From a game developer’s point of view, creating games for the PC has never been especially efficient. With so many combinations of hardware possible in a PC, it’s not practical to create specialized programming for every possible configuration. What they do instead is write simplified code that gets translated on-the-fly into something the computer can work with.
Just as when two people communicate through a translator, this works, but it isn’t efficient. And it’s the CPU that has to do all this extra work, translating and queuing data for the graphics card to process. PCs are meant to be the ultimate gaming platform — they have the power — but all this translation slows things down, and game developers approached AMD asking for something better.
What Mantle does
Mantle is the harmony of three essential ingredients:
A driver within the AMD Catalyst™ software suite that lets applications speak directly to the Graphics Core Next architecture;
A GPU or APU enabled with the Graphics Core Next architecture;
An application or game written to take advantage of Mantle.
Mantle reduces the CPU’s workload by giving developers a way to talk to the GPU directly with much less translation. With less work for the CPU to do, programmers can squeeze much more performance from a system, delivering the greatest benefits in gaming systems where the CPU can be the bottleneck.
What it means for gamers
Now that Mantle has freed up some extra CPU capacity, we expect Mantle will lead to better games, and more of them, since Mantle makes game development easier.
That’s not all Mantle will do for gamers. By shifting work to the GPU, a mid-range or older CPU isn’t the same handicap it was before. With Mantle, the GPU becomes the critical part of the system, and GPU upgrades will have a bigger impact than before.
AMD: We're going after Nvidia graphics with Kaveri APU
The war is nothing new, but Kaveri is
To give a taste of its new Kaveri APU's kick, AMD ran a side-by-side Battlefield 4 demo during its opening APU13 keynote last week. The game ran on two machines; frames per second ticked away in the upper left hand corners as the opening, "Total Eclipse of the Heart"-sound tracked sequence rolled.
Kaveri came away as the clear winner, holding nearly double the frame rates and running with hardly a hiccup. The second machine, equipped with an Intel Core i7 4470K CPU and a Nvidia GeForce GT630 GPU, stuttered and lurched from scene to scene.
It was an effective, visceral demonstration, but the question quickly circulated why AMD would pit the top-end A10-7850K Kaveri APU against this CPU/GPU combo. Adam Kozak, senior product marketing manager at AMD, laid the company's logic out for TechRadar.
"We want to position an A10 versus Intel plus a graphics card," he told us during a post-keynote rendezvous. "Obviously you can go up so high before the graphics card gets faster, and that's why we picked the 630. There is a new 640 we're looking at, and we'll take a look at that as we get closer to launch."
It's not just CPU then that AMD is targeting with its first APU of 2014.
"We want to go after the idea where it makes sense, at least from our perspective, that you don't need to buy a certain graphics card," Kozak said. "In fact, Nvidia probably sells 70-80% of their entire stack at 630 and below. People kind of know that Intel is very weak [with] GPU, so now we're going after something people think is strong."
AMD's Adam Kozak talks the Kaveri talk
Kaveri in action, but what about Mantle?
We had caught up with Kozak to see another BF4 demo played on a different desktop - one presumably less meaty than the machine used for the keynote head-to-head. The settings were on medium except for a custom graphics quality setting. AMD Inclusion, which controls how shadows are placed on overlapping objects, was turned off.
"The difference between low and medium is huge, and then from medium to high, you see a little bit more details in the soft and shadows," Kozak said of how visible Kaveri's footprint becomes on different settings. "From high to ultra, it's more of a post-processing so you're light rays and everything are kind of blended a little more.
"For me, the biggest jump is from low to medium, and then from there it just gradually looks nicer and nicer, depending on what you've got."
The frame rates hit 39 or so as they did during the keynote, and the play never lapsed. Granted, this wasn't a particularly action-heavy demo we were being shown - Kozak was really just wondering through a ruined building.
The first Kaveri demo was more graphics intensive, but neither it nor the one Kozak played ran a Mantle-optimized version of BF4. AMD's Mantle API, developed with the help of EA's DICE, is designed to push frame rates higher and improve graphics fidelity. Pair it with Kaveri, and the hope is for near-perfect renderings.
That's the idea, of course, and Kozak for one is keeping his forecasts on a more even keel.
"Personally my expectations are low," he said of a Mantle-plus-Kaveri combo. "But there is an Oxide demo here and they are seeing substantial speed-ups, beyond what anyone internally has guessed at. I'm optimistic it's going to be more than the 5% I'm hoping for and more towards the double digits."
Busting through the frame rates
In fact, we're told Mantle is still fairly new for AMD internally, and it's partners like DICE who are seeing frame gaps vaporize.
"What I have heard from DICE is that what [Mantle] does with the discrete card is it equalizes the CPUs," Kozak explained. "It was only a couple of frames faster before because the CPU doesn't really play into things like that, but [Mantle] eliminates any gap. And essentially it does that by allowing the graphics card to do more, so it becomes the bottle neck."
When a Mantle optimization-bearing update arrives for Battlefield 4 in December, we'll have an accurate idea of just how much the API improves graphics performance in the real world.
Masters of productivity
Our meeting with Kozak held more than just a BF4 run through.
