascination with TSMC 3 nm is understandable, seeking out TSMC's 6 nm node raises eyebrows. Internally referred to as "N6," the 6 nm silicon fabrication node at TSMC is expected to go live either towards the end of 2020 or early 2021, which is when Intel's 10 nm+ node is expected to pick up volume production, beginning with the company's "Tiger Lake" processors. Perhaps a decision has been made internally to ensure that Xe doesn't eat too much into Intel's own foundry capacities meant for processor manufacturing, and to instead outsource Xe manufacturing to third-party foundries like TSMC and Samsung eventually. Way back in April 2019 it was rumored that Intel was evaluating Samsung as a foundry partner for Xe.

[SteelCrysis]

https://www.techpowerup.com/265073/intel...stic-goals

Intel is revisiting the concept of asymmetric multi-GPU introduced with DirectX 12. The company posted an elaborate technical slide-deck it originally planned to present to game developers at the now-cancelled GDC 2020. The technology shows promise because the company isn't insulting developers' intelligence by proposing that the iGPU lying dormant be made to shoulder the game's entire rendering pipeline for a single-digit percentage performance boost. Rather, it has come up with innovating augments to the rendering path such that only certain lightweight compute aspects of the game's rendering be passed on to the iGPU's execution units, so it has a more meaningful contribution to overall performance. To that effect, Intel is on the path of coming up with SDK that can be integrated with existing game engines.

Microsoft DirectX 12 introduced the holy grail of multi-GPU technology, under its Explicit Multi-Adapter specification. This allows game engines to send rendering traffic to any combinations or makes of GPUs that support the API, to achieve a performance uplift over single GPU. This was met with lukewarm reception from AMD and NVIDIA, and far too few DirectX 12 games actually support it. Intel proposes a specialization of explicit multi-adapter approach, in which the iGPU's execution units are made to process various low-bandwidth elements both during the rendering and post-processing stages, such as Occlusion Culling, AI, game physics, etc. Intel's method leverages cross-adapter shared resources sitting in system memory (main memory), and D3D12 asynchronous compute, which creates separate processing queues for rendering and compute.
...
Intel iGPUs are approaching the 1 TFLOPs compute power barrier, with Gen11 and the upcoming Xe-based iGPU debuting with "Tiger Lake." That's a lot of compute power not to take advantage of. Intel's tech can prove particularly useful with notebooks that have entry- thru mid-range discrete GPUs, as all Intel mobile processors pack iGPUs and implement dynamic switching between iGPU and dGPU.