CPUS GPUS MOBOS & CHIPSETS CASES & POWER STORAGE MOBILE FORUMS

Contemplating SpeedStep and Cool'n'Quiet in performance testing
by Scott Wasson — 11:07 AM on August 28, 2009

Howdy all.  I've been hard at work in Damage Labs setting up new test systems with Windows 7.  This is a particularly agonizing chore for me, because I want to be sure to set up everything properly and perfectly the same (as much as possible) between the different systems.  Also, generally what happens is I go into this process looking to incorporate as many new benchmarks as possible, but then for various reasons (time constraints, poor application performance scaling, lack of counters for timing operations, software licensing/DRM restrictions), I end up using many of the same tests as in the past generation of results.  That kind of looks to be repeating itself in this new round of CPU tests, although I do expect to add a few new games, 7-Zip, and Windows Live Movie Maker, at least—along with new versions of a great many applications.

The question of the day, however, has to do with power management features.  Typically, we've left features like SpeedStep and Cool'n'Quiet disabled for our general performance tests, only enabling them when we do power efficiency testing.  We've disabled them for multiple reasons, mainly because they can affect performance results in some tests.  The picCOLOR application benchmark we've used for a while, for instance, does many short, quick operations spaced a little bit apart.  As a result, the CPUs don't have time to ramp up their clock speeds for each operation, and the results come out lower than with power management disabled.

I also have a sense that, generally speaking, when cases like this occur, AMD processors are more likely to be negatively affected than Intel processors, in part because AMD chips tend to drop to lower clock speeds at idle and in part because of a history of real problems with AMD CPUs, CnQ, and performance.

The question is: Isn't that fair game, though?  "Balanced" is the default power profile in Vista and Win7, and the vast majority of folks are going to want to have these power-saving features enabled on their systems in order to cut down on the noise, heat, and power consumption of their systems.  In a case like the picCOLOR benchmark, I think we have a simple solution: use a workload more like a real user would, with higher-resolution images and longer operations.  We can work with the software's developer on that.  In other cases, well, most of our benchmarks already reflect real-world use pretty well and aren't so affected by SpeedStep and Cool'n'Quiet.  If they are, the odds are pretty good that a real user might experience the same drop-off in performance, and even if it's not perceptible, there's no reason not to include it in our performance measurements.

I'm of two minds on this question.  The one sticking point that keeps me from making to switch and leaving power management features enabled is the possibility that our test results will be rendered unreliable in some cases, either due to big differences in outcome from one run to the next or to the occasional outright incompatibility between these features and a piece of software (which we've seen in some games in days past).  And I'm on a deadline here, so that prospect frightens me more than you might expect.  Still, it's 2009, I've been using CnQ and SpeedStep on my own desktop and laptop systems for years, and I consider them integral features of a modern CPU.  Seems like it might be time to test 'em like we use 'em.

Hmm.  What do you all think?

32 comments —
Last by derFunkenstein at 8:08 PM on 09/14/09

Gearing up for Lynnfield and Win7
by Scott Wasson — 11:34 AM on August 24, 2009

59 comments —
Last by Clint Torres at 6:31 PM on 08/31/09

On fireworks
by Scott Wasson — 3:41 PM on July 6, 2009

In the past few years, my brother and I have arrived at a place where we have the resources and venue to purchase and use, well, real fireworks, of the sort that aren't legal in an urban or suburban setting.

These days, with Safety Nazis on the prowl with ever-increasing assertiveness, you might be thinking I'm talking about sparklers or snakes. Such a mistake could be forgiven.

But I am not, and I'm not talking about ladyfingers and bottle rockets, either. I'm talking about the sort of fireworks that, used improperly, could blow your foot off or set the house on fire. You know, the fun ones.

We've been able to buy and detonate an increasing number of such devices because, for the past couple of years, we've had access to a house in the country, where no one particularly cares what sort of fire-related things—fireworks, firearms, brush fires, fire whiskey, various combinations of the preceding elements—one does for amusement.

During this span of time, we've begun to develop an understanding of what sort of fireworks we tend to enjoy and, quite literally, how to get the most bang for your buck. After this year's multi-hour extravaganza of flame and smoke, once the kids were down to bed, my brother and I sat outside on his front porch, beer in hand, considering the good and the bad of our most recent efforts. Some of this knowledge is a culmination of a life-long trajectory of blowing stuff up in mid-summer for my brother and me, but keep in mind that we are still relative newcomers to the higher-budget, higher-payload adult versions of these things. Make of this info what you will; perhaps you'll find it useful should you find yourself in possession of some rural land and a few hundred bucks to spare around the Fourth of July.

