Slashdot Mirror
AMD Announces Radeon VII, Its Next-Generation $699 Graphics Card (theverge.com)
An anonymous reader shares a report: AMD has been lagging behind Nvidia for years in the high-end gaming graphics card race, to the point that it's primarily been pushing bang-for-the-buck cards like the RX 580 instead. But at CES, the company says it has a GPU that's competitive with Nvidia's RTX 2080. It's called the Radeon VII ("Seven"), and it uses the company's first 7nm graphics chip that we'd seen teased previously. It'll ship on February 7th for $699, according to the company. That's the same price as a standard Nvidia RTX 2080. [...] AMD says the second-gen Vega architecture offers 25 percent more performance at the same power as previous Vega graphics, and the company showed it running Devil May Cry 5 here at 4K resolution, ultra settings, and frame rates "way above 60 fps." AMD says it has a terabyte-per-second of memory bandwidth.
Yes, AMD is lagging behind but I will still go with AMD graphics over NVIDIA because NVIDIA has an anti-open source stance. It's good news that AMD's graphics chipsets are getting better.
This is rather disappointing really. It's just a Vega refresh that offers ~30% improved performance in most workloads or frame rates, but only at 40% additional cost compared to Vega 64. I suppose it's nice if you need more than 8GB of memory, but this isn't anything to get excited about as far as I'm concerned.
At least the sneak peek at the new Ryzen CPUs looked promising.
Comment removed based on user account deletion
Dmc5 isn't even out yet. So the fact that it runs it on ultra at above 60 FPS in 4K at that. Is a good thing.
Realistically it will be 5 to 7 percent if history is any indicator.
I mean, if it is supposedly "competitive" with the Nvidia RTX 2080, which means.. almost as fast and the same price as the Nvidia RTX 2080, why would you not buy the Nvidia RTX 2080?
It's almost like you haven't seen what a new 1080ti or rtx2800 costs right now
Yeah, real time ray tracing isn't really a thing yet. They haven't really built video cards powerful enough to enable it without a huge performance hit.
Maybe the next gen cards will support it in 2020... but it's not gonna happen now.
apple mac pro price $999
Vega was a bit cheaper because of stiff competition from Nvidia, but Nvidia isn't all that competitive right now except in power utilization.
That's what's got me interested. There's reviews of the 590 where folks found it was throttling on a 500 watt power supply and they had to put a 600 watt in to fix it. As an adult I pay for all that power and it does add up. So for me the question is are the competitive with Nvidia on power consumption now?
Oh, and DMC at 4k/60fps? It's a beat'em'up/spectacle brawler ala God of War I/II. How badly have PC game port optimizations gotten that running that kind of game at max settings is a big deal? I'd be more impressed to see Fallout 4 or even The Division 2 pushing those kinds of frame rates.
Hi! I make Firefox Plug-ins. Check 'em out @ https://addons.mozilla.org/en-US/firefox/addon/youtube-mp3-podcaster/
And the 2080 is basically the same as the 1080 Ti.
And that's if you believe AMD's benchmarks. Sure, it'll win in some Vulkan games if you compare stock clocks to stock clocks, but it'll lose or barely keep up in everything else. And if you ever overclock, the Nvidia option will pull miles ahead while still using less power. This is with AMD having a node shrink advantage! Vega is trash.
The graphics card is NOT the device that controls when the new frame starts. Not if you want anything to work, that is.
If you use a format or timing the display doesn't support, you'll get nothing or you'll get a blinking (and possibly scrolling), unusable mess.
If you just vblank/vsync/rsync/whatever willy nilly on an LCD, it will laugh at you.
Variable refresh rates have been in the standards for ages, sure. But they were not implemented in any way that someone could drive a continuous, uninterrupted video stream with. Not until AMD came along and pushed for it.
The advantage over VSYNC is clear.
With FreeSync / VRR, you eliminate tearing and minimize delay as long as you stay within the VRR range of the display.
With VSYNC, if you're at 60 Hz you're adding an unnecessary delay whenever you could render faster than that, and you're dropping down to 30 Hz whenever you dip just under 60 FPS. Or 15 Hz if you drop under 30 FPS.
I see the 1080 ti going from almost $800 to almost $1680 (newegg) I can find some Nvidia RTX 2080 for as cheap as 699.99 to $1,699.00.
I fail to see what your issue is?
