All aboard the Intel Arc! Or Ark? Intel's first true dedicated graphics cards are here, and we've got the Arc A750 review along with the Arc A770 review. Intel has set its sights on the value midrange market with the A750, hoping to earn a spot among the best graphics cards. We've thoroughly tested the card and have come away far more impressed than we were with the budget-friendly Arc A380.
This is a companion review to the A770, so there's a lot of additional detail in that article. You can also check out our deep dive into the Intel Arc Alchemist architecture. For the A750, we'll skip straight to the important bits and render our verdict.
Intel Arc A750 Specifications
Here's the quick rundown of the full Intel Arc desktop card lineup. The A580 hasn't launched yet, so we're missing some information like the price, but everything else should now be available. That's not quite true, as the A750 and A770 will actually go on sale next week, on October 12 — right alongside the Nvidia RTX 4090, which promises gobs more performance for slightly more than five times the price of the Arc A750. It's going to be a tough decision, we know!
|Graphics Card||Arc A770 16GB||Arc A770 8GB||Arc A750||Arc A580||Arc A380|
|Process Technology||TSMC N6||TSMC N6||TSMC N6||TSMC N6||TSMC N6|
|Die size (mm^2)||406||406||406||406||157|
|Ray Tracing Units||32||32||28||24||8|
|Boost Clock (MHz)||2100||2100||2050||1700||2000|
|VRAM Speed (Gbps)||17.5||16||16||16||15.5|
|VRAM Bus Width||256||256||256||256||96|
|TFLOPS FP16 (INT8)||138 (275)||138 (275)||118 (235)||84 (167)||33 (66)|
|Launch Date||October 2022||October 2022||October 2022||?||June 2022|
The Intel Arc A750 follows the familiar pattern of taking the same GPU and core design of a more expensive model and then trimming down some features. The A750 disables four of the potential 32 Xe-Cores and has a slightly lower boost clock, giving it about 85% of the theoretical compute of the A770. It also has half the memory of the 16GB A770 Limited Edition, clocked at 16 Gbps instead of 17.5 Gbps, so that's 9% less memory bandwidth. Memory capacity is our bigger concern.
8GB of VRAM was great back in 2016 when the GTX 1070 and 1080 launched. However, six years later, we're not quite as keen on only having 8GB of memory. Granted, the RX 6600-series cards from AMD are all packing 8GB, and everything from the RTX 3050 through the RTX 3070 Ti — with the exception of the RTX 3060 — also has 8GB. But one look at where games are heading with VRAM use, and we can't help but think the A770 16GB is probably worth the extra $60. The extra memory could definitely help big brother leave his little sibling sucking wind.
Looking at the direct competition, which, based on current GPU prices, would be the AMD RX 6600 or RX 6650 XT and the Nvidia RTX 3050, things are a bit messy. Let's just get this out of the way and say that the RTX 3050 ends up hopelessly outclassed. That was already true with the RX 6600, and the Arc A750 can pour some salt into the wound. But the AMD cards aren't going to roll over so easily.
In short, AMD promises good performance and excellent value for people that aren't worried about ray tracing or fancy schmancy AI upscaling technologies. On the other hand, Intel and Nvidia offer much better ray tracing performance along with matrix cores that can boost machine learning and AI performance.
With 28 Xe-Cores and a nominal 2050 MHz boost clock — and we say "nominal" because, as you'll see later, the Arc A750 easily exceeded that mark in our testing — Intel offers 14.7 teraflops of graphics compute performance. For deep learning and AI workloads, the A750 can perform 118 teraflops of FP16 calculations, or double that for INT8 work with 235 teraops of compute. Don't worry too much about the loss in precision, as companies like Google and Facebook have proven that 8-bits is sufficient for such work.
By comparison, AMD's RX 6650 XT has 10.8 teraflops of compute, while the RTX 3050 putters along with a meager 9.0 teraflops. Of course, that's on paper, and you can't just look at theoretical numbers to determine a winner — we've seen AMD's RDNA 2 chips punch well above their theoretical specs over the past two years. And that's why we run the benchmarks.
- MORE: Best Graphics Cards
- MORE: GPU Benchmarks and Hierarchy
- MORE: All Graphics Content
Power usage and temperatures are less than desired.
My biggest concern is future support. They said they're committed to dGPUs, but this product line clearly didn't live up to their expectations. Unless we're all being horribly lied to on GPU pricing, it doesn't seem like Intel is making much/any money on the A750/770. Certainly not as much as they'd hoped. If next gen is a flop also.....who knows, maybe they call it quits. Then what? Would they still provide driver updates? For how long?
I do wonder what % of games released in the past 2 years (say top 100 from each year) are DX12....
I don't really think they're going to ax the GPU division, though. Intel needs high density compute, just like Nvidia needs its own CPU. There are big enterprise markets that Intel has been locked out of for years due to not having a proper solution. Larrabee was supposed to be that option, but when it morphed into Xeon Phi and then eventually got axed, Intel needed a different alternative. And x86 compatibility on something like a GPU (or Xeon Phi) is going to be more of a curse than a blessing.
I really do want Intel to stay in the GPU market. Having a third competitor will be good. Hopefully Battlemage rights many of the wrongs in Alchemist.
I mean I love Control and everything, but I've been done with it for years. I googled "upcoming ray tracing games" and the top result was still that original list from 2019.
There's so few noteworthy RT games, that I'm surprised that Intel and the next gen cards are even bothering to support it.
Also, I'm not really understanding how the hypothetical system cost that was discussed would be factored into the math.
As more affordable yet sufficiently powerful RT hardware becomes capable of pushing 60+FPS at FHD or higher resolutions, we'll see more games using.
It was the same story with pixel/vertex shaders and unified shaders. Took a while for software developers to migrate from hard-wired T&L to shaders, give it a few year and now fixed-function T&L hardware is deprecated.
Give it another 5-7 years and we'll likely get new APIs designed with RT as the primary render flow.
Hi, I have some questions and a request
Does this support PCIe 3.0x16.
For Low end GPU could you select a low end GPU like my Ryzen 5700G. this would tell me 3 things -- support for AMD, Support for PCIe 3.0, and use for low end CPU
I have absolutely no idea what this says.