He also showed us a JPEG decode accelerator that overrides the usual routines found in Windows and speeds them up with Kaveri. In one thumbnail decode run, performance increased by 110.1%.
When the first Kaveri APU desktops become available January 14, they'll have the decoder built in. More than simply decoding family photos faster, the driver shines a light on an area AMD wants to target with Kaveri.
"Productivity is sort of a new one for us," Kozak told us.
The company plans to have additional productivity compute acceleration examples at launch, and Kozak said that "the idea there is to gain interest, [to] get Microsoft and others aware that you can make things a lot faster on tasks that people still care about in a professional level with these crazy spreadsheets that the normal consumer level may not be much of a big deal."
No more trips to the kitchen while JPEGs decode
Kaveri price and scalability
Not all is known about Kaveri - CES 2014 is the APU's "big coming out party," as we've been told by AMD.
With the first Kaveri desktops due early next year, the all-important price question will be answered in short order. Until then Kozak and AMD are keeping mum on cost, but we suspect Kaveri will be priced in the same range as Richland desktop was when that APU was released in the channel.
Kozak noted, as AMD has, that the company has reversed its normal APU release order and is taking Kaveri first to desktop.
"It's sort of a chicken and egg thing," he said of the decision. "We're really interested in getting Kaveri out there as fast as possible. If it is just the desktop, and not the bread and butter of mobile, it's because we need guys to start programming for it. We give them the fastest implementation and they can start optimizing their code, and obviously from there can start optimizing for lower TDPs."
While scalability with Kaveri is a big selling point, Kozak said not to expect it in something as small as a smartphone anytime soon.
"Right now we're going as low as 15W, which is not a phone, all the way up to the typical desktop," he said. Though Kozak didn't mention it, AMD has stated Kaveri will head to embedded systems and servers as well.
Snipping through Battlefield 4 and JPEGs is all well and good, but when it comes to real-world implementation, CES and the days following are going to be Kaveri's true gauntlet run.
Early showings have been impressive, and Kozak said more work is being done to fine-tune Kaveri.
"I expect this one to get even better for us," he said, referring to the Battlefield 4 desktop demo. "We still have our engineers working with DICE on Kaveri optimizations. That's going on as we sit here, and we still have DICE working with Mantle. There's a two-step prong that's going to make this even better."
Victory snatched from the jaws of defeat. Or is it the other way round? Either way, only AMD could pull it off with such perverted panache.
I speak of the new AMD Radeon R9 290. Yes, specifically the 290, not the 290X. By most metrics, it's by far and away and without a shadow of a doubt the best graphics card you can currently buy. And yet somehow, AMD managed to launch it in a state that some leading review sites felt they couldn't recommend it.
Let me re-emphasise that. AMD conspired to create the best graphics card on the market and yet make it sufficiently flawed that some experts advised PC gamers not to buy it.
Let's remind ourselves first of what makes the 290 great. It's not actually the fastest graphics chip in the world. That accolade falls to Nvidia's might GeForce GTX 780 Ti. It's not even AMD's fastest. The Radeon R9 290X takes that prize.
Huge performance, plausible price
But what it does do is deliver frame rates that I suspect are largely in distinguishable from those faster chipset in subjective gaming terms. And it does so at a fraction of the price.
As I type these words, an Nvidia GeForce GTX 780 Ti will sock you for about £500. AMD Radeon R9 290Xs start at about £420. But the 290 is yours for just over £300. Nice.
"To get right to the point then, this is one of a handful of cards we've ever had to recommend against."
So, that's one of a handful of cards Anandtech has ever unambiguously recommended against in around a decade of graphics card reviews.
What on earth is wrong with the 290? As it happens, I think Anandtech blundered pretty badly to put the 290 into such undistinguished company. But the 290 is undoubtedly flawed.
There are two closely related problems. The GPU at the heart of the 290 runs very hot and it sports a very noisy fan. The latter problem, if it is a problem, depends on your point of view. Some won't be too worried about a bit of din when the GPU is under heavy load. To be clear, the card is only noisy when rendered detailed 3D graphics.
Piping hot pixel pumper
The temperature issue is potentially more serious and raises concerns about the long-term reliability of 290 boards. AMD says the running temps are fine, but history shows computer chips soaking up 90 degrees-plus with regularity tend to go pop eventually.
The full story is a little more complicated yet and involves some last minute tweaking of the settings controlling the 290's fan. But the details really don't matter too much. The bottom line is that AMD had managed to launch a card that simultaneously all comers while looking seriously flawed, at least to some.
The irony here is that all AMD needed to have done was launch with a quieter fan. The temps could have remained the same and while they would have been remarked upon, I doubt they would have become the main story.
But the hurricane-force fan did a stand up of grabbing all the headlines and spoiling the excellent work of the engineers who designed the GPU.
Fortunately, a solution to all this is on its way. Pretty soon, 290 boards with custom cooling will become available and the din will die down, both figuratively and literally. And then the 290 will take its rightful place as the best graphics card in the world.
In the meantime, I can only marvel at how AMD could make such a mess of such an inherently great product.
