The AMD Radeon R9 290X Review
by Ryan Smith on October 24, 2013 12:01 AM EST- Posted in
- GPUs
- AMD
- Radeon
- Hawaii
- Radeon 200
A Bit More On Graphics Core Next 1.1
With the launch of Hawaii, AMD is finally opening up a bit more on what Graphics Core Next 1.1 entails. No, they still aren’t giving us an official name – most references to GCN 1.1 are noting that 290X (Hawaii) and 260X (Bonaire) are part of the same IP pool – but now that AMD is in a position where they have their new flagship out they’re at least willing to discuss the official feature set.
So what does it mean to be Graphics Core Next 1.1? As it turns out, the leaked “AMD Sea Islands Instruction Set Architecture” from February appears to be spot on. Naming issues with Sea Islands aside, everything AMD has discussed as being new architecture features in Hawaii (and therefore also in Bonaire) previously showed up in that document.
As such the bulk of the changes that come with GCN 1.1 are compute oriented, and clearly are intended to play into AMD’s plans for HSA by adding features that are especially useful for the style of heterogeneous computing AMD is shooting for.
The biggest change here is support for flat (generic) addressing support, which will be critical to enabling effective use of pointers within a heterogeneous compute context. Coupled with that is a subtle change to how the ACEs (compute queues) work, allowing GPUs to have more ACEs and more queues in each ACE, versus the hard limit of 2 we’ve seen in Southern Islands. The number of ACEs is not fixed – Hawaii has 8 while Bonaire only has 2 – but it means it can be scaled up for higher-end GPUs, console APUs, etc. Finally GCN 1.1 also introduces some new instructions, including a Masked Quad Sum of Absolute Differences (MQSAD) and some FP64 floor/ceiling/truncation vector functions.
Along with these architectural changes, there are a couple of other hardware features that at this time we feel are best lumped under the GCN 1.1 banner when talking about PC GPUs, as GCN 1.1 parts were the first parts to introduce this features and every GCN 1.1 part (at least thus) far has that feature. AMD’s TrueAudio would be a prime example of this, as both Hawaii and Bonaire have integrated TrueAudio hardware, with AMD setting clear expectations that we should also see TrueAudio on future GPUs and future APUs.
AMD’s Crossfire XDMA engine is another feature that is best lumped under the GCN 1.1 banner. We’ll get to the full details of its operation in a bit, but the important part is that it’s a hardware level change (specifically an addition to their display controller functionality) that’s once again present in Hawaii and Bonaire, although only Hawaii is making full use of it at this time.
Finally we’d also roll AMD’s power management changes into the general GCN 1.1 family, again for the basic reasons listed above. AMD’s new Serial VID interface (SIV2), necessary for the large number of power states Hawaii and Bonaire support and the fast switching between them, is something that only shows up starting with GCN 1.1. AMD has implemented power management a bit differently in each product from an end user perspective – Bonaire parts have the states but lack the fine grained throttling controls that Hawaii introduces – but the underlying hardware is identical.
With that in mind, that’s a short but essential summary of what’s new with GCN 1.1. As we noted way back when Bonaire launched as the 7790, the underlying architecture isn’t going through any massive changes, and as such the differences are of primarily of interest to programmers more than end users. But they are distinct differences that will play an important role as AMD gears up to launch HSA next year. Consequently what limited fracturing there is between GCN 1.0 and GCN 1.1 is primarily due to the ancillary features, which unlike the core architectural changes are going to be of importance to end users. The addition of XDMA, TrueAudio, and improved power management (SIV2) are all small features on their own, but they are features that make GCN 1.1 a more capable, more reliable, and more feature-filled design than GCN 1.0.
Facebook
Twitter
RSS
396 Comments
View All Comments
ninjaquick - Thursday, October 24, 2013 - link
so 4-5% faster than Titan?Drumsticks - Thursday, October 24, 2013 - link
If the 780Ti is $599, then that means the 780 should see at least a $150 (nearly 25%!) price drop, which is good with me.DMCalloway - Thursday, October 24, 2013 - link
So, what you are telling me is Nvidia is going to stop laughing- all- the- way- to-the-bank and price the 780ti for less than current 780 prices? Current 780 owners are going to get HOT and flood the market with used 780's.dragonsqrrl - Thursday, October 24, 2013 - link
Why is it that this is only ever the case when Nvidia performs a massive price drop? Nvidia price drop = early adopters getting screwed (even though 780 has been out for ~6 months now). AMD price drop = great value for enthusiasts, go AMD! ... lolz.Minion4Hire - Thursday, October 24, 2013 - link
Titan is a COMPUTE card. A poor man's (relatively speaking) proper compute solution. The fact that it is also a great gaming card is almost incidental. No one needs a 6GB frame buffer for gaming right now. The Titan comparisons are nearly meaningless.The "nearly" part is the unknown 780 TI. Nvidia could enable the remaining CUs on 780 to at least give the TI comparable performance to Titan. But who cares that Titan is $1000? It isn't really relevant.
