|
||
|
|
|
              |
|
 |
 |
|||||||||||
|
 |
|
|||||||||
By William Van Winkle |
|||||||||||
AMD vs. Intel: TESTING NOTES We don't do a lot of benchmark testing at RAM for two reasons: First, because you can find benchmark tests in ample supply on the Web (AnandTech, Tom's Hardware, HotHardware, on and on). Second, because benchmark roundups often have little relevance to actual value when selling. Unless you're catering to that top 5% of enthusiasts, the difference between a part that enables 227 fps and 214 fps is about nil in the real world. That said, what we can infer from such numbers if discussing GPUs is the amount of horsepower available to perform physics processing, antialiasing, and so on. But since we're talking about CPUs alone, the situation gets a bit easier. If all other system factors are left constant, you know that processor X producing faster numbers than processor Y means you've got a chip better suited to the application type you're benchmarking. But this number is somewhat meaningless if you don't factor in price at the same time. With performance and price, you can begin to get a better sense of overall value, and value is what you're pitching to most buyers. Comparing two different processor platforms is always a muddy affair, and we made several choices when planning these tests that might have kept our results from being the absolute highest possible. (Again, poke around the Web for higher results, although you'll find we're in the same ballpark.) The first choice to make was chipsets. Most FX vs. Conroe match-ups to date have pitted something like the ATI Radeon Xpress 3200 versus Intel's 975X. However, NVIDIA was generous enough to provide us with a near-production reference board for its nForce 590 SLI Intel Edition chipset along with matching 7800 GT cards to run in SLI mode on it. This offered a rare opportunity to perform AMD vs. Intel comparisons on the same chipset and thus have a more level playing field. Moreover, since we were now dealing with an SLI platform, it only made sense to run tests with a moderately high-level dual-GPU setup as this is likely what buyers of such high-end chips will also be running. The decision to use NVIDIA GPUs also reflects our bias toward using as many components from the same manufacturer as is prudent in order to better assure driver stability and efficiency. Suffice it to say that we had zero graphics-related trouble during our testing.
From here, other component choices fell into place. On the AMD side, we opted for ASUS' flagship AM2-based motherboard, the M2N32-SLI Deluxe/Wireless. Enthusiasts would be hard pressed to choose between this board and a hot date with the Swedish Bikini Team—both have a similar number of impressive features and dazzling performance. From integrated wireless to eight-phase power to wall-to-wall copper heat piping, this board has everything you could ask. That also includes twin dedicated x16 SLI slots and a BIOS with enough overclocking options to make your head spin. We didn't even attempt to mess with the advanced OC options. This board is a recent release, though, which means you can expect the BIOS updates to arrive in rapid succession. Version 0403 came out of the box, and we installed 0603 during setup. Even still, we discovered that the board would not install Windows XP Pro SP2 to our Hitachi Deskstar 500GB SATA drive. Swapping out to a 500GB Seagate Barracuda 7200.8 immediately solved the problem, so that's the drive we used. Also be warned that the first SATA port is blocked if you have a full-length graphics card, such as our reference 7800 GT. (Of course, that still leaves five primary SATA ports open, plus another internal SATA behind the PS/2 connector block and an eSATA port on the backplane, all of which can be RAIDed courtesy of the 590 chipset.) For memory, we went with a 2GB kit of OCZ's PC2-7200 (900 MHz) Platinum SLI-Ready Edition DDR2 modules. This was a dicey choice. We wanted OCZ on the basis of its aggressive timings, impressive history of voltage tolerance, and, most of all, SLI certification. However, after bringing in the OCZ samples, NVIDIA intimated that the 590 SLI chipset would perform best at 1T settings, but to get it we'd need to use a select set of Corsair modules, the CM2X1024-6400C3 kit (3-4-3-9). So having given our nod to Corsair here, we went ahead with the OCZ modules for the reasons given above, because we'd already completed the FX-62 testing, and because we're reluctant to optimize a platform for one select memory SKU. As it turns out, our NVIDIA reference Intel board would not POST with 1T selected and the OCZ SLI modules installed. We should also mention that we used OCZ's new Tempest heat pipe cooler, which comes with mountings for both socket types, as well as OCZ Ultra 5+ silver thermal grease, which has always given us excellent results. Also, the OCZ PowerStream 520W power supply that has sat on our test bench for years has finally been replaced by the just-released 700W OCZ GameXStream power supply, which lacks the adjustable rails of the older unit but is quieter, supports both ATX12V v2.2 and EPS12V, has more device connectors, is remarkably lighter in weight, and boasts 80% power efficiency—an increasingly important spec in our opinion. Let's admit right up front