AMD REDEFINES PERFORMANCE
By William Van Winkle
Everybody knows there’s an “enthusiast” group at the top of the consumer food chain. This is the elite minority, the gamers and other niche hobbyists who make up maybe 5% of the market but pay whatever it takes for the latest and greatest. Below this is the “mainstream,” the majority of the market, the buyers who want an affordable system that does most things fairly well. AMD says there’s a group in the middle. We’re not sure if you’d call these enthusiasts on a budget or mainstreamers who aspire to be enthusiasts. Whichever way you dice it up, AMD calls this the “performance” group.
Not so long ago, I would have defined “performance” as the fastest subset of the mainstream. This user would simply buy the mainstream platform du jour—something like a P45 if we were discussing Intel—and pair it with the fastest CPU he could afford, end of story. But these performance users are a tricky lot to pin down, in part because they’re comfortable with the concept of upgrading. They may want to start modestly, then upgrade the system as needs and budget allow. In comparison, typical mainstream users will either make do with what they have or buy a new PC.
Unlike enthusiasts, who tend to buy their products via etail, performance users are still a strong channel play. Because they often lack the deep expertise of many enthusiasts, savvy resellers can step in with the guidance and products this group needs to meet their exact needs. You don’t see the word “scalability” often applied to consumers, but that’s precisely what performance buyers are after. They want platforms that can scale, and this is something that conventional mainstream systems don’t address well.
AMD is a company filled with surprises. No one who knows processors would dispute that the company’s success, is built on the ingenious work of its engineers. In fact, there are so many perks and cool facets in AMD’s designs that the marketing side of the company sometimes can’t keep up. For example, a couple of weeks ago, I was researching the Phenom X4 Black Edition processors. You might already know that the key difference between a Black Edition chip and its non-Black counterpart is that the Black Edition has an unlocked multiplier for more flexible, robust overclocking.
The Intel equivalent of this capability rests in the Extreme Edition processors. What’s not equivalent between these two CPU groups is the price. Current generation Black Edition chips start at about $200. The most expensive Black Edition part, the 2.6 GHz model 9950, can be had online currently for $235. Intel starts its modern Extreme Edition parts at $1,000. In other words, if your buyer likes to overclock as a hobby—just the sort of person you’d expect to find in the performance crowd—he can get a chip with unlimited overclocking capabilities for $800 less on the AMD side of the fence. If AMD ever tried to market this fact, I flat-out missed it, but it’s a great story to tell your performance customers. Yes, the Intel Extreme chips will generally benchmark out faster in head-to-head comparisons against the Black Edition parts, but $800 faster? Top-shelf enthusiasts are willing to drop a grand for an extra few percent in performance, but performance sensibilities are more rooted in the mainstream, and this group will find the Black Edition a much smarter fit.
Naturally, a good CPU needs to be paired with a suitable chipset, and AMD finally has a perfect fit for the X4 Black Edition—the 790GX. Now, let’s be honest. We all know that a lot of times companies will promote the “perfect” this or that and try to BS you into thinking a square peg goes into a round hole. I don’t think this is one of those cases. Hang with me, and I think you’ll see why.
If scalability is the theme of this chipset, then there are three main ways in which the 790GX proves its case. The first is graphics. With the 780G (see also this month’s cover story for this chip’s mobile version), AMD integrated an HD 3200 graphics core into the northbridge. With the 790GX, AMD updates this to the HD 3300. OK, maybe that’s not like night and day, but remember that Intel is the 800-pound gorilla of IGPs, and while the G45's X4500 IGP has 10 unified shader processors, AMD’s HD 3300 has 40. AMD’s own benchmarking against the 780G and Intel G45 turned up the following scores in 3DMark06: The G45 scored 1075, the 780G hit 1540, and the 790GX smoked both of them at 2150.
What makes the HD 3300 better? While every other IGP design we’ve seen to date has pulled its video memory resources from shared system memory—a notoriously slower design—the 790GX takes the bold step of integrating a DDR3 chip directly on the motherboard for the IGP to use as video memory. (Just so there’s no confusion, that’s DDR3 for video memory but still DDR2 for system memory under AMD’s current platform.)
