This article is more than 1 year old
Intel challenges Nvidia, AMD with trio of workstation GPUs
For those that just can't wait for AV1 encoding
Intel unveiled its answer to the AMD's FirePro and what used to be Nvidia's Quadro workstation GPUs this week with the launch of a trio of new graphics cards aimed at professional applications like architectural design, engineering, and content creation.
Intel's professional lineup — marketed as the Arc Pro series — includes a mobile-workstation GPU called the A30M, the single slot A40, and the dual slot A50. Both desktop cards feature 6GB of GDDR6 memory, with the biggest differentiator between the two being the A50's dual-slot cooler allows for a higher 75W power envelope. Meanwhile, the A30M is saddled with 4GB of GDDR6 and a TDP of between 35-50W.
Curiously, Nvidia also sells an A40 GPU, but that card ships with 48GB of GDDR6 and claims between 37.4 and 74.8 teraflops of FP32 performance while consuming 300W.
Despite the similarities in naming, Intel's Arc Pro cards target a very different demographic — mostly low-end workstations and applications that can take advantage of the platform's strengths.
Did someone say AV1
Intel's top-tier Arc Pro A50 offers peak performance of 4.8 teraflops of FP32 performance, which according to Moor Insights and Strategy analyst Anshel Sag, puts the card somewhere between Nvidia's T1000 (2.5 teraflops) and its recently released A2000 (8 teraflops).
However, some of that performance delta may be attributable to a limitation required to fit within the PCIe slot's 75W power budget. For example, the Nvidia GA106 used in the A2000 is the same used in the RTX 3060, which boasts more than 12 teraflops of FP32 performance but consumes roughly twice the power in doing so.
However, Intel's performance isn't all that surprising or even unexpected for a first-gen product, Sag notes.
"Arc, in general, was never intended — at least at launch — to compete with the highest performance graphics cards from AMD or Nvidia," he said. "They were always going to start entering the market at a more mainstream target profile, as far as consumers go, because building a flagship GPU is very difficult; it's very expensive; and its actually a fairly low-volume business."
He explained that while products like Nvidia's 3090 TI may capture people's attention, they are very low volume, decent margin, and ultimately make up very little of the company's profits. "In reality, the bulk of the revenue and profit is in the middle," Sag said.
But while Sag doesn't find it surprising that Intel went after the low-to-mid-tier graphics market for its first crack at Arc, he was surprised by the company's decision to launch a workstation-class chip at this time.
- Too little, too late: Intel's legacy is eroding
- Intel's first discrete GPUs won't be a home run
- Intel tried selling software before. Will it succeed this time?
- Intel delivers first discrete Arc desktop GPUs ... in China
However, he notes that capabilities like AV1 encoding - which Intel claims it's the first to offer - may be valuable to application developers and content creators that want to get a head start working with the codec. "When it comes to any kind of rendering, faster is always better, as long as you don't compromise on quality," Sag added.
Arc Pro also enables Intel to approach OEMs with a relatively inexpensive workstation part that will be good enough for a variety of workloads. However, Intel clearly believes these cards won't be limited to OEM builds and makes a point of calling out their PCIe 4.0 8x interfaces and PCIe 3.0 backward compatibility.
Given the relatively low performance target for these cards, Intel likely could have gotten away with a PCIe 4.0 4x interface, as AMD has done with many of its recent entry-level gaming and workstation cards. Doing so likely would have compromised performance on older systems.
Can Intel oust AMD in workstation graphics?
Sag notes that while Intel has its work cut out when comes to competing with AMD and Nvidia in gaming graphics, the company stands a far better chance of being competitive in the workstation market.
Here, AMD's offerings are relatively weak compared to that of Nvidia, Sag said. By comparison, Intel has something of a advantage because its "already spent a lot of time and effort qualifying their CPUs for these exact workloads and has existing relationships with all the ISVs."
In its marketing materials, Intel highlighted support for several popular software platforms and frameworks, including Handbrake, Premier Pro, Davinci Resolve, and Gigapixel AI.
This is particularly true around Intel's media-encoding capabilities, which it calls QuickSync on its onboard graphics. And with Arc Pro, Intel claims it can split encode tasks up between the onboard and discrete graphics when paired with a compatible GPU.
Don't count out Intel Arc yet
The launch also casts doubt on rumors that Intel's graphics division was destined for the chopping block following a painful second-quarter earnings. Late last month, the company reported a $454 million net loss and axed its Optane persistent memory and SSD business.
"They're definitely struggling a little bit on the drivers side," Sag said, noting that it's not unusual for drivers to improve considerable over the course of a GPU's lifespan.
"This workstation launch kind of negates a lot of the rumors that they're going to shutdown their discrete business that they haven't even really spun up yet," he added, arguing that many of these rumors are likely borne out of fear that Intel could pose a threat to Nvidia and AMD's duopoly.
Sag adds, that in his calculation, Intel's graphics business is on a different trajectory compared to its other product divisions and has been separated out into its own product division for that very reason. ®