that we're not overclocking zealots. Every board and processor will yield different results, so spending countless hours trying to pin down the most precise setting possible is an exercise in pointless minutia. This is why we do OC testing in 5 MHz frequency jumps and only apply modest voltage increases to the CPU and memory. (Specifically, for all three configurations, we maxed the CPU at 1.55V and the memory at 2.3V.) We're out to make a point, not split hairs. Similarly, we have a policy of running memory modules at the vendor's default timings. So when OCZ says its PC2-7200 parts are 4-4-4, we go into the BIOS and manually set for 4-4-4. After all, these are the timings customers pay for. However, this carries a potential downside. Both motherboards offer options for linking the memory bus speed to the FSB, and we enabled this on both platforms. But like most boards, there is an option for loading "optimized settings" which no doubt leverage the Enhanced Performance Profile (EPP) settings outlined by NVIDIA for SLI-approved modules. On our NVIDIA reference board, this resulted in 4-5-4 timings, which let us benchmark reliably at a 1280 MHz bus speed. Forcing the timings back to 4-4-4 resulted in lock-ups all the way down to just over 1245 MHz. Also note that some applications show slower performance as you edge into unstable speeds, 3DMark06 and SPECviewperf chief among them in our tests. Examining the Results There are few surprises here. Special kudos go to AMD for providing us with its flagship desktop processor knowing full well what we would find. That takes chutzpah and should be commended. The truth is that the FX-62 is a great chip. If we were to confine our discussion to comparing the FX against anything based on NetBurst, most would have to agree that AMD has the superior product. Or to put it in the words of AMD's Damon Muzny: "Just because Intel finally doesn't suck doesn't mean that AMD inherits all their baggage or badness. Our products are still excellent products."
But the timing is what it is, and third quarter is Intel's moment to shine after five years of hard lessons and deep soul searching in the R&D department. We've seen these two companies leapfrog each other many times, and this is the turn when Intel gets to jump ahead. The difference this time around is that Intel is out for blood, margins be damned. It's not enough that the company's mainstream CPUs now beat AMD's top of the line without breaking a sweat; Intel is now delivering this performance at far below half of AMD's retail point. The situation is unprecedented if not absurd. In fact, one recent comparison over at Digit-Life.com shows the FX-62 averaging about even to slightly behind the 2.13 GHz E6400, which has a 1,000-unit price of $224—a mere 27% of the FX-62. AMD almost manages to pull even with the E6700 in our DOOM 3 test and falls only a step or two behind in 3DMark, which may lean more heavily on the shoulders of our SLI cards than most of the other tests. Generally, the E6700 shows at least a 10% edge over the FX-62 at stock speeds. The AMD chip only delivered overclocking headroom of just over 10 percent, which is relatively low for an Athlon chip but not necessarily out of expectations for a top-bin SKU. Stock performance is already near the high end of that core's present spectrum. What surprised us was that Intel's E6700 and X6800 only offered 18% and 17% overclocking headroom respectively within our testing conditions, although recall that the Extreme part uses an 11X multiplier rather than the E6700's 10X, is unlocked, and we were testing on a board that did not allow for multiplier changes. (In fact, we had to change the FSB in the memory settings screen. Weird.) Intel's D975XBX will doubtless provide more impressive overclocking results for the Extreme part, as will other third-party 975 boards that provide for multiplier changes. (NVIDIA's subsequent BIOS update does provide for multiplier adjusting, placing NVIDIA on par with Intel's D975XBX for unleashing the Core 2 Duo Extreme's overclocking potential.) Still, having seen Yonah chips overclock well in excess of 40% on our bench, we have to wonder if Intel is being more stringent in its Conroe binning in order to keep a tighter reign on product positioning and margins...what's left of them. Which brings us to the inevitable comparison between the mainstream Core 2 Duo chips and the Extreme Edition parts. As covered before, there is no architectural difference between them. The XE merely uses higher grade silicon and comes set out of the box at a higher frequency. At $999 versus $530, the X6800 posits a nearly 100% price premium over the E6700 while averaging less than a 10% performance gain at stock settings. For most people, those are hard numbers to swallow, especially when you consider that an E6700 with a modest 10% overclock under air cooling can generally outperform the X6800 out of the box. That's $500 that can be put to good use elsewhere in the system, such as beefing up the multi-GPU rigging and stepping up from integrated audio to Creative's higher fidelity X-Fi cards. That said, it's plain that the X6800 is still the crowned title holder in the performance race. Top dog is top dog, and some buyers have the ability and motivation to settle for nothing less.