Much of what we've learned simply involves types of fireworks—which ones are a sham, and which ones are good and worthy. Here's what I think of each:

Kiddie stuff — Sparklers, snakes, tanks, flowers, party poppers, snaps, and the like. A necessary evil, but handled with an economy of time and money by buying a "kiddie pack" at the country store. 12 bucks, five seconds, and you can move on to inspecting the more important things.

Oh, and I suppose I should have mentioned: If you are buying your fireworks at a suburban tent run by the local Boy Scout troop, you're doing it wrong. Find a semi-rural store in a real building that's open all year round. These places sell to the fund-raising crowd, pre-markup, and they will sell to you, as well. A bit of a drive will be worth it.

Fountains — Although technically not considered kiddie fireworks, these devices do not explode. I've looked into it, and this behavior is apparently by design. As a result, during this firework's moments of action, your mother may take the opportunity to turn to your wife and tell her a story of some sort, perhaps about when you were a baby. Fortunately, armed with this knowledge, you may avoid this class of firework entirely.

Rockets and parachutes — My two absolute favorite sorts of fireworks since childhood have fallen sharply out of favor with me in recent years. Fireworks parachutes have maybe a 30% chance of actually opening properly with current quality control standards, which is a heck of a buzzkill. Worse, the gunpowder used in fireworks turns out to be a lousy rocket propellant. You'll get a larger payload up to higher altitudes with an artillery-style mortar shell any time.

Instead, to get your rocket-and-parachute fix, go pick up a model rocket kit from the local hobby store, along with some extra motors. If you have the room to detonate real fireworks, you probably have the room to launch a model rocket. Even the cheap ones will reach ten times the altitude of a skyrocket, and if you fold it right, the parachute will deploy correctly over 90% of the time. Larger model rockets require FAA clearance for launch, which pretty much says it all. A vastly superior alternative.

Spinners — A large spinner, when nailed to a utility pole, can produce a spectacular display for a few minutes that will surely please a crowd. Yet even large spinners can be relatively cheap. What's more, the incredibly strong sense of grass-fire danger more than makes up for the fact that most spinners don't actually detonate.

Bees — Helicopter-style bees, when placed on a flat surface, spin around and lift off into the air, scattering showers of sparks wherever they fly, sometimes to considerable altitudes. Oftentimes, as they begin to burn out, a partially-powered bee will suddenly careen off in a random direction, say toward a toddler, a fireworks stash, or a propane tank. Thrilling for all involved.

And cheap thrills! We got a pack of six large bees for $3.75 this year. Given the chance, I'd take that deal over again several times next year.

"Homemade" — A key component of any country fireworks display is at least one firework not sanctioned by any government body, something large and menacing, about which grown men speak only in hushed tones. There are two keys here. One is finding a source for such a device, preferably through a friend of one's cousin's in-laws or some such, who can ensure quality and who probably has never heard the words "plausible deniability," yet understands the concept perfectly. The other key is escalation: each year's "special feature" must outdo the last, hopefully via a larger force, and most optimally in surprising fashion.

I'm not saying we had such a special feature in this year's display, but if we had, I suppose it would have created a breathtakingly powerful blast, coupled with a 3'x6' white flash, that penetrated the sheet of plywood on which it sat and created a crater underneath. Had that happened, I might have giggled uncontrollably for two minutes or more in the aftermath.

Cakes — A fireworks display in a box, or cardboard cylinder, or a little house made to look like a gazebo, cakes are typically not cheap. They are, however, often very good values. Even medium-sized ones can burn for a few minutes and incorporate a dazzling amount of variety. Each year, we buy a larger, heavier, and more expensive cake to use as our grand finale, and each year, we are pleased with the outcome. When presented with a choice between multiple cakes of the same basic size and price, we have taken to lifting them up to determine which is heaviest, since it likely includes the most powder.

In addition to a large cake finale, we had several medium-sized cakes and even a few small ones this year, and we enjoyed them all. One goal for future years is to incorporate more cakes by saving on rockets and small stuff. Cakes are to be preferred to fountains in all cases, regardless of size.

Mortars — Artillery-style mortar-shell fireworks have become the heart of our annual fireworks extravaganza. These are typically smaller versions of the sort of fireworks used in commercial or municipal displays. The payload is launched out of a tube, and it then detonates in the sky in some colorful, round pattern. The right kind will get you arrested in most suburban settings, but where there's room, these will let you pretty well replicate the experience of a professional display on your own terms.

