AMD's Radeon boss has talked about the RDNA 3 GPU power efficiency, 12VHPWR on Radeon RX 7000 GPUs & ray tracing capabilities.
AMD's Radeon boss has talked about the RDNA 3 GPU power efficiency, 12VHPWR on Radeon RX 7000 GPUs & ray tracing capabilities.
The interview is very detailed and we would like you to visit Club386 and read the full thing here but some interesting comments were made regarding a few aspects of the RDNA 3 "Radeon RX 7000" GPU family and what we can expect in the coming generation.
Back when AMD was in the process of launching its RDNA 3 GPU architecture, the company promised a monumental +54% increase in power efficiency vs. RDNA 2 GPUs through the use of chiplets and other changes. However, the launch saw little gains in the efficiency department, all the while NVIDIA took their efficiency to a whole new level with the Ada GPU architecture. Scott says that AMD believes in offering good performance per watt across their GPU lineup & that it matters more on the notebook front. So far, AMD has only introduced its non-chiplet Navi 33 to laptops.
I was gonna say that it matters on laptops, and I don't know if I'd say that most people playing games are doing so on the desktop, but the actual quote is more reasonable than the title:
In notebooks, it matters greatly. In desktop, however, it matters, but not to everyone.
Damn. That was my last hope for a company putting efficiency as a priority.
(OP Asked)
I care about efficiency because I hate having to deal with the waste heat from my computer. It makes it more challenging to regulate indoor temp in summer where I live. Humidity is always high so keeping the house cool and dry for as cheap as possible is my highest priority.
I've seen people experimenting with using insulated flex duct out of a PC case exhaust out a window.
I would imagine that one could hypothetically have an intake, a case that seals well, and just have one of those window vent panels and route outside air in, through the PC, and then back outside.
It's pretty sizeable if you're running it 24/7 without checks.
A recently common example might be someone interested in running Stable Diffusion locally. Run the program overnight that's drawing ~300-400 watts for 12 hours. For a comparison, an electric heater can run up to 1,200 watts and those are known to absolutely raise the electric bill gone unchecked (like putting it on a timer vs. running over night).
For gaming 20 hours a week? Probably not too much. For gaming 20 hours a week and running AI a few times a week overnight (40 hours) it's noticeable for sure.
However there's also the factor of ambient heating, so there's technically some offset cost if you need heating... And it's also not something to ignore, I had a room consistently 68F with bad insulation up to 85F with a window open, hotter in the summers of course.
Overall, yes but also no. Like with most things it's really about the use case and consistency. NVIDIA GPU in a media server? Higher energy costs than something like an intel quicksync for limited realistic gains but somewhat noticeable cost increase. Gaming GPU running high idle all the time just browsing and watching videos? Definitely more expensive than just a laptop, but with proper checks like the gaming computer put in eco mode they're more equatable.
as with everything "it depends" on many aspects. In the US it's relatively cheap in many places (~0.12-0.15 USD per kwh), and high-end cards running at high settings can suck ~300w. Averaging 3hrs a day could cost ~36-45 cents a day just on the GPU alone for high-end settings. Not a huge deal, but in places where electricity is 2-3x the price it could be more of an issue (or at surge times for folks in Texas).
But with the new atx3.0 and 600w power cables, we could see double those costs for high end cards in the next few years, and putting full pc power at ~900+ watts.
Efficiency does affect more than just electricity costs. less efficient chips also means more heat and more massive coolers. Many of the higher end cards today have thick, heavy coolers that we now have anti-sag braces. It also potentially means more noise for fan coolers or requiring more expensive coolers and more electronics to regulate that power in stages.
Efficiency matters, it has always mattered. A card that uses less power, requires less cooling. Therefore it can be quieter, less warm(important during the summer), cheaper to make(smaller heatsink) and in the long run, significantly cheaper(less power consumption).
Efficiency equals performance. The more efficient your architecture is, the more performance you can get out of it. They are the same thing. Just because nvidia decided to scam everyone by significantly downgrading all non 4090 gpus, doesnt mean that this isnt true. Amd simply isnt competitive anymore so nvidia can afford to scam everyone and everyone continues buying nvidia.
Amd is barely cheaper(per performance) than nvidia. I am tired of all this bullshit internet propaganda about amd's offerings. All the reviewers are talking about how much better amd is, while amd cards offer 10% better rasterization performance at best(for the same money). And while all those reviewers run nvidia gpus and havent run an amd gpu on their daily machine in 10 years.
Just imagine if tomorrow the roles were reversed and nvidia was making gpus that used infinitely more power for equal performance with less features and shittier performance on cutting edge features. In fact, you dont need to imagine, this was the case for almost 15 years with dx9 and even dx12 cards. Take an amd 290 and compare it with an nvidia 780. Hell, nowadays even the 280x, a card that cost 50% the price of a 780, beats the nvidia 780.
