A few days ago, I conducted the first part of my Interview with Robert Hallock, Technical Communications, Desktop Gaming & Graphics over at AMD. During this interview, we discussed topics such as GPU computing, the GCN (Graphic Core Next Architecture) and the future of PC gaming according to AMD. In this second part of the interview, we follow up with a few of the questions which didn’t make it in time for print, discussing Mantle’s performance, more info on ROP and Texture Unit count, and AMD’s opinion on Valve’s Steam Machine configurations.
Could you offer any comments regarding FreeSync technology and how it functions?
Robert Hallock: I’ll have to get back with you soon. We only just made the announcement at CES, and I actually have to learn a little more about the technology myself.”
Because the interview took place pretty much a day after the CES event, much of the technology behind Free-Sync is still being worked on. AMD are currently in working on ways for it to work with Desktop monitors, and are making progress. See news here. They’re not quite ready to speak about this yet.
Regarding Mantle’s performance, it was mentioned at CES that the performance increase with Battlefield 4 was up to 45 percent. Will this sort of performance boost be the normal for titles under Mantle?
Robert Hallock: Every game is different, of course, but we’ve said all along that we didn’t undertake the Mantle project to chase two or three percent performance improvements. With each game, we’ll be publishing a full set of data and we’re confident people will be pleased with the uplift.”
Even if the performance of the Mantle API isn’t up to 45 percent per game, and only works out to be half as much, that’s still going to be a significant difference. This can turn a title barely running in 30FPS into the comfortable to play territory, or titles which are just shy of the target of 60 shooting way past that number. Obviously it’s still too early to make judgement calls quite yet, but for titles that have AMD’s Mantle API available, Nvidia could be in for a fight.
I’m hoping there’ll be news of Nvidia embracing the technology, or some alternative to start to resolve this.
How long do you believe it will take before mobile / slim form factor gaming performance can easily outperform the last generation consoles in terms of what we see on screen?
Robert Hallock: Hmm, I really couldn’t say. I don’t follow mobile/sff gaming all that closely, so I wouldn’t be able to provide an intelligent answer.
Following up to the GPU performance question. Many have noticed that AMD are extremely generous in ROP count for your GPU’s. For example, the R9 range of cards has more ROPS than your competition. How many ROPS and Texture Units are required for a resolution of 1080P and 60FPS, and 1440P?
Robert Hallock: I think it’s easiest to look at hardware we already offer, in this instance, as they are demonstrative of what we think are ideal hardware configurations for certain performance targets within this generation of gaming. For example, the AMD Radeon™ R9 270X is the fastest card we make specifically designed for 1080p gaming: 32 ROPs and 80 texture units. Stepping up to the R9 280X, this being our product specifically designed for 1440p gaming, you’ll find a 32/128 configuration. But I do want to emphasize that these figures are in the context of today’s GCN Architecture, and may not be true in the future, so don’t hang your hat on them five years down the road or what have you.”
The remaining specs of the R9 270X for those curious is a core clock of 1000MHZ, 1280 Stream Processors, 2GB GDDR5 RAM running at 5.6GHZ effective on a 256 bit bus. This kind of card runs titles at around the 60FPS mark generally at full quality at the 1080P. So it’s interesting from AMD that they believe that the GCN architecture requires roughly this spec (in terms of ROP / TMU) to perform this well at 1080P.
Do you feel that the Steam Machines (being pre-configured and ‘ready to go’) will help make PC gaming mass market. One of the common complaints I get from friends is that “pc gaming is more complex”. I do know that AMD are also trying to tackle this however.
Robert Hallock: Sure, we’re trying to tackle this, for example with the AMD Gaming Evolved App Powered by Raptr. It doesn’t necessarily simplify the process of building a gaming PC, but we’re trying to take the legwork out of optimizing a game and getting it running smoothly by crowdsourcing a load of data for users, then making that available to them in the form of one-click optimization.
As for Steam Machines, that is their biggest appeal in my own personal opinion. Valve and Steam have such prestigious names in our industry, deservedly so, and I think it’s wonderful Valve is using that prestige to make PC gaming more accessible.”
From my own perspective as a PC gamer, I believe the belief that PC gaming is more hassle is one of the core problems that must be tackled. With both AMD (using thei Gaming Evolved app) and Nvidia with their Geforce experience App helping to pre-configure games, it’s becoming easier. The problems with beginners is that they often don’t understand how to best configure the graphics options with games, and that puts them off.
The other issue is hardware – inexperienced gamers just don’t know what to buy. Fortunately, there are less ‘lemon’ products out there than there used to be. Back several years ago, it was fairly common practice to sell GPU’s with high amounts of slower RAM. These cards would seem like bargains, often with double the RAM, but the customer didn’t know the memory was often extremely slow, crippling the cards performance due to bandwidth constraints. Fortunately this is less of a problem now, but pre-configured PC are still somewhat hit and miss when it comes to the value to the customer. Hopefully with standards set by Valve’s Steam Machines, we’ll see some of this getting resolved.
Concerning PCIE bandwidth and latency. I’ve seen numerous comparisons that show that PCIE bandwidth isn’t currently fully saturated on a PCIE 3 bus, and on high end cards we’re currently seeing very little performance dip between PCIE3 and 2. Given how XDMA works with Crossfire on PCIE 3.0, how much of a bottleneck does the PCIE bus pose for GPU compute and for high graphics performance in terms of latency and bandwidth?
Robert Hallock: In independent analysis, such as tests conducted by Anandtech, a very small but notable performance falloff has been demonstrated on PCIe 2.0 x16 in full-bore compute scenarios with multiple GPUs. The same could not be replicated for gaming scenarios. Even so, I would argue that PCIe is not a bottleneck at all.”
Just to clarify, XDMA does away with the traditional bridge which connects two GPU’s (from AMD) and instead feeds all of this data through the PCI-E 3.0 bus. AMD used this technique to help remove frame pacing issues which were a major thorn in the side of the Radeon 7xxx series. Although AMD have managed to fix a lot of these frame rate pacing problems through new drivers, the best fix ultimately was the XDMA technology. For more info on this please click here.
And that’s it for this particular interview. Thanks to Robert and the rest of AMD for the insights, and I look forward to interviewing them again soon regarding CPU and APU specifics.