If you’ve bought a laptop computer or pill with an AMD Ryzen chip inside, there’s a efficiency tweak you completely have to learn about. Apply it to AMD’s AI/gaming powerhouse Ryzen AI Max chip, and you may rating monumental efficiency positive factors in simply seconds!
Savvy players know instinctively that you would be able to increase your recreation’s body price by reducing the decision or the visible high quality, or by making an adjustment to the Windows power-performance slider. But the Ryzen AI Max is a brand new type of machine: a killer cell processor that may run trendy video games at elevated body charges, and function an AI powerhouse.
What’s the key? A easy adjustment of the Ryzen AI Max’s unified body buffer, or obtainable graphics reminiscence. While it’s a easy repair, in my assessments, it made an unlimited distinction: as much as a 60 % efficiency increase in some circumstances.
When I used to be evaluating Intel’s “Arrow Lake” mobile processor to AMD’s Ryzen AI 300 processor, I used to be unable to carry out sure AI assessments due to the reminiscence and VRAM necessities these assessments required. After I revealed the story, AMD reached out to recommend dialing up the VRAM by way of the laptop computer BIOS to allow the check to run. (As it turned out, this didn’t make a distinction in that particular check.)
When Asus ROG supplied me an Asus ROG Flow Z13 gaming pill for testing AMD’s Ryzen AI Max+ chip, the subject surfaced once more. For no matter motive, Asus had configured its pill with simply 4GB of accessible video reminiscence. The firm really helpful we check with 8GB as a substitute. It gave the impression to be an opportune time to see what results adjusting the Ryzen AI Max’s graphics reminiscence would have.
Why does video reminiscence matter?
VRAM stands for video RAM. If you employ a discrete GPU, the quantity of VRAM is predetermined by your graphics-card producer, and the VRAM chips are soldered proper onto the board. Some folks use “VRAM” as shorthand when speaking about built-in graphics, as nicely. That’s not solely correct; built-in graphics share reminiscence between the PC’s important reminiscence and the video logic, and the “VRAM” on this case is named the “UMA frame buffer.”
VRAM (and the UMA body buffer) shops the textures utilized by a PC recreation inside your graphics card, permitting them to be shortly accessed and rendered upon the display screen. If the feel measurement exceeds the quantity of VRAM, your PC should pull them in from some place else (RAM or the SSD), slowing down gameplay.
In the case of AI, the VRAM acts loads like commonplace PC RAM, storing the weights and fashions of an AI algorithm, and permitting it to run. In many circumstances, inadequate VRAM signifies that a selected LLM may not run. In each circumstances, although, the quantity of reminiscence obtainable to the GPU issues.
If you personal a discrete graphics card from any producer, the quantity of VRAM can’t be adjusted. On Intel Core techniques, the obtainable UMA body buffer can also be sometimes mounted, to half of the available system memory. In different phrases, in case you have an Intel Arrow Lake laptop computer with 16GB of whole reminiscence, as much as 8GB is accessible by the built-in GPU. And sure, including extra system reminiscence will increase the dimensions of the UMA body buffer. However, you continue to can’t modify the UMA body buffer’s measurement to your personal preferences.
Mark Hachman / Foundry
AMD, however, sometimes permits you to adjust the size of the UMA frame buffer on its Ryzen processors (and never simply the AI Max!), sometimes by means of changes made to the firmware. In my case, I obtained entry to the UEFI/BIOS simply by rebooting the pill and tapping the F2 key because it rebooted.
On the ROG pill, I discovered the adjustment within the superior settings. Several choices had been obtainable, from an “Auto” setting to 24GB, with a number of increments in between. As I famous above, Asus had unintentionally despatched out its tablets with the 4GB setting enabled. All I did was choose a brand new measurement for the body buffer, save, and the boot cycle continued.
Mark Hachman / Foundry
Warning: If you personal a handheld PC like a Steam Deck, adjusting the UMA body buffer to enhance recreation efficiency isn’t that uncommon. Making changes to the BIOS/UEFI does carry some danger, nevertheless. In this case, allocating an excessive amount of reminiscence to the GPU may not enable an utility to run in any respect, or trigger instability. You’re most likely protected leaving 8GB reserved for important reminiscence.