Thats because amd's gpu architecture used to be amazing and infinitely more futureproof. But the current amd gpu architecture is clearly inferior and power consumption is a big indicator of that. Amd is forced to push so much power through their cards in order to get competitive performance. While nvidia is literally chopping the legs of its cards and lobotomizes them, in order to maximize profits and despite all that nvidia cards are still competitive. Thats how far ahead nvidia's architecture is atm.
Tbf, a part of this difference is because of the manufacturing node but thats a choice AMD made. The soon to be released 7800xt is similar in size to a 4080, a card that is basically twice as fast(and expensive). A card that has a big part of its architecture dedicated to ray tracing and ai cores. Nvidia isnt even trying, like at all.
You're right, and I upvoted you, but one issue is that unless people buy AMD cards anyway, they'll never have the research funds to ever match, much less beat Nvidia in the future. All things being equal (and they usually are for my needs) I tend to buy AMD over Nvidia to keep them in the race, so that GPU's don't turn into a monopoly. As it is, the duopoly we have now (excluding Intel, which I'm not sure will continue making consumer GPU's in the long run) isn't great for consumers right now.
The only reason AMD GPU's are even as good as they are now is due to the research money they got from Ryzen taking the fight to Intel.
Asking people to buy amd because it is "good enough" in order to avoid monopoly, isnt a sound business strategy. Thats why amd's gpu share has plummeted. There are basically zero amd 7xxx cards in steam charts, while all nvidia 4xxx cards are skyrocketing in market share. Every year, as more and more old amd gpus die, amd's market share drops a percent.
Intel cant even run the most anticipated game of the year, Starfield. How the fuck does that happen? Did they not contact Bethesda to ask for the code before the game was released?
The only reason AMD GPU's are even as good as they are now is due to the research money they got from Ryzen taking the fight to Intel.
Amd was getting Ls in the cpu market too, after intel released the core 2 duo. It wasnt till ryzen and especially ryzen 2 that amd bounced back. They need to do that with gpus too. And when they were taking Ls, at least they could offer more cores. What does amd gpus offer? More vram? Maybe the 7800xt will be good but it needs to be faster than 4070, while offering 16GB vram, for 100$ less.
Even if it has same rasterization performance for 100$ less than the 4070 might not be good enough. I dont think 12gB is going to be a significant bottlenecked in the future(outside of specific scenarios, which you can adjust a few settings) and people are willing to pay a bit more to get access to nvidia features(dlss, better ray tracing performance, lower power consumption, etc). I am still hopeful for the 7800xt, though the 7900xt has 40% more cores, so it is significantly cut down. But the 7800xt is using a higher frequency as a crutch.
People used to make fun of intel and nvidia for decades, for using so much power and pushing frequency just for their shitty architectures to be somewhat competitive. But when it comes to amd gpus, i guess that is acceptable because they are the underdog.
But personally, as someone living in a house without an ac, i'd rather not die while gaming during the summer. In fact, i am trying cloud gaming atm and i love how quiet and not hot my pc is while i am gaming on the cloud.
Well he's not wrong, people don't really factor in various power efficiency statistics when they buy graphics cards. Ultimately I feel PC gamers just want whatever Nvidia tells them to want. Everybody wants ray tracing and DLSS, even when these features are version-locked behind newer cards and only really useful for low-end (DLSS) and high-end (ray-tracing) cards at that. Going by Steam hardware survey, grand majority of people have little use for these features and yet you always see people clamor for them. I used to own a 2080 and I don't think I used either feature once.
What I want is better rasterization performance per dollar.
“Ray tracing” is a silly gimmick that contributes almost nothing and halves your frame rate. DLSS/FSR is upscaling, and upscaling is for consoles. To hell with them both.
I say “ray tracing” in scare quotes because what AMD and NVIDIA call “ray tracing” is nothing like the full-scene path tracing used in CGI movies. That is absolutely awesome, but it's far too slow to do in real time.
Also, I've heard a lot of complaints from CUDA programmers that AMD GPUs are pretty much useless for anything other than graphics, and NVIDIA is basically the only GPGPU game in town. I don't know the details, but AMD should probably work on whatever the problem is.
NVIDIA GPUs are definitely the go-to these days, but the world's most powerful supercomputer is using AMD GPUs. I wouldn't be surprised if AMD picks up speed (though they probably won't beat or meet NVIDIA). NVIDIA got started way sooner, so the fact AMD is behind is only natural.
AMD has partially been caught with its pants down on its GPUs. They're claiming that they're more power efficient, but really only in the context of laptops, handhelds and other embedded devices. They are not talking about the power/performance ratio, in which case nVidia easily beats them on the desktop.
They simply need a new GPU architecture that can handle all of the newer rendering concept being utilized these days. Desktop GPUs only use more power because we are looking at very demanding games that are ran at maximum settings. Eventually, these rendering ideas will show up on other platforms. At which point, either they admit graphical inferiority or waste more energy running those features than the competition.