Recently, the folks over at Chips and Cheese put the Asus ROG Ally Z1 up against the Steam Deck in a variety of GPU compute benchmarks, following a similar run of CPU benchmarking for the two entry-level PC handhelds. Chips and Cheese used Nemes' Vulkan benchmark suite for this testing, since their own OpenCL tests won't run on the Steam Deck's iGPU.
Below, we'll break down the tests run by Chips and Cheese in a list, including an explanation of what each test is actually measuring before discussing their results. For a more detailed look at their benchmarking finds (particularly their proper charts), consider the original piece— just be aware that their charts are also quite tall and full of many sub-results.
- Vulkan Floating Point Compute Throughput — One of the most common operations used in games.
- Vulkan Integer Compute Throughput — Slightly-less common than floating point operations, but proper integer compute still matters in games, as well as in AI workloads.
- Vulkan Vector Cache and Memory Latency — The lowest level of Cache on an RDNA GPU is Vector cache.
- Vulkan Scalar Cache and Memory Latency — AMD GPUs since GCN have used a separate scalar memory path with better latency optimizations than vector cache.
- Vulkan Cache and Memory Bandwidth — Both systems are using a configuration requiring shared memory with the CPU, thus this pertains to RAM used and not the GPU die necessarily.
- Memory Bandwidth to Compute Ratio — The metric is somewhat self-explanatory, though its relevance in isolation is questionable. More on this one a bit.
- CPU to GPU Link Bandwidth — This metric is also somewhat self-explanatory, and shows quite similar results across units due to similar unified SoC/APU design.
How does the Z1 ROG Ally compare to the Steam Deck?
In the Vulkan Throughput testing, Z1 ROG Ally traded blows with Deck depending on the workload. The newer iGPU architecture nets a lead in some areas (like FP32 and INT64 operations), but the Deck remains competitive and in some tests outright wins due to having more Compute Units on hand.
In Vulkan Cache and Memory testing, Z1 ROG Ally consistently offered lower latency than the Steam Deck. With Cache and Memory Bandwidth testing, Z1 ROG Ally maintains this lead.
Things get a little more complicated with the Memory Bandwidth to Compute Ratio testing, since it's more about architectural efficiency than raw performance. The Radeon 740M iGPU manages a .03 Memory Bandwidth to Compute Ratio while the full-fat RX 6900 XT desktop GPU manages .032— obviously, these devices are otherwise not directly comparable. More on that in the original piece.
Finally, in assessing the CPU to GPU Link bandwidth across six tests, the ROG Ally Z1 wins three out of six tests. The test results are generally pretty close, even with the Z1 Extreme Ally— which makes sense, as they're all using a unified CPU+GPU design.
In terms of the raw gaming performance touched upon at the end, the ROG Ally Z1 presents "mixed feelings". The high iGPU clocks make it surprisingly comparable to the Deck (particularly when at 30W), but the release of the Deck OLED after the Ally Z1 make it a far more questionable choice, and the Deck still has better overall performance.
The common wisdom for handheld PC gaming purposes at this point is that the Steam Deck LCD or Steam Deck OLED are your best bet for an entry-level experience. The ROG Ally Z1 Extreme performs quite similarly to these units when on battery, and only gets its big performance wins when plugged into a wall— as seen in one of our benchmarks of the Ally Z1 Extreme vs the Steam Deck below.
Thus, if you want to go with the Asus Ally for some extra performance, saving up the extra $100 for the Z1 Extreme is the better use of your money — and these results from Chips and Cheese further reinforce just how suboptimal the non-Extreme Ally is, as it loses to the Deck in gaming performance.