If you don’t wish to fiddle round in your laptop computer’s BIOS, you may discover that your laptop computer or pill ships with the AMD Adrenalin utility. In this case, you may be capable to modify the graphics reminiscence from inside the utility itself. I discovered it inside the Performance > Tuning tab. If you’re confused about how a lot reminiscence you’re allocating to graphics and the way a lot is left to your PC, the utility helps make that clear.
Mark Hachman / Foundry
Tested: Does adjusting the UMA body buffer make a distinction?
I’m presently within the technique of reviewing the Asus ROG Flow Z13 and its AMD Ryzen AI Max+ chip, however I paused to test to see what enhancements, if any, could possibly be made by adjusting the dimensions of the UMA body buffer. In common, I discovered that adjusting it made probably the most enchancment inside video games and particularly AI.
I didn’t see monumental positive factors in artificial graphics benchmarks: Tweaking the UMA body buffer utilizing 3DMark’s Steel Nomad check boosted scores from 1,643 to 1,825, or about 11 %. Ditto for PCMark 10, which measures productiveness: The needle barely moved. I turned to Cyberpunk: 2077 to check how adjusting the UMA body buffer utilized to video games.
I ran 4 benchmarks, every utilizing a UMA body buffer of 4GB, 8GB, or 16GB. The concept was to see what the chip might do by itself, simply rasterizing the sport, in addition to turning on all the GPU’s AI enhancements to maximise body price:
- 1080p decision, Ultra settings, with all enhancements turned off
- 1080p decision, Ultra settings, with all enhancements (AMD FSR 3 scaling and body technology) turned on
- 1080p decision, Ray Tracing Overdrive settings, with all enhancements turned off
- 1080p decision, Ray Tracing Overdrive settings, with all enhancements turned on
Mark Hachman / Foundry
You can instantly see that the biggest enhancements in general body price merely come from turning on AMD’s FidelityFX Super Sampling 3 (FSR 3). But adjusting the body buffer offers you a couple of 10 % increase within the Ray Tracing Overdrive mode. More importantly, it bumped up the body price on the 1080p Ultra settings, with FSR 3 on, by 17 %. That’s all from a free, straightforward adjustment that prices you nothing.
It’s price stating that adjusting the UMA body buffer doesn’t all the time scale. It appears to underneath the Ray Tracing Overdrive setting, however the identical positive factors don’t play out elsewhere. I’ve discovered Cyberpunk‘s tests to be quite reproducible, so I’m inclined to consider that it’s not simply statistical variance.
I noticed the identical factor when testing AI, too. I used UL Procyon’s AI Text Generation benchmark, which hundreds in 4 separate LLMs or AI chatbots, and asks a collection of questions. UL generates its personal scores, based mostly on velocity and token output. Here, there was a tremendous soar in general efficiency in locations: 64 % in Llama 2 simply by dialing up the body buffer! But it was additionally fascinating that the efficiency enhance was “capped” at 16GB. Selecting a 24GB body buffer really lowered the efficiency barely.
Mark Hachman / Foundry
(It is perhaps price noting that Procyon didn’t assume it might run on the Ryzen AI Max, as a result of the system didn’t report sufficient VRAM. It ran simply high quality, after all.)
I additionally checked my work utilizing MLCommons’ personal consumer benchmark, which makes use of a 7 billion-parameter LLM to generate “scores” of tokens per second and an preliminary “time to first token.” I didn’t see as a lot enchancment — nearly 10 % from the 4GB to 8GB buffer, after which just about nothing after I examined it with a 16GB buffer as a substitute.
Integrated graphics issues once more
Until now, adjusting the UMA body buffer was largely irrelevant: Integrated GPUs had been merely not highly effective sufficient for that tweak to matter. Things are completely different now.
Again, if you happen to’re a handheld PC consumer, you’ll have seen others adjusting their UMA body buffer to eke out a bit extra efficiency. But as my colleague Adam Patrick Murray commented, the Ryzen AI Max+ processor contained in the Asus ROG Flow Z13 pill is an odd hybrid of a handheld and a pill. It’s greater than that, after all, on condition that corporations like Framework are placing it in small form-factor PCs.
But classes realized from the hand held PC house do apply — and nicely! — to this new class of AI powerhouse chips. Adjusting the UMA body buffer on an AMD Ryzen AI Max+ chip is usually a terrific free tweak that may increase your efficiency by dramatic quantities.