"For what could amount to a 10% to 20% performance gain, people who are buying in that space will pay it," says Intel's Todd Garrigues. "Yes, it's a $1,000 processor, but they're probably buying a PC or workstation that costs at least $2,500. So in the workstation case, say you got 10% or 20% performance increase in your CAD application, for example. If that were true, and you could save that much of your processing time every day, is that worth a few hundred bucks? It likely is if it means additional projects or workloads getting completed so your clients can generate more revenue. For the consumer gamer, it's all about bragging rights. If they've got the money and want to spend it—I mean, up to 20 percent? I remember when the difference between $800 and $400 was three or five percent! Bottom line: There are people willing to pay for the best performance on the planet, and that part will give it to them."
Garrigues raises an intriguing point. Does the Core 2 Duo XE have a role to play in an application space generally served by workstations, meaning the Xeon family? Garrigues senses that the majority of contemporary workstations are uniprocessor systems owing to the relatively strong performance of dual-core/single-processor machines versus dual-core/dual-processor configs. Also, multi-GPU horsepower is an increasingly big deal in the rendering world, but there are few dual-processor, dual-graphics platforms on today's market—hence AMD's 4x4 move. What remains unclear is how much benefit the pertinent applications will show once a once a dual-processor/dual-graphics solution arrives, much less when it will arrive in quantity. Keeping it Real We've talked about a lot of technical and testing details so far, but now we need to pull back and examine these topics from a market perspective. What really matters today? How can AM2 and Core impact your business? Some of the answer hinges on what you'll actually be able to buy. A quick peek at Newegg as of late July shows the full complement of AM2 parts in stock while no Core-based SKUs had yet to arrive. Could Core actually be a paper launch? "Conroe looks great on paper and in the hands of some reviewers, but what can you go out and buy right now?" asks AMD's Damon Muzny. "You can go out and buy Pentium D now, and the reviewers are bludgeoning Pentium D to death now that Conroe's out. The question is, how long until you can buy Conroe and how long until it's a significant part of the product mix? Because all the scuttlebutt I'm seeing says that even by the end of the year, it's still going to mostly be Pentium D and Pentium 4. Is it a bait and switch where you come in after the Conroe review, but when it's not there you say, "Aww, I guess I'll buy the Pentium D instead"? We laugh at that, but mom and pop just might do it." "Based on what I've heard," adds AMD's Moorhead, "only 25% of the Intel desktop mix will be Conroe by the fourth quarter, and I've got a pretty good idea who will have those. It'll be the guys in Round Rock and the guys in Houston. But I'm sure some enthusiasts will have them, as well." You can't buy better F.U.D. than that. Score one more for AMD's press machine. However, we hopped across the fence and asked Intel's Garrigues for an official reply. He stated that AMD's statement is partly true—on a global level, including chips in irrelevant markets to this discussion, such as emerging and embedded markets. By the end of July, there were essentially zero Conroe chips sold to dealers. But by September, Garrigues expects that practically all desktop chips priced over $180 will be Conroe. Looking purely at North American sales through distribution—not to tier-ones, e-tail, etc.—total Core desktop numbers in Q3 will be in the mid-teens. By early October, this number will be around 30%, and by the end of 2006 fully half of all Intel desktop chips should be Conroe models. "Let them believe what they want to," says Garrigues of AMD. "Our intention is that this will be the fastest ramp in our history. Seriously. We wouldn't have priced Conroe starting in the $180 range if we didn't intend to sell a ton of them. And as for the Pentium D, what's wrong with that? If customers are price sensitive, this is one heck of a part. It outperforms all but the top end of our competitor's stack at much lower pricing. We will have a 3.4 GHz, dual-core, 4MB L2 part at $169 or so by the time you're in print. That's one major workhorse for the money." Next, we need to consider positioning. Given the sweeping breadth of Conroe's pricing and performance, it remains the undeniable choice for performance buyers and enthusiasts at least until the holidays when AMD's 4x4 is set to arrive. But the processor is only one part of the package you need to entice this audience. Probably the leading second piece of the puzzle is graphics, and both NVIDIA and ATI (until recently) have been scrambling to align themselves as the de facto option to pair with Conroe. In NVIDIA's case, this means planting Conroe on a motherboard outfitted with an nForce4 or nForce 500-series SLI chipset for Intel processors.