Mortars generally fire a single shot skyward, which may then detonate one to three times. Large cakes are often multi-shot mortars launched in sequence, but they tend to cost more for each shot. We understand that mortars are the most cost-effective way to get the job done, but we are still learning which ones are best. This year, we took dual tracks: the low road and the high road.

The low road consisted of a six-shot box of Black Cat-brand fireworks for $3.50, special deal. We took this deal four times, giving us 24 medium-sized, single-shot shells. As expected, these shells weren't the largest, didn't travel the highest, and weren't the most spectacular of displays, but for the money, they were a fantastic supplement to our larger selections.

The high road consisted of a fairly pricey box of artillery-style mortars with one to three detonations per shell. Our selection was from the Shogun brand and had the words "Commercial grade" emblazoned on the box in large letters—certain proof that they were not, you know, really commercial grade. Still, they were large, menacing, and recommended by the folks at the fireworks shop.

And I have to say, for sheer power, they did not disappoint. Both the launch detonation and the subsequent explosions had a concussive power beyond anything we'd used before.

Unfortunately, for all of that, they didn't reach great heights before exploding, and those detonations didn't yield much that was particularly creative or memorable compared to, uh, whatever brand it was we used last year or, crucially, whatever brand the surrounding neighbors had chosen. We would have gladly traded some of the pop for more sizzle. Next year, we'll probably try a few different brands of mortars, if we have the means. We know this is the way to go, but the path holds perils we did not anticipate.

Of course, if you have suggestions for a brand we should try our a source we should investigate, feel free to let us know by posting the comments below. With your help, we'll have loads of fun, one blown-off finger at a time.

26 comments —
Last by eric93se at 12:50 PM on 07/12/09

A new addition in Damage Labs
by Scott Wasson — 3:05 PM on July 2, 2009

The results of my latest project are somewhat complicated to interpret at times but very positive overall.  Ladies and gents, please welcome Charles Jacob Wasson, born Sunday morning at 8:39AM.

"C.J." weighed nine pounds, two ounces at birth.  He and his mom are both happy and healthy, and we're all at home now.  This baby promises to be relatively easy as these things go, because, well....

Helpers!  His big brother and sister are often contending over who gets to hold him, and they've both been great help so far.  I've not yet instituted diaper-changing training, but that's on the agenda when the time is right.  For now, the whir of systems in Damage Labs has ground to a halt as I take care of C.J. and his mom.

On a related note, congrats to Marco at HotHardware on his new addition, too.  We are brothers in lack of sleep.

64 comments —
Last by PerfectCr at 7:37 AM on 07/09/09

A compact, fast, and quiet PC for the kitchen
by Scott Wasson — 12:54 PM on June 25, 2009

If you've read our summer system guide, you might enjoy hearing the story of how our very cool Pocket Swiss Army Knife config came to be.  Those of you who have been around here for a while will know that my affinity for small form-factor systems like Shuttle's XPC series goes way, way back.  Simple computers that are small and quiet can go a lot of places a bigger box cannot.  My favorite SFF build is an excellent example.  We've had a system in service continually on our kitchen counter, in one form or another, since the debut of the original Kitchen PC a shocking five years ago.  Since then, my enthusiasm for Shuttle XPCs has waned for a smattering of reasons, including their limited upgradeability, the fact that Shuttle no longer innovates or targets enthusiasts with bare-bones boxes in quite the same way, and some disturbing long-term quality and reliability problems with XPCs—which I got to witness first-hand through different incarnations of the Kitchen PC.

Recently, the last incarnation of the Kitchen PC was suffering from a painful collection of age-related problems.  The box whined and growled in a creative array of scary sounds indicating friction, it would occasionally reboot itself out of spite, and it gave off the faint odor of Ben Gay at all times.  The thing was not a good citizen on our kitchen counter any longer, and it threatened to croak at any time.  So I sat down and ordered a bunch of new parts to replace it, and darned if the thing didn't up and die before the packages could arrive from Newegg, as if it knew what was coming.  After a quick dissection, I determined that the power supply had given up the ghost, most likely because it was worn out from having to supply too much juice to the hard drive, which was making a horrid hissing noise (I thought it was just a bad fan!) and probably would only spin its platters with some effort.  Since the thing's XPC enclosure used a proprietary power supply, it had to operate in Borg mode, with a different PSU hanging out of the side, until the new bits came to us.

This is an attractive ornament for the ol' kitchen countertop, let me testify.  Great for entertaining.

