The Intel Arc A380 must be one of many worst graphics card launches in historical past — not the {hardware} itself, essentially, however the retail launch of the {hardware}. By all indications, Intel knew the drivers had been damaged when the {hardware} was prepared for launch earlier this 12 months. Slightly than taking ample time to repair the drivers earlier than the retail launch, and with the clock ticking as new AMD and Nvidia GPUs are on the horizon, Intel determined to ship its Arc GPUs first in China — probably not the form of method an organization would take if the product had been worthy of creating our checklist of the finest graphics playing cards.
A number of months later, after loads of unfavorable publicity courtesy of GPUs that made their strategy to different shores, and with quite a few driver updates come and gone, Arc A380 has formally launched within the US with a beginning worth of $139 (opens in new tab). The one providing on Newegg bought out and is at present again ordered, however that is probably extra to do with restricted provides than excessive demand. Nonetheless, the A380’s not all unhealthy, and we’re blissful to see Group Blue rejoin the devoted GPU marketplace for the primary time in over 24 years. (And no, I do not actually matter the Intel DG1 from final 12 months, because it solely labored on particular motherboards.)
How does the Arc A380 stack as much as competing AMD and Nvidia GPUs, and what’s all of the hype about AV1 {hardware} encoding acceleration? You possibly can see the place it lands in our GPU benchmarks hierarchy, which if you need a spoiler is… not good. However let’s get to the small print.
Arc Alchemist Structure Recap
We have offered in depth protection on Intel’s Arc Alchemist structure, courting again to about one 12 months in the past. On the time we first wrote that piece, we had been anticipating a late 2021 or early 2022 launch. That morphed right into a deliberate March 2022 launch, then finally a mid-2022 launch — and it is not even a full launch, at the very least not but. Arc A380 is merely the primary salvo, on the very backside of the worth and efficiency ladder. We have seen loads of hints of the quicker Arc A750, which seems to be near RTX 3060 efficiency primarily based on Intel’s personal benchmarks, and that ought to launch throughout the subsequent month or so. What in regards to the quicker nonetheless Arc A770 or mid-tier Arc A580 and different merchandise? Solely time will inform.
Arc Alchemist represents a divergence from Intel’s earlier graphics designs. There’s in all probability loads of overlap in sure components, however Intel has modified names for a number of the core constructing blocks. Gone are the “Execution Items (EUs),” which are actually known as Vector Engines (VEs). Every VE can compute eight FP32 operations per cycle, which will get loosely translated into “GPU cores” or GPU shaders and is roughly equal to the AMD and Nvidia shaders.
Intel teams 16 VEs right into a single Xe-Core, which additionally contains different performance. Every Xe-Core thus has 128 shader cores and roughly interprets as equal to an AMD Compute Unit (CU) or Nvidia Streaming Multiprocessor (SM). They’re mainly all SIMD (single instruction a number of information) designs, and just like the competitors, Arc Alchemist has enhanced the shaders to satisfy the complete DirectX 12 Final characteristic set.
That naturally means having ray tracing {hardware} integrated into the design, and Intel has one Ray Tracing Unit (RTU) per Xe-Core. The precise particulars of the ray tracing {hardware} aren’t fully clear but, although primarily based on testing every Intel RTU may match up decently in opposition to an Nvidia Ampere RT core.
Intel did not cease there. Alongside the VEs and RTUs and different typical graphics {hardware}, Intel additionally added Matrix Engines, which it calls XMX Engines (Xe Matrix eXtensions). These are comparable in precept to Nvidia’s Tensor cores and are designed to crunch although a lot of much less exact information for machine studying and different makes use of. An XMX Engine is 1024-bits broad and might course of both 64 FP16 operations or 128 INT8 operations per cycle, giving Arc GPUs a comparatively great amount of compute energy.
Intel Arc A380 Specs
With that transient overview of the structure out of the way in which, listed here are the specs for the Arc A380, in comparison with a few competing AMD and Nvidia GPUs. Whereas we offer theoretical efficiency right here, keep in mind that not all teraflops and teraops are created equal. We’d like real-world testing to see what kind of precise efficiency the structure can ship.