ddriver - Thursday, October 24, 2013 - link
Even much cheaper radeons compeltely destroy the titan as well as every other nvidia gpu in compute, do not be fooled by a single, poorly implemented test, the nvidia architecture plainly sucks in double precision performance.ShieTar - Thursday, October 24, 2013 - link
Since "much cheaper" Radeons tend to deliver 1/16th DP performance, you seem to not really know what you are talking about. Go read up on a relevant benchmark suite on professional and compute cards, e.g. http://www.tomshardware.com/reviews/best-workstati... The only tasks where AMD cards shine are those implemented in OpenCL.ddriver - Thursday, October 24, 2013 - link
"Much cheaper" relative to the price of the titan, not entry level radeons... You clutched onto a straw and drowned...OpenCL is THE open and portable industry standard for parallel computing, did you expect radeons to shine at .. CUDA workloads LOL, I'd say OpenCL performance is all I really need, it has been a while since I played or cared about games.
Pontius - Tuesday, October 29, 2013 - link
I'm in the same boat as you ddriver, all I care about is OpenCL in these articles. I go straight to that section usually =)TheJian - Friday, October 25, 2013 - link
You're neglecting the fact that everything you can do professionally in openCL you can already do faster in cuda. Cuda is taught in 600+ universities for a reason. It is in over 200 pro apps and has been funded for 7+yrs unlike opencl which is funded by a broke company hoping people will catch on one day :) Anandtech refuses to show cuda (gee they do have an AMD portal after all...LOL) but it exists and is ultra fast. You really can't name a pro app that doesn't have direct support or support via plugin for Cuda. And if you're buying NV and running opencl instead of cuda (like anand shows calling it compute crap) you're an idiot. Why don't they run Premiere instead of Sony crap for video editing? Because Cuda works great for years in it. Same with Photoshop etc...You didn't look at folding@home DP benchmark here in this review either I guess. 2.5x faster than 290x. As you can see it depends on what you do and the app you use. I consider F@H stupid use of electricity but that's just me...LOL. Find anything where OpenCL (or any AMD stuff, directx, opengl) beats CUDA. Compute doesn't just mean OpenCL, it means CUDA too! Dumb sites just push openCL because its OPEN...LOL. People making money use CUDA and generally buy quadro or tesla (they own 90% of the market for a reason, or people would just buy radeons right?).
http://www.anandtech.com/show/7457/the-radeon-r9-2...
DP in F@H here. Titan sort of wins right? 2.5x or so over 290x :) It's comic both here and toms uses a bunch of junk synthetic crap (bitmining, Asics do that now, basemark junk, F@H, etc) to show how good AMD is, but forget you can do real work with Cuda (heck even bitmining can be done with cuda)
When you say compute, I think CUDA, not opencl on NV. As soon as you toss in Cuda the compute story changes completely. Unfortunately even Toms refuses to pit OpenCL vs. Cuda just like here at anandtech (but that's because both love OpenCL and hate proprietary stuff). But at least they show you in ShieTar's link (which craps out, remove the . at the end of the link) that Titan kills even the top quadro cards (it's a Tesla remember for $1500 off). It's 2x+ faster than quadro's in almost everything they tested. So yeah, Titan is very worth it for people who do PRO stuff AND game.
http://www.tomshardware.com/reviews/best-workstati...
For the lazy, fixed ShieTar's link.
All these sites need to do is fire up 3dsmax, cinema4d, Blender, adobe (pick your app, After Effect, Premiere, Photoshop) and pit Cuda vs. OpenCL. Just pick an opencl plugin for AMD (luxrender) and Octane/furryball etc for NV then run the tests. Does AMD pay all these sites to NOT do this? I comment and ask on every workstation/vid card article etc at toms, they never respond...LOL. They run pure cuda, then pure opencl, but act like they never meet. They run crap like basemark for photo/video editing opencl junk (you can't make money on that), instead of running adobe and choosing opencl(or directx/opengl) for AMD and Cuda for NV. Anandtech runs Sony Vegas which a quick google shows has tons of problems with NV. Heck pit Sony/AMD vs. Adobe/NV. You can run the same tests in both on video, though it would be better to just use adobe for both but they won't do that until AMD gets done optimizing for the next rev...ROFL. Can't show AMD in a bad light here...LOL. OpenCL sucks compared to Cuda (proprietary or not...just the truth).