Anyhow, when I ordered the parts, I was determined to make sure our next Kitchen PC would be darn near silent and at least as small as the last one, but what to do?  I considered a range of options, and was sorely tempted to fork over a few hundred bucks for an Eee Box and be done with it.  But my wife wanted something with a DVD drive it it—the Kitchen PC is her main computer—and I wanted something standards-based that I could build from existing parts and upgrade over time.  I'd been watching the development of Mini-ITX hardware for a while, but I was worried by the choice of enclosures.  Many of them seemed to be cheap, build wise, and yet some were pretty expensive at the same time.  The user reviews on Newegg were disconcertingly mixed for almost all of them, and I couldn't be sure that even the most attractive choices would be truly quiet.

Then I happened upon the Silverstone SG05.

The SG05 is about the size of an XPC enclosure, is sized to accept Mini-ITX motherboards, and comes from Silverstone, whom I've learned to trust for quiet PC enclosures and PSUs.  This case has a single, large (120mm) fan in the front that cools both the enclosed PC and the built-in 300W power supply unit.  The thing is affordable, too (only $99 right now).  Best of all, at the time, Newegg had a combo deal featuring the Zotac GeForce 9300 Mini-ITX mobo.  Using that board would deliver me from my extended flirtation with a dual-core Atom processor and allow me to use a faster off-the-shelf Core 2 processor with upgrade potential down the road, to boot.

Of course, the Zotac board wasn't my only option, because Mini-ITX mobos come in many flavors.  It was just the best option.  But the SG05 closed the deal.  At long last, it enables an SFF ecosystem that is standards-based, upgradeable, and has the footprint and quiet cooling potential of a Shuttle XPC.  I could build in this case and rebuild a few years down the road with a new mobo, no problem.

Thus was born the Kitchen PC Mk III:

The SG05 and Zotac 9300 board supply most of what you need to build a complete PC, including GeForce graphics and Wi-Fi.  I threw in a Core 2 Duo E7200 processor, a hard drive, and a couple of gigs of RAM.  My choice for the hard drive was a 320GB Caviar SE16, since it was the quietest drive I had on hand, according to Geoff's tests.  I was ready to order up an SSD if it was too loud, but that proved to be unnecessary.

The hardest part of the build, on which I spent the most time, was making sure the system was quiet.  The only real bone of contention in this case was the CPU cooler; a stock Intel model made more noise than I liked, and the Zotac board's fan speed control pretty much had a mind of its own, ignoring the choices I set in the BIOS.  I believe those problems have since been fixed with a BIOS update, but I found my own fix ahead of time, using a Zalman fan speed controller cranked to its lowest setting to limit the fan speed on an old LGA775 cooler that's pretty large.  I can't find that same cooler on sale at Newegg now, but I'm hopeful the Masscool unit we picked for the guide performs similarly.  Because, wow, that thing is blessedly quiet and yet cools the E7200 just fine with both cores cranking away on Prime95.  Sweet success—and near silence.

Here's how the new Kitchen PC looks in its native environment.

As I've done before, I was able to hide most of the wires behind the counter and refrigerator to keep the whole setup looking clean.  I'm very pleased to report that this system is literally inaudible over the gentle hum of the refrigerator next to it.  This box is much faster than an Atom-based system would be, and the prospect of upgrading it down the road at will, like any normal-sized PC, makes me feel like a longer term problem has been solved.

One more thing.  Do pay attention to the guide's inclusion of the special SATA power adapter for the slim-line optical drive if you decide to build a system in the SG05.  You will need it, and most of us don't have one of those handy.

56 comments —
Last by grantmeaname at 11:12 AM on 06/30/09

Laptoppery
by Scott Wasson — 11:36 AM on June 11, 2009

Time for some laptoppery. After spending the past month working on servers and desktop CPUs, I have the urge to get out of the office and go mobile. Since my own main laptop, the Sharp M4000 WideNote, is three-and-a-half years and aging, I have the upgrade itch pretty badly. The M4000 has been excellent in almost every way, but age has taken its toll: the touchpad lists to one side and periodically loses its scroll area, the Wi-Fi isn't up to today's standards for range and reliability, and it's on battery number three and counting. Meanwhile, Sharp has pulled out of the mobile market in the U.S. Feels like it's time to look elsewhere.