“Using a separate video chip allows us to save energy, because now we can put the CPU and memory into a lower power state and just refresh the screen from this little piece of memory,” says AMD chipset product manager Adam Kozak. “The chip will usually be 128MB or 64MB. That may sound small for graphics, but when we’re talking IGP graphics, it doesn’t matter if you go higher than that. We’ve tested it and seen that there’s a bottleneck in other places in the system such that increasing the video memory doesn’t help.”
So as an integrated graphics platform for the mainstream, the 790GX is the new king. But what if our buyer has a few extra bucks at the end of the month and decides he’d like to beef up his graphics? Between $30 and $50 buys a new Radeon HD 3450 card. Add this into the 790GX board’s first graphics slot and you’ve suddenly got the ability to run in a Hybrid CrossFireX configuration, which bonds the discrete and integrated GPUs. There is one caveat. The way AMD’s drivers are built today, the graphics cores in play must all be of similar generation and speed, such as both engines being from the 3200/3300/3400 group, although this limitation may relax in the future. Still, according to AMD, a hybrid configuration such as this can obtain up to 70% better performance than if the system were running with only the discrete card.
AMD takes an interesting tack with demonstrating performance. I like that the company is moving us away from seeing everything in terms of absolute frames per second benchmarks. Does anyone need 212 fps instead of 190 fps when the human eye can’t really perceive anything more than 30 fps? No, instead AMD conveys the scalability of the 790GX’s graphics in terms of visual quality. For example, in the popular game Mass Effect, the 790GX’s integrated graphics can deliver 32 fps at 800 x 600 resolution, ultra settings, with no shadows enabled. That’s not stunning stuff, but it’s a lot better than some other recent IGPs that might have only shown a black screen when the user tried to play the title. Using an HD 3470 card in hybrid mode, AMD plays the same game at the same 32 fps, only now it’s at 1024 x 768 resolution, ultra settings, and shadows enabled. That’s a respectable leap in the gaming experience. Taking things up to the top with a pair of HD 4850 cards, we see 32 fps, 1920x1200 resolution, ultra settings, shadows enabled, film grain, and 2X antialiasing.
I really like the hybrid graphics concept. So often in computing, the user has to abandon part of his investment if he wants to upgrade, particularly if he starts out on a budget. Hybrid won’t give you 1:1 scaling—you always have to sacrifice a part of the total compute capability of the two cores for overhead—but it’s a lot better than the usual case of having to disable the IGP when a discrete card is added.
But hybrid isn’t the end of the graphics road with the 790GX. The platform also supports two PCI Express 2.0 graphics slots. Specifically, there are 16 PCIe 2.0 lanes coming off the 790GX northbridge dedicated to discrete graphics. When the first graphics slot is populated, all 16 lanes are utilized by the slot. If another Radeon card is added to the second graphics slot, then the two share the 16 lanes, with eight lanes for each.
Today, there aren’t any graphics cards able to surpass the bandwidth of a x8 PCIe 2.0 connection. So don’t feel like this is any kind of bottleneck. Because the 790GX supports CrossFireX, users can plant two of the highest-end cards available (that’s the HD 4870 presently) on the motherboard and get some of the fastest graphics performance available on a desktop anywhere. That’s pretty hot for a platform that started out as a humble IGP setup.
Of course, there’s more to the 790GX than just graphics. Users get DirectX 10 support, a move to the new and more power-friendly HyperTransport 3.0 bus, dual independent display off the IGP, and output port options including DVI, HDMI, and DisplayPort. (Note that with an HD 3400-series card, the user gets quad-head output capability for practically pennies. Productivity-minded buyers should eat this up.) Also built in is ATI’s UVD engine, the industry’s top integrated, hardware-based decode accelerator for playback of MPEG-2, VC-1, and H.264 video. This serves to keep CPU utilization low and thereby help with reduced cooling.
