Intel’s Kaby Lake-G Radeons Might Use Hybrid Polaris-Vega Solution

Ever since AMD and Intel announced they’d be releasing an Intel CPU married to an AMD GPU core, they’ve maintained the IP AMD contributed to the GPU in question was based on Vega, not Polaris. The chip is marketed as RX Vega and is baked into several Intel SKUs, including the recently-launched Hades Canyon NUC (we’ve got one on the test bench and will be discussing it in the not-too-distant future).

All of the initial messaging around the part from AMD and Intel have labeled it a Vega-class GPU, as have a number of diagnostic applications. But as PCWorld notes, AIDA64 reports the Core i7-8809G as being based on a “Polaris 22.” In and of itself, this doesn’t mean much. Applications like AIDA64 (and SiSoft Sandra, HWMonitor, etc) use a database of application IDs to determine what kind of functionality is supported by a given processor or graphics card. Sometimes these apps will return incorrect identifiers or flags if they haven’t been properly updated.

But then others noticed that the integrated RX Vega on Kaby Lake-G identified itself as supporting DirectX 12_0 features. Polaris, AMD’s earlier GPU core, offers DirectX 12_0 support. Vega, in contrast, is a DirectX 12_1 chip (for more on DirectX 12 features levels, see this article). Again, these capabilities are self-reported, so it doesn’t prove anything, but it’s a further oddity. Finally, Tom’s Hardware set out to check whether the Core i7-8809G supports a key Vega feature, rapid packed math. Rapid packed math allows Vega to pack two 16-bit operations into a single 32-bit register, and while most games don’t support it, SiSoft Sandra has a baked-in test.

Now, keep in mind, these results don’t tell us that Kaby Lake-G can’t perform rapid-packed math — but they do indicate RPM is not currently enabled on the chip. When reached for comment, Intel’s response was:

This is a custom Radeon graphics solution built for Intel. It is similar to the desktop Radeon RX Vega solution with a high bandwidth memory cache controller and enhanced compute units with additional ROPs.

So where does that leave us?

The GPU baked into Intel’s Core i7-8809G absolutely has some Vega-class features. HBM2, for example, is clearly Vega-only. Intel refers to enhanced compute units, an HBM cache controller, and additional ROPs — all capabilities that reflect a higher-end GPU than anything we ever saw from the Polaris family (the Core i7-8809G has more ROPs than AMD’s Polaris-derived GPUs).

If it Waddles Like a Vaguely Duck-Shaped Object…

It could be a duck. Alternately, it might be one of these:

It’s not unusual for GPU companies to pull old silicon into new GPU families, sometimes for multiple generations in a row. AMD and Intel have both signed off on this RX Vega branding, and clearly neither were in the dark about what that entailed. We know that no other integrated GPU offers the kind of performance you’ll find with the Core i7-8809G, which makes the entire argument a bit of a moot point. After all, if you’re happy with the performance Intel’s latest NUC offers, you’re happy with it. The difference between DirectX 12_0 and 12_1 isn’t one any gamer is going to notice, and the GPU component of the 8809G absolutely has certain Vega features, like the HBMCC, HBM2 itself, and higher ROP counts. Intel’s reference to Vega’s enhanced compute units imply at least some Vega compute capabilities are baked in as well.

The final piece of the puzzle is that AMD’s semi-custom abilities and expertise mean that not every GPU is going to cleanly conform to a single concept. This chip may be a hybrid solution ultimately marketed under the Vega brand, but if AMD and Intel are happy calling it a Vega, it’s probably fair game to do so — especially since there’s really nothing else like it in-market in the first place. Of course, it’s also possible certain functions like rapid packed math might be enabled at a later date as well.

Continue reading

Intel’s Raja Koduri to Present at Samsung Foundry’s Upcoming Conference

Intel's Raja Koduri will speak at a Samsung foundry event this week — and that's not something that would happen if Intel didn't have something to say.

Ryzen 9 5950X and 5900X Review: AMD Unleashes Zen 3 Against Intel’s Last Performance Bastions

AMD continues its onslaught on what was once Intel's undisputed turf.

Leaked Benchmarks Paint Conflicting Picture of Intel’s Rocket Lake

Rumors about Rocket Lake have pointed in two opposite directions recently, but the more competitive figures are more likely to be true.

Intel’s Iris Xe Max Discrete GPU Is Slower Than the Integrated Version

Intel's Iris Xe Max has debuted, but the discrete GPUs performance is a bit odd, as new benchmarks show.