Intel Arc A770 16GB and Arc A750 8GB in review: Are Intel’s gaming debutants convincing?

It’s about time: After more than five years of development and many a postponement, the gaming graphics cards of the Arc series are finally going on sale. Intel’s risky foray into new, yet old, territory is being watched on all sides with a wild mix of excitement, skepticism, and anticipation. Around 23 years after the last Intel graphics card and extensive preparation, the processor market leader dares to attack the apparently eternal duopoly of AMD Radeon and Nvidia Geforce. With the first gaming graphics cards from the Arc-A7 series, Intel is targeting Nvidia’s Geforce RTX 3060 and wants to beat it in terms of value for money. In the test, we not only compare the Intel Arc A770 and Intel Arc A750 with the Geforce mid-range, but of course also with the Radeon RX 6600 XT and RX 6650 XT. So much in advance: Intel’s debut is good for some surprises – positive as well as negative. Before we get down to business, we have an important message for all readers and those interested in Arc:

The lead time for this test was anything but optimal, because Intel was suddenly in a hurry after Nvidia’s announcement of the Geforce RTX 4000. The (first) driver was provided by Intel on the evening of September 30th – a Friday before what was supposed to be a long weekend. We wouldn’t be the PCGH if that kept us from an extended extra shift, but five days or around 50 hours of testing is by far not enough to properly feel every tooth of a completely new architecture. We cover all the important points around performance, loudness, efficiency and compatibility, but not in the depth that we would like. If you miss something, let us know. Subsequent articles will expand on various topics (including XeSS, ray tracing, ReBAR and tuning potential) in more detail.

To start with, a summary of the most important cornerstones. The first generation of Intel Arc graphics cards is codenamed Alchemist and, corresponding to the letter, is referred to as the A series. The spotlight today is on the Arc A770 and Arc A750, with which Intel wants to pick up both Full HD and WQHD gamers – including ray tracing usability. We’ll see if that goal is met below. To get you in the mood, we recommend the following video, in which we unpack and examine the graphics cards:

Arc A770 and Arc A750: That’s what’s in it

Just a few weeks ago, we tested the A380 entry-level graphics card, checked the graphics driver’s gaming compatibility and interviewed Intel about the plans for the eagerly awaited “real” gaming graphics cards – these are now, with some delay, going on their maiden voyage. Both are based on the best that Intel’s graphics division currently has to offer: the ACM-G10, also known as the DG2 (Desktop Graphics 2). The “ACM” stands for “Alchemist”, the codename of the first arc series. ACM-G10 accommodates a proud 21.7 billion circuits on a 406 mm² core area and is manufactured in the N6 process at the Taiwanese chip forge TSMC.

The Arc flagship goes by the name A770 and houses a fully active ACM-G10 with 4,096 FP32 arithmetic units, 512 XMX units (comparable to Nvidia’s tensor cores) and 32 ray tracing units (RTUs). This graphics card will be available with both 16 and 8 GiByte and, according to Intel, will be sold over the counter for 349 or 329 US dollars (excl. taxes) – Intel does not name official euro prices. In addition, the Arc A750 is courting the favor of price-conscious gamers: It is based on a slightly slimmed-down ACM-G10 with 28 Xe cores, just as many ray tracing units, 3,584 FP32 ALUs and 448 XMX units, which, among other things, accelerate Intel’s XeSS upsampling be able. In combination with a slightly reduced clock, this graphics card should compete with Nvidia’s Geforce RTX 3060 at an attractive price of 289 US dollars.

The Arc A580, which was also announced, will not appear together with the A770 and A750. That’s unfortunate, because it leaves a gap in the conclusive, “hopefully eternal” naming scheme according to Intel. The plan is to split each Arc generation – as well as the Intel processors – into categories 3, 5 and 7. Whether a 9 will follow is just as open as the release date of the Arc A580. In any case, a more trimmed ACM-G10 than the A750 is used here, with which the A580 could compete with the Geforce RTX 3050. But that’s a long way off – let’s take a look at the Arc Launch portfolio alongside comparable AMD and Nvidia graphics cards:

Specifications of the performance with typical GPU boost according to the manufacturer. In practice, the frequency varies (usually it is higher) and therefore the throughput.

In case you missed the previous articles and are wondering about the “Limited Edition” designation: This is the equivalent of Nvidia’s “Founders Edition”, ergo the branding of the reference cards manufactured by Intel. This will exist both in retail and e-tail, i.e. both in regular and online trade. Despite the name, there is no artificial limitation to, for example, 5,000 pieces without replenishment, so there is no need to rush.

The Arc A770 Limited Edition (LE) is only available with 16 GiByte memory, with all eight 2 GiByte GDDR6 memory chips on the front of the board. Technically possible and approved by Intel are also 8 GiByte versions of the A770 from board partners. The latter may also offer 16 GiByte versions of the A750; the Limited Edition, however, always carries 8 GiByte. AMD and Nvidia used to let customers choose between two memory configurations, but this is the exception today. Intel is pursuing an aggressive pricing strategy: The surcharge for 16 instead of 8 GiByte on the Arc A770 Limited Edition reference card is only 20 US dollars, which represents an interesting added value for everyone who uses their graphics card for more than a few months. Incidentally, the A380 installs an appropriate 6 GiByte in any case.

The architecture of the Alchemist graphics cards was discussed by Intel more than a year ago, but it still deserves a few words of classification. Nvidia is not only the opponent chosen by Intel marketing, the Geforce makers also seem to serve as a source of inspiration at the chip level. Of course, the innards are only roughly described, but the Xe Render slices look like Nvidia’s shader multiprocessors in many respects. Worth mentioning is not only the very generous 16 MiByte level 2 cache – not to be confused with AMD’s “Infinity” level 3 cache – but also the dedicated matrix units for inferencing/AI calculations (XMX) as well the emphatically powerful ray tracing cores along with the thread sorting unit. The latter evoke memories of the Ada/RTX 40 presentation: Nvidia advertises the new GPUs with Shader Engine Reordering (SER), a feature that is intended to make ray tracing calculations much more efficient by sorting the tasks that are urgently needed. Interestingly, Arc’s TSU is supposed to work automatically, while Ada’s SER has to be implemented specifically by developers. More on that later.