Of course, the point of all that decoding is to enjoy a great multimedia entertainment experience. Typically, IGPs have topped out at 1920 resolution, which is fine if you’re on a 21" monitor. But once we start talking about new interfaces like DisplayPort and screens larger than 30”, you need more pixels than most IGPs can crank out. That’s why it’s important to remember that the HD 3300 core can deliver 2560 resolution, so as the display gets bigger, the user doesn’t have to take a hit on quality.
GO SOUTH TO OC
We all know about how to overclock a system by digging into the BIOS settings and changing factors such as the CPU frequency, voltage, bus speeds, and northbridge. The southbridge, to the best of my knowledge, has never once factored into the overclocking discussion—until now. When used with a Phenom Black Edition CPU (told you we’d get back to that), the SB750 is able to communicate directly with the processor and obtain a hitherto impossible overclock through a feature called Advanced Clock Calibration. I’ll let AMD’s Kozak use his own words.
“Because we’re one company now, the engineers worked on a low enough level to figure out how to create a communication path between the two components. It’s not a HyperTransport link. The CPU has pins for ins and outs, right? It is able to send data and communicate to the northbridge, as every CPU does today. We just take some of those signals and match them up to our southbridge, as well. I don’t think there’s a brand or anything on it. It’s just a communication path. We talk to the pins on the CPU.
“We know how the CPU works, and we can give it hints on how to overclock better just by changing some internal settings. It’s not playing with voltages. It’s not playing with anything that’s typically been done before, so these options aren’t normally exposed. Because no one’s done it before, we don’t really want to give out hints on how it’s done. But in our current slides, we’ve been showing a bar that says on any platform, this is the highest this CPU is capable of overclocking. Like you cannot go beyond this. And I take that CPU with the same setup, and with the same voltage settings on this platform, you can actually push it higher. Keeping everything else the same, it’ll go higher because of this extra function we have. You turn on the function and it gives you more frequency headroom, usually a couple speed grades—two, three, or four. Now we can push Phenom to 3.3, 3.4, 3.5, even on air cooling—territory that today we can’t normally hit.”
Kozak told us as much as he could, but the fact is that AMD is keeping a very tight lid on Advanced Clock Calibration. It’s more or less the competitively advantaged secret sauce out to capture the attention of today’s enthusiasts. Previously, a voltage setting of 1.35V could get a current Black Edition chip up to, say, 3.0 GHz. You could increase the voltage, but the CPU had hit its frequency brick wall, and trying to push it further only sacrificed system stability. Now, combining increased CPU voltage with Advanced Clock Calibration delivers unprecedented results. A 1.375V setting will get that same chip up to roughly 3.2 GHz; going to 1.45V will likely reach 3.4 GHz, according to Kozak. Getting another 10% or so above the previous overclocking limits is a huge deal for those who either like to tweak or simply get more bang for their buck. Just remember that if your user is going to make a habit of running with extra voltage to consider adding a water cooler on that CPU
In last month’s cover story, we discussed AMD’s OverDrive utility, the Window’s-based UI for both novices and experts that gives users full control over every facet of processor and memory overclocking. OverDrive is now updated with an option to enable or disable Advanced Clock Calibration.
I could be wrong, but I don’t think AMD is set to displace Intel as the desktop performance king, not with the kinds of benchmarks being pulled in by Extreme Edition platforms. However, that’s not the enthusiast group AMD is after. AMD wants to let mainstream buyers feel and act like enthusiasts, but at mainstream price points. Consider the Black Edition 9850 on a 790GX board versus the Intel Core 2 Quad Q9300 on a P45 board. According to AMD, you’re going to get considerably better performance from the AMD solution for over $100 less. Based on my own examination of current prices and independent reviews, I’d say this is about right. Non-overclocked, Intel’s Q9300 performs about one speed grade above the Black Edition, but with Advanced Clock Calibration in play, I suspect we’d either reach performance parity or show AMD with a slight advantage. Add all of the 790GX’s other scalability benefits on top of that, and you have AMD with a clear edge on total value.
AMD says it offers more for less. Normally, I chalk up such statements to marketing hyperbole. With the 790GX, though, I think AMD may have finally hit the mark.
|Back to top|
Copyright © 2008 RAM Magazine. All rights reserved.
Do not duplicate or redistribute in any form.