The spirit of resistance to government is so valuable on certain occasions that I wish it to be always kept alive
AMD still doesnt release drivers as frequently as nvidia
That should be a positive. Frequent driver updates are indicative of shitty quality and a broken system where hardware manufacturers bend over backwards to add in game-specific hacks to deal with shitty code put out by developers / engines. Yes, occasionally the games are running into flaws in the driver (or hardware) and the updated drivers fix / work around that. For those instances, see the prior point regarding shitty quality.
I'm all for dedicated graphic chips as long as they cost less than $100.
Time is what keeps everything from happening all at once.
is so you don't have to upgrade. There's videos on youtube of folks benchmarking 7 year old flagships and still hitting 60fps. So spend $700 now and pay $100/yr for a card or $300 every 2-3 years and hit somewhere around $800-$900. Plus the flagships tend to hold their value better, so you'll probably get $200 for it in 7 years when you sell it.
Also if you replace it with something just as power hungry that kinda defeats the point...
Thing is, if I keep a card 4 years (which I usually do) and save $15/yr on power (little less since I buy lower mid-range, $180-$230) that's $60 bucks gone. If I take that $60 and put it into a GPU it gets me into the upper mid range ($290-$300). That means anything AMD puts out has to outperform Nvidia by a big margin or it has to offer some other advantage (better image quality, better features, etc).
I think AMD was starting to kill it with cheap Freesync monitors, but as usual Nvidia noticed and responded, making their cards work with Freesync.
Hi! I make Firefox Plug-ins. Check 'em out @ https://addons.mozilla.org/en-US/firefox/addon/youtube-mp3-podcaster/
it's currently killing frame rates. Like in half. Cards that can do ray tracing at 60 fps are $1000+. The $700 Nvidia cards that can do ray tracing are hitting 35-50 depending on workload. If I drop a grand on a video card I want 60 fps...
Hi! I make Firefox Plug-ins. Check 'em out @ https://addons.mozilla.org/en-US/firefox/addon/youtube-mp3-podcaster/
Except that's not how it works. Screen updates are not instantaneous, and vsync originates with the monitor, not the video card - you can think of it as the monitor saying "okay, I'm drawing the next frame, start feeding me the data NOW", and the video card is expected to comply and start sending whatever is in the frame buffer, because the monitor is already busy displaying it.
What Gsync and Freesync essentially do is let the computer say back "Wait, not yet... okay now", so that if your rendering engine is only rendering at 57fps, while your monitor updates at 60fps, then every frame you get the monitor saying "Feed Me!", the computer says "wait...okay, now" and then starts sending sending that frame's image data. The result being that the monitor cleanly refreshes the full screen at 57 Hz, rather than creating a crawling horizontal "tear line", wherever it was at in the update cycle when you flipped the buffer - with the old frame being displayed above the tear line, and the new frame below.
--- Most topics have many sides worth arguing, allow me to take one opposite you.
"No marketing gimmicks, and drivers that are released when updates are relevant".
Where's the downside?
(Before you pretend that releasing drivers often lets them fix problems, tell us, how many driver releases did they kick the the firefox browser corruption bug down the road again?)
that made the programming easier. Right now AAA games cost a fortune and they're kind of simplistic. Compare any modern game to Deus Ex. The stupid complexity of modern graphics are a big part of that. Having to hand code shaders for every little look and effect gets really pricey really fast....
Hi! I make Firefox Plug-ins. Check 'em out @ https://addons.mozilla.org/en-US/firefox/addon/youtube-mp3-podcaster/
That's straight up marketing BS. Reality is, unfixed bugs stay across multiple driver releases with remarkable consistency. In some cases, they persist for many months, even when they're so well documented that nvidia has to literally copy/paste it in documentation errata in each release for months, like the firefox cursor corruption issue.
....I guess I'm still waiting for the "glut" of Nvidia top end cards to hit the market, somehow I can't comprehend how Nvidia sitting on thousands and thousands of cards in inventory and that hasn't impacted their prices.
-Styopa
That's straight up marketing BS. Reality is, unfixed bugs stay across multiple driver releases with remarkable consistency.
