Intel wants a piece of the graphics card cake and will enter the market in 2022 with the GPU division Arc. The first mobile offshoots were launched a few months ago and the first desktop GPU has recently become available in China and the USA. The first cards should soon make it to Europe. We summarize all information.
It feels a bit strange to call Intel a newcomer or an underdog, but that’s no different when the manufacturer entered the dedicated graphics card market at the end of March to compete with the top dogs Nvidia and AMD. Initially, however, only the entry-level models for the mobile sector were launched, followed on June 22nd by the Intel Arc A380, the first desktop graphics card, but it was only released in China. Intel announced the underlying architecture during Architecture Day 2021 (read the article) already revealed some details. The GPUs should be composed of scalable X e cores and master both ray tracing and the in-house X e Super Sampling function. In terms of performance, the first GPUs can convince in synthetic benchmarks but have major problems in the gaming tests depending on the render engine, which Intel also openly communicates.
All information summarized
In the following table, we have summarized all information regarding the specifications of the Intel Arc GPUs. In addition, Intel regularly publishes updates for the upcoming graphics cards on the official Arc homepage.
| A770 | A750 | A580 | A380 | |
|---|---|---|---|---|
| graphics chip | ACM-G10 | ACM-G10 | ACM-G10 | ACM-G11 |
| Xe cores | 32 | 28 | 24 | 8th |
| FP32 units | 4096 | 3584 | 3072 | 1024 |
| XMX units | 512 | 448 | 384 | 128 |
| base clock | 2100MHz | 2050MHz | 1700MHz | 2000MHz |
| video memory | 8/16GB GDDR6 | 8GB GDDR6 | 8GB GDDR6 | 6GB GDDR6 |
| memory clock | 17.5GB/s | 16GB/s | 16GB/s | 15.5GB/s |
| memory bus | 256-bit | 256-bit | 256-bit | 96-bit |
| TDP | 225W | 225W | 175W | 75W |
| FP32 performance | 17.2 TFLOPs | 14.7 TFLOPs | 10.4 TFLOPs | 4.1 TFLOPSs |
| release | shortly | shortly | later | June 22 (China) |
| Comparable performance | RTX 3060 Ti / RX 6650 XT | RTX 3060 / RX 6600 | RTX 3050 | GTX 1650 / RX 6500 XT |
| Price | 349-399$ | 279-349$ | 199-249$ | $129-139 |
When is the release of the first Intel Arc graphics cards?
Intel’s first dedicated Arc models have already been launched. Initially, however, only the two entry-level mobile GPUs Arc A350M and Arc A370M were available before the Arc A380 desktop GPU followed on June 22nd, which is still only available in China and the USA. We have all the information about the two mobile GPUs and what their release means for gamers. Intel was, therefore, unable to meet the originally planned second quarter, at least in Europe.
EnlargeIntel Arc Alchemist GPUs roadmap ©Intel
The first generation runs under the code name “Alchemist” – formerly Intel DG2 – and is based on the Xe HPG architecture. Intel not only changes its naming of the manufacturing processes (to the article), but also the naming of the graphics cards. Intel has also already revealed the names of the next generations after Alchemist: “Battlemage”, “Celestial” and “Druid”. In short, all the names that we know from one or the other computer game are in alphabetical order, reminding us of Android. Battlemage is already planned for 2023-2024 and according to Tom Fellow, the majority of the development department should already take care of Battlemage. Work on its successor, Celestial, has also already begun.
EnlargeIntel Visual Compute Roadmap ©Intel
Meanwhile, the manufacturer is also working on the so-called “Project Endgame”, which should already offer gamers Arc GPUs as a service at the end of 2022, which sounds like a comparable offer to Nvidia’s GeForce Now. As things stand at present, we strongly doubt that Intel will achieve this goal.