| Graphics Card | Arc A380 | RX 6500 XT | RX 6400 | GTX 1650 Tremendous | GTX 1650 |
|---|---|---|---|---|---|
| Structure | ACM-G11 | Navi 24 | Navi 24 | TU116 | TU117 |
| Course of Know-how | TSMC N6 | TSMC N6 | TSMC N6 | TSMC 12FFN | TSMC 12FFN |
| Transistors (Billion) | 7.2 | 5.4 | 5.4 | 6.6 | 4.7 |
| Die measurement (mm^2) | 157 | 107 | 107 | 284 | 200 |
| SMs / CUs / Xe-Cores | 8 | 16 | 12 | 20 | 14 |
| GPU Cores (Shaders) | 1024 | 1024 | 768 | 1280 | 896 |
| Tensor Cores | 128 | — | — | — | — |
| Ray Tracing ‘Cores’ | 8 | 16 | 12 | — | — |
| Base Clock (MHz) | 2000 | 2310 | 1923 | 1530 | 1485 |
| Increase Clock (MHz) | 2450 | 2815 | 2321 | 1725 | 1665 |
| VRAM Pace (Gbps) | 15.5 | 18 | 16 | 12 | 8 |
| VRAM (GB) | 6 | 4 | 4 | 4 | 4 |
| VRAM Bus Width | 96 | 64 | 64 | 128 | 128 |
| ROPs | 32 | 32 | 32 | 48 | 32 |
| TMUs | 64 | 64 | 48 | 80 | 56 |
| TFLOPS FP32 (Increase) | 5 | 5.8 | 3.6 | 4.4 | 3 |
| TFLOPS FP16 (MXM/Tensor if Obtainable) | 40 | 11.6 | 7.2 | 8.8 | 6 |
| Bandwidth (GBps) | 186 | 144 | 128 | 192 | 128 |
| Video Encoding | H.264, H.265, AV1, VP9 | — | — | H.264, H.265 (Turing) | H.264, H.265 (Volta) |
| TDP (watts) | 75 | 107 | 53 | 100 | 75 |
| Launch Date | Jun 2022 | Jan 2022 | Jan 2022 | Nov 2019 | Apr 2019 |
| Launch Value | $139 | $199 | $159 | $159 | $149 |
On paper, Intel’s Arc A380 mainly competes in opposition to AMD’s RX 6500 XT and RX 6400, or Nvidia’s GTX 1650 Tremendous and GTX 1650. It is priced barely decrease than the competitors, particularly present on-line costs for brand new playing cards, with roughly comparable options. There are some vital {qualifications} to notice, nonetheless.
Nvidia would not have ray tracing {hardware} under the RTX 3050 (or RTX 2060). Equally, not one of the AMD or Nvidia GPUs on this phase help tensor {hardware} both, giving Intel a possible benefit in deep studying and AI functions — we have included FP16 throughput for the GPU cores on the AMD and Nvidia playing cards by the use of reference, although that is not fully apples-to-apples.
Intel is the one GPU firm that at present has AV1 and VP9 {hardware} accelerated video encoding. We’re anticipating AMD and Nvidia so as to add AV1 help to their upcoming RDNA 3 and Ada architectures, and probably VP9 as properly, however we do not have official affirmation on how that may play out. We’ll take a look at encoding efficiency and high quality later on this evaluation as properly, although be aware that the GTX 1650 makes use of Nvidia’s older NVENC {hardware} that delivers a decrease high quality output than the newer Turing (and Ampere) model.
The Arc A380 has theoretical compute efficiency of 5.0 teraflops, which places it barely behind the RX 6500 XT however forward of all the pieces else. It is also the one GPU on this worth class to ship with 6GB of GDDR6 reminiscence, with a 96-bit reminiscence interface. That offers the A380 extra reminiscence bandwidth than AMD however with out Infinity Cache, and fewer reminiscence bandwidth than Nvidia’s GPUs. Energy use targets 75W, although overclocked playing cards can exceed that, similar to with AMD and Nvidia GPUs.
The ray tracing capabilities are tougher to pin down. To shortly recap, Nvidia’s Turing structure on the RTX 20-series GPUs had full {hardware} ray tracing capabilities, and every RT core can do 4 ray/triangle intersection calculations per cycle, plus there’s {hardware} help for BVH (bounding quantity hierarchy) traversal. Nvidia’s Ampere structure added a second ray/triangle intersection unit to the RT cores, doubtlessly doubling the throughput. In observe, Nvidia says Ampere’s RT cores are usually 75% quicker, as it may well’t all the time fill all of the execution slots. AMD’s RDNA 2 structure is much like Turing in that it may well do 4 ray/triangle intersection calculations per Ray Accelerator. Nevertheless, it makes use of GPU shaders for BVH traversal, which is slower and extra reminiscence intensive.
From what we will confirm (and we have requested Intel for clarification), Intel’s RTUs are much like Nvidia’s Turing RT cores in that they will do 4 ray/triangle intersections per cycle. We’re nonetheless making an attempt to find out in the event that they embrace BVH traversal {hardware}, or in the event that they’re extra like AMD’s Ray Accelerators and use GPU shaders for the BVH traversal. It sounds as if they embrace BVH {hardware} and may really be fairly first rate.
Nonetheless, with solely eight RTUs, the A380 undoubtedly will not be a ray tracing powerhouse — Nvidia for instance has 20 or extra RT cores in its RTX lineup, or 16 should you embrace the cell RTX 3050 within the checklist, Nvidia’s slowest RTX chip. AMD however has as few as 12 Ray Accelerators in its RX 6000-series elements, and built-in RDNA 2 implementations just like the Steam Deck can have as few as eight RAs, although RT efficiency understandably suffers quite a bit — not that you simply want ray tracing, even 4 years after {hardware} first supported the performance.