I think you're both right, or at least, in my experience both things are true. nvidia does have more releases that fix my problems, and also in the past I have needed DnA (patched) drivers to get my ATI/AMD GPUs to work correctly.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
There's a bit of a terminology snafu here:
v(ertical)sync's original meaning is the once-per-frame signal between monitor and video card that lets them stay coordinated and display a stable image (in combination with h(orizontal)sync, which happens once per line)
Then there's the more modern usage where you "turn on vsync" in a game. But what that's actually turning on is "vsynced rendering", or perhaps better expressed "sync the framebuffer update to the monitor vsync signal"
And yes,"turning on vsync" does stop the tear line, because waiting for vsync to flip the framebuffer means the buffer doesn't change while the monitor is refreshing. But it does so at the expense of drastically reducing the frame rate, at least intermittently. Framerate can only be adjusted in integer multiples of frames, so if you're only able to render at 57Hz, then syncing the rendering engine to vsync will drop the frame rate to 60/2 = 30Hz, either permanently or as "stuttering", depending on exactly how the synchronization is implemented.
(Hmm, I'm a bit rusty, but I think you need triple-buffering to only see stuttering - pretty sure that double-buffering requires you to wait until vsync to free up the "old" framebuffer before you can start rendering the next frame, which means *every* new frame will take slightly too long to render, and thus *every* old frame will be displayed for an extra refresh while your computer sits idle, dropping you to 1/2 your montors refresh rate. Triple buffering lets you render one frame while a second waits for the monitor to finish displaying the third. Can anyone confidently confirm or correct me?)
That's actually one of the big draws for high frame rate monitors - smoother vsynced rendering. If your monitor refreshes at 120Hz instead of 60Hz, then a 57Hz rendering engine will cause the framerate to stutter between 120/2=60Hz and 120/3=40Hz. Or alternately, your game might be able to lock to a slower refresh rate - e.g. a nice stable 40 Hz in this case (which is kind of slow, and probably probably why rates like 144Hz are more popular - you can run that at 72 or 48Hz, either of which is pretty solidly above the motion perceptual threshold)
--- Most topics have many sides worth arguing, allow me to take one opposite you.
"Navi" should be the new one, presumably on 7nm and GDDR6. Why not now?
Because the Navi team is working on PS5. When that's done, they'll work on the PC Navi hardware.
Also related: Everyone talking about Ray tracing are playing right into NVidia's hands; they would want any talking points regarding AMD to include "but where's the ray tracing?" even though it's a vendor specific thing that not everyone will use, even if they have the hardware, because using it drops your frame rate.
Meanwhile, the game developers are going to primarily target AMD based GPUs for most of their work, because Playstation and XBox contain AMD chips.
Adding RTX support to their games as well will incur more development and maintenance cost, to support a subset of the PC gaming market that's going RTX off anyway due to frame rates. Why bother doing that extra work for nothing?
The raytracing Nvidia is touting is DXR, or DirectX Raytracing. It's a standard AMD can support, rather than a proprietary Nvidia-only thing. Honestly I don't care about the raytracing that much (in this gen of video cards) because it's too slow and poorly-supported. However, if I were dropping $700 on a new top-end video card, I'd question paying the same money for fewer features. If the Radeon 7 were a couple hundred cheaper, I might decide to skip DXR for a few years.
Another thing to keep in mind is that the second AMD cards DO support DXR, Nvidia will stop giving money to devs to implement DXR support... meaning that DXR support will likely be worse afterward.
Corruption is convincing someone that the selfless ideal is the same as their selfish ideal.
I completely agree that in the past, some games required patches from AMD to run properly. Same is true in reverse, where they needed patches from nvidia to run properly. This is mainly linked to each manufacturers "best played on our cards" campaigns, which tend to block the other GPU manufacturer access from the title to make relevant driver adjustments ahead of time.
As AMD's program is notably smaller than nvidia's, it tends to run into this problem more than nvidia.
No it's not, it really adds some nice visuals.. It's more in line with bumpmapping etc in the old days.. Also due to reflections you can actually use it to your advantage.
Argument: AMD wouldn't want to support DXR right now because they have no properly DXR compliant HW, and their next targets are PS5 and Apple (Navi) where they will focusing on Vulcan and Metal
Because:
16G vs 8G
3 games
not nVidia
The 2080 costs around $800. If you compare based on price/performance, the Vega VII is more attractive if it can match the performance of the 2080. The 16 GB might also be more future-proof with constantly increasing memory requirements in gaming.
C - the footgun of programming languages
Yeah, there are a few games out there that support it. Turning it on often cuts the game performance by half, though. It's basically the same impact as jumping from 1080p to 4K resolution at the moment.
"I completely agree that in the past, some games required patches from AMD to run properly."
Lets be clear here, DnA drivers were third party patched. The fact is that amd was so bad at writing working drivers that they literally could not do it, and it was necessary to get drivers which had been modified by someone else to get the GPU working correctly. Meanwhile Nvidia was putting out fairly regular releases with their own fixes.