Intel Arc Alchemist – the X e-Core in detail and performance assessment
Structure of Intel Arc Alchemist: The Alchemist SoC, Intel’s first GPU under the Arc branding, is based on the Xe HPG architecture. A Xe Core itself consists of 16 vector engines with 256 bits each and 16 matrix engines with 1024 bits each. The so-called X e HPG Render Slice consists of four such X e -Cores. There are also four ray tracing units for real-time ray calculation. On top of that, there are four sampler blocks for the textures, a block with the geometry units, and a block with the units for the rasterization. X e HPG, i.e. the graphics chip itself, is in turn made up of up to eight render slices. The memory fabric with a shared L2 cache is used for communication between the slices. According to Intel, the clock of X e HPG should be different compared to X e Increase LP by 50 percent at the same voltage. The same applies to the performance-per-watt comparison, were the performance.
Introducing Intel Xe HPG
Intel Xe CoreIntel Xe Core ©Intel
In fact, Intel does not rely on its own production for the production of the graphics chips, but on TSMC’s N6 node. In an interview conducted by the Japanese magazine ASCII (website) with Raja Koduri – the head of Intel’s GPU division – the latter said: “[….] These three, ie the cost, performance, and capacity, are taken into account when deciding which process to use “In other words, the first product, Alchemist, had the best record on TSMC’s N6.” In the interview, Koduri also makes it clear that apart from Intel’s reference designs, there will also be custom designs from the board partners.
No high-end GPU: Rumors about the Arc Alchemist desktop graphics cards have been circulating on the internet for a long time. It is now certain that there will be two different graphics chips, namely the ACM-G10 and the ACM-G11, on the basis of which Intel wants to launch four models – and not seven to eight models as initially assumed. This emerges from an official presentation slide that Intel recently published. In addition, the secret of Intel’s Limited Edition has finally been revealed, although it is not a limited edition, as the name suggests, but simply the reference design, i.e. the counterpart to Nvidia’s Founders Edition.
The top model is the Arc A770, which is based on the full ACM-G10. That means 32 X e -Cores with 32 ray tracing units and 4,096 FP32 units. The GPU has a 256-bit memory interface, to which either 16 GB or 8 GB GDDR6 video memory is connected, depending on the version, which clocks at 17.5 GB/s. The manufacturer specifies the graphics clock as 2100 MHz, whereby the gaming clock should be a good corner higher, and the power loss should be 225 watts. In terms of performance, Intel sees the Arc A770 ahead of AMD’s RX 6600 XT and between Nvidia’s RTX 3060 and RTX 3060 Ti. Intel should aim for roughly QHD gaming with high to very high settings. However, this also means that Intel cannot achieve the performance level of an RTX 3070 or RX 6700 XT with the first Arc generation, which initial rumors have assumed.
EnlargeIntel Arc A770 and A750 ©Intel
Among them, the Arc A750 is listed as a competitor to the RX 6600 or RTX 3060. Intel has a few selected benchmark results for this GPU in a blog post published, according to which the Arc A750 should be ahead of the RTX 3060. But here, too, the TGP is 225 watts, which is significantly more than the 170 watts of the RTX 3060 and the 132 watts of the RX 6600. The manufacturer also states that this performance level is not achieved in all games. The graphics card relies on a partially cropped ACM-G10 with 28 active X e -Cores resulting in 3,584 FP32 units and 28 ray tracing units. The video memory is only half as large with 8 GB GDDR6 and clocks a little slower at 16 GB/s but is also connected with 256-bit. Intel lowered the base clock of the GPU to 2050 MHz despite the same TDP. The price should range from 279 to 349 US dollars.
The Arc A580 with the ACM-G10 cut to 24 X e cores can then be assigned to the lower performance segment. As usual, the 8 GB VRAM is connected to a 256-bit memory interface, with a power consumption of 175 watts and a GPU clock rate of 1700 MHz. Intel itself sees the mid-range card on a performance level with Nvidia’s RTX 3050. Accordingly, the target price of 199 to 249 US dollars is very similar.
All GPUs located below rely on the ACM-G11 graphics chip. This only offers eight Xe cores and, as a result, up to 1,024 shader units. The full expansion is used in the Arc A380 paired with 6 GB GDDR6 video memory on a 96-bit memory bus. The performance should be on par with Nvidia’s GTX 1650 or AMD’s RX 6500 XT.