"SLI is a huge deal for the channel," says NVIDIA spokesperson Bryan Del Rizzo, "because resellers went back to Intel and said, ‘We can't sell Conroe with CrossFire. There's not enough demand for CrossFire or ATI high-end graphics cards.' So they want a combination of SLI and Conroe to be a compelling thing for them to sell consumers. That's why you're seeing a bunch of promotions from Intel and NVIDIA for both Conroe with NVIDIA GPUs as well as Conroe with nForce motherboards." Specifically, NVIDIA is launching several Conroe promotions. Some of the most aggressive focus on bundling the nForce4-based ASUS P5N32-SLI SE with Conroe and one or two GeForce 7900 GTXs, two 7800 GTs, or a single 7950 GX2. This raises the issue of Quad SLI, drivers for which went live in the channel on August 7th. Many resellers want "ultimate" gaming configurations, and while modern reviews show ATI's X1900 XTX edging past NVIDIA's 7900 GTX, there's no denying that four top-end NVIDIA GPUs harnessed for Quad SLI blow past two ATI GPUs in a CrossFire rig provided the applications being run can take advantage of the extra horsepower. Quad SLI, which today means a pair of 7950 GX2 cards, can run on any SLI motherboard, but users will see the best results on a platform with two dedicated x16 PCIe slots. Currently, this covers either the nForce 4 SLI X16 or nForce 590 SLI, although expect more options in the near future. Of course, there are two sides to every story. ATI director of North American channel sales Larry McIntosh quipped in return to us that "[NVIDIA] barely has a working [Quad] solution, and about 95% of the bigger system builders are launching [Conroe] with ATI CrossFire systems." It doubtless helps that CrossFire is supported on Intel's flagship motherboard, the D975XBX, which has been promoted by Intel up until now as the ultimate bedrock to put under a Conroe chip. But that takes us to the recent news about AMD's acquisition of ATI, which throws everything into doubt. As of this writing, the merger announcement is only days old, and none of our Intel contacts cared to speculate about the impact that the manufacturer realignment would have on Intel's motherboard products or the ways in which Conroe would be promoted heading into the fourth quarter. (That said, it seems rather telling that Conroe's launch event on July 27th was absolutely devoid of any CrossFire or ATI products.) We speculate further on this matter in this issue's What Matters column, but suffice it to say here that with the AMD/ATI merger not being finalized until the fourth quarter, both SLI and CrossFire will remain excellent ways to add value to either AMD or Intel high-performance processors. All finger-pointing and chest-pounding aside, Core and AM2 both have substantial implications for your sales today. Core represents a performance leap of roughly 40% over the prior NetBurst parts, and AM2 has a platform roadmap full of great promise. Both offer power savings benefits, and both have great stories to take before your clients now looking to replace systems that are three or four years old. From Vista compatibility to 64-bit support to virtualization to interplay with the Viiv and LIVE! digital home platforms, these two chip families have the ability to reshape how people can use their PCs and elevate them to entirely new levels of functionality and productivity. No matter which camp you or your customers prefer, interest in these two platforms is only going to elevate over the coming months. For those who may not keep abreast of developments in the CPU world, you will rarely have windows as good as this to crank up the outside sales engine and approach your clients with the value propositions carried by these processors. For example, the AM2 chips people buy today can drop into 4x4 upgrade boards in six months. Because Conroe shares the same architecture as Woodcrest, many of the Intel Active Management Technology (iAMT) benefits that make Xeon so persuasive now also apply to desktops, which may prove to be a huge plus for businesses of any size. The days of lackluster announcements about tiny speed increases are over (for now). Seize this opportunity to stoke a new flaring of interest in new CPU technologies and the benefits they can enable. Core, AM2, or even C7, the news is all good and bound to make the second half of 2006 unseasonably hot. |
|||||||||||
 Back to top
|
|||||||||||
Copyright © 2007 RAM Magazine. All rights reserved.
Do not duplicate or redistribute in any form. |
|||||||||||