Everyone says that amd is much better now, but most of their customers claimed that their drivers worked when they clearly didn't, so now I remain hesitant. Nobody I trust has yet told me that the problems are truly over.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
Okay, so I'm specifically talking about rendering *without* tearing - which means that you can never alter the front framebuffer while it's active, only during the brief pause between when the monitor finishes displaying one frame and begins displaying the next. That includes no swapping buffers. Do so, and the screen will tear, displaying the top part of the old frame buffer, and the bottom part of the new one.
No buffering means that the only way to avoid tearing is to render the entire frame during that brief pause between refreshes, while the frame buffer is not in use. That was the standard procedure all through the pre-VGA days, and very often with VGA and early SVGA (VGA and early SVGA cards didn't have enough memory for a second buffer at higher resolutions). Obviously though that means that you can only use the small fraction of each 1/60th of a second for rendering - not ideal if rendering is a big part of the per-frame workload. There was actually a lot of effort spent on techniques to render the screen from the top down to buy more time - after all you didn't actually have to finish rendering a line until just before it was fed to the monitor. That was quite successful for most 2D games, but mostly didn't work for 3D games which required the screen to be rendered in a somewhat arbitrary order for efficiency reasons.
Double buffering was a way to address that, and means that you can render a second frame into the back buffer while the front buffer is being fed to the monitor, and can thus use the entire 1/60th of a second for rendering without any render-sequencing tricks. You can't instantly switch buffers when you're done with the next frame though, or you'll cause tearing. You have to wait until the next vsync so that the front buffer is no longer in use.
That works great so long as you can render each frame before the next vsync happens. It breaks down if you can't though. If you start rendering a frame the instant the vsynced buffer flip frees up the old front buffer, and it takes you just slightly longer than 1/60th of a second to render, then the next monitor refresh will have already begun before you're finished. That means that you can't flip buffers immediately or you'll cause tearing, so you have to wait until the NEXT vsync, halving your frame rate for that frame. And then you're stalled out - you can't start rendering a third frame during that time, because you only have two buffers - one currently being displayed, and one waiting to be. You just have to wait until the next vsync to free up a framebuffer - which means you'll start the next frame in the same situation, and continue rendering each frame just slightly too late, so that the previous frame always has to be displayed for two monitor refreshes, and your animation frame rate drops to half the monitor refresh rate. In that case it's probably better to just deal with the tearing (unless your monitor refresh rate is high enough that the animation rate can step down more gradually)
Triple buffering solves the problem by giving you a third buffer to work with, so that when you finish rendering a frame just slightly too late, you can at least immediately start rendering a third frame so that you've got a head start on the next cycle. You still drop one frame of animation, but you'll be almost done rendering frame 3 by the time frame 2 gets displayed.
Timing diagram starting at having just swapped frame zero to the front buffer and started rendering frame 1. Rendering a frame takes 6 ticks, monitor refreshes every 5 ticks
Double buffering halves frame rate since only two buffers can be in use at a time ...
Render-- 11111 1xxxx 22222 2xxxx 33333 3xxxx...
Display- 00000 00000 11111 11111 22222 22222
Triple Buffering has stutter
Render-- 11111 12222 22333 33344 44445 55555 66666 67777 77888...
Display- 00000 00000 11111 22222 33333 44444 55555 55555 66666...
I'm fairly certain that's correct
--- Most topics have many sides worth arguing, allow me to take one opposite you.
Well, time wasted on my reply then I guess.
And yes, triple buffering is no longer super memory intensive, though it used to be. 24MB for three 32-bit 1080p frames, 96MB for 3x4k... I've gone through a LOT of video cards that had far less RAM than that.
Triple buffering does though add a variable amount of lag though - which can futz with your reflexes. Just look at the time between starting to render a frame (last input actions processed), and the frame being displayed (visual feedback received)
Double buffering: 10,10,10...
Triple buffering: 10,9,8,7,6,5,10,9,8...
It also carries that extra full frame of lag when you *can* render frames fast enough, since you're rendering two frames ahead of the monitor instead of only one (at least it used to - though it seems to me that it would be easy enough to only use the third buffer when actually needed)
--- Most topics have many sides worth arguing, allow me to take one opposite you.
But anyway, not about gaming performance, but about bandwidth.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
Comment removed based on user account deletion
Comment removed based on user account deletion
Comment removed based on user account deletion