Intel’s performance classification: Intel itself has already provided extensive benchmark results for the Arc A770, A750, and A380. At this point, it must of course be taken into account that these are not independent test results and Intel even states that the performance in games depends very much on the programming interface used. In current AAA titles, for which Intel has already been able to optimize its drivers, the manufacturer sees itself ahead of the competition from a price-performance point of view. The performance is also said to be very good in games based on DirectX 12 or Vulkan. With DirectX 11 and older DirectX versions, however, the performance drops sharply. The example of Shadow of the Tomb Raider shows that the FPS is almost halved. In addition, Intel states.
EnlargeArc A770 vs RTX 3060 ©Intel
According to Intel’s measurement data, the Arc A770 should be an average of 14 percent ahead of the RTX 3060 in 17 games in 1080p resolution with activated ray tracing, which means that it should be a good ten percent behind the RTX 3060 Ti. By activating Xe Super Sampling, the FPS can be significantly increased again. Unfortunately, Intel does not provide any information on pure rasterization performance. It would also be interesting to see how Intel’s top model performs at 1440p.
EnlargeArc A750 vs RTX 3060 ©Intel
Intel has provided this information on the Arc A750 compared to the RTX 3060, specifically the measured values for 43 DirectX 12 and 6 Vulkan games in the resolutions 1080p and 1440p. According to this data, the Intel GPU should be three to five percent ahead of the direct competition. The measurement results look very promising, but the power consumption of the Arc A750 should be significantly higher and unfortunately, no final euro prices are known yet.
Raytracing and XeSS: Intel Arc also supports all modern features: The upcoming graphics cards should not only offer support for mesh shaders, sampler feedback, and raytracing but also their own super sampling function, which can draw on artificial intelligence. Intel’s approach goes by the name of X-e Super Sampling and is similar to Nvidia’s DLSS approach by considering temporal image information and using a neural network. You can find more information about XeSS on Intels can be found on the official website, and the manufacturer has also announced the first 19 games to be implemented with XeSS.
Introducing Intel Xe Super Sampling
Intel Arc – Alchemist SoCIntel Arc – Alchemist SoC ©Intel
The first test results are disappointing: the embargo on the Intel Arc A380 fell on June 22nd, although it is initially only available in China. According to the first test reports, the Intel GPU is only partially convincing. As already written above, the performance of an RX 6400 or GTX 1650 was expected in advance. According to a review by Shenmedounengce via bilibili the Arc A380 is around 1000 points ahead of the GTX 1650 and 400 points behind the RX 6400 in the 3DMark Firestrike with a score of 11005. In the DirectX 12 test, the Intel graphics card can clearly beat the competition with 5170 to 3620 or 3650 points and in Port Royal, the Arc A380 performs almost twice as well as the RX 6400.
Intel Arc A380 performance according to Shenmedounengce
Intel Arc A380 in 3DMark Fire StrikeIntel Arc A380 in 3DMark Fire Striketo
The raw computing power seems to be right, but the gaming tests show a completely different picture. Testing was done with a Core i5-12400 on a B660 motherboard and DDR4 RAM in LoL, PUBG, GTA5, Shadow of the Tomb Raider, Forza Horizon 5, and Red Dead Redemption II games at 1080p resolution. In none of the tested games does the performance of the Arc A380 come close to that of the competition. In the best case, the Intel GPU performs seven percent slower than the GTX 1650, in the worst case the gap is 24 percent. Intel seems to be struggling massively with driver optimization here, which is probably also the reason why the release is being pushed back further and further and the A380 is initially only available in China. Considering the first tests.
In this regard, Tom Petersen, a graphics engineer from Intel, stated in a video by Gamers Nexus that the main focus when optimizing games is initially on current blockbusters that support DirectX12 and/or Vulkan. Accordingly, in older games that are still based on DirectX 9 or DirectX 11, an ideal gaming experience cannot be expected at first. The YouTuber PRO Hi-Tech got a few more interesting impressions. The power consumption of his tested Arc A380 for gaming was only 35 watts and not the 75 watts specified by Intel. By raising the “GPU Performance Slider” in the driver, the power consumption increased to 55 watts, but the gaming performance increased by an average of 37 percent, which means that the Arc A380 was able to compete with the GTX 1650 to compete.
TechAint 