I did make one concession to the M4000's age last summer, when I bought an Eee PC 1000H. I really like the 1000H, and I use it with regularity for surfing on the couch and the like, but the 1024x600 display is just too small for extended work. I can write on it, but I can't edit or do anything involving page layout. Not without feeling cramped very quickly. I had hoped to learn from having the 1000H around how I would use such a device compared to the aging M4000, and the answer is clear: the 10" netbook simply can't replace my 13.3" ultraportable, mainly due to the screen resolution and perhaps partially due to the small keyboard. Those are the only two barriers, really, which is remarkable.

The question is what to do next.

One approach, which I really like, is to continue with my upgrade installment plan. I figure I could buy a new netbook (or something similar) every 6-8 months and wind up spending no more than I would on upgrading an expensive laptop (as the Sharp originally was) every three years. This approach has many merits, especially with the quick progress being made right now on things like fancy touchpads, displays, and battery life. One could step up to the latest and greatest thing at appropriate inflection points once or twice a year and never be outmoded, whereas even the most expensive ultraportable will seem a little tired in the back half of a three- to four-year lifespan.

This way of doing it also meshes with my sense that the right laptop for me may just be "as many as possible."

Continuing with this approach, to me, means moving into larger-than-1024x600 displays and affordable ultraportable systems that may not quite qualify as "netbooks" anymore. Only a few are on the market, but many more are coming. The ones selling now include the Nano-based Samsung NC20, the sleek Mac Air knockoff and CULV-based MSI X340, and the 11.6" version of the Acer Aspire One, which is apparently now available at Walmart but not online yet. The HP Pavilion dv2 isn't in the running because I need longer battery life and, well, I hate its keyboard. (Glossy key caps? Please.) Now is a troubling time to buy, though. These few systems are just the first fruits of a wave of promising ultraportables on the way, including Asus' 11.6" Eee PC and a number of systems based on the CULV processor.

A more traditional laptop would better handle HD video content, and there are options for well under the traditional two-grand mark now. I'm intrigued by the Samsung X360, with a 13.3" screen, 2.8 lbs weight, and a claimed 10-hour battery life for $1500. Other choices include the Portege A600 and, going more expensive and exotic, the Thinkpad X300 series and the MacBook Air. (No X200, thanks. Need touchpad.) Were it not for the fact that I could have three Samsung NC20s, with excellent build quality and similar dimensions, for that same price, I'd be more enthused about these options. One of these pricier systems would have to remove all doubt about its superiority in every way in order to justify itself. I don't see that happening.

Part of the problem is that my ideal system isn't out yet, but the potential is there. I basically want a larger version of the Eee PC 1000H, whose little 10" display is shockingly good in terms of color reproduction and off-angle viewing, and whose multi-touch touchpad is addictively satisfying to use. Give it the 1000HE's seven-hour battery life, of course. If they could add in the excellent keyboard of the Samsung NC20, that'd be nice. And let's keep the weight well under four pounds, thanks. That mix right there would do it for me.

Then again, I'd pay more for decent integrated graphics from Nvidia or AMD, along with a CULV or Athlon Neo processor. Second core is optional.

The thing is, one can dream up just about the perfect system built from components currently on the market. We're just waiting now for someone to put them all together into a nice design. Surely that can happen soon, right?

—

On another note, I should mention that Lenovo has almost too predictably heard a huge outcry over its discontinuation of ThinkVantage System Update service and has brought it back. (Thanks to TR reader Prototyped for the heads up.) Good for them, I suppose, but they finally admitted (for real) why they did it, and it's as lame as you might fear: the bandwidth and server costs were too much for them. Really, in this age of hosted services and incredibly cheap CPU power and bandwidth, I find that explanation incredibly weak.

Also, re-activating System Update now requires (as far as I can tell) finding out that it's back via some outside source (I got no Message Center notices about it) and manually downloading the new version from Lenovo's web site.  What do you think their attrition rate is on that? More than likely, just stepping through this process bought Lenovo an 80%-plus reduction in update traffic. Clever, perhaps, but at the continued expense of their reputation.

Still, the new System Update software seems to work well on the T60 I have here, and that's much better than nothing.

57 comments —
Last by rhema83 at 8:40 PM on 06/24/09

Some speculation about that Larrabee die shot
by Scott Wasson — 1:44 PM on May 14, 2009

We don't yet know as much as we'd like to about Intel's upcoming Larrabee GPU-CPU hybrid, but enough useful information has leaked out over the past little while to give us the ability to speculate a bit. Intel has disclosed many of the architecture fundamentals, but one of the big missing pieces of the puzzle has been the specific number of cores and other types of hardware that the first implementations will have. The release of a fuzzy die shot yesterday, therefore, caused a bit of a stir around here, with the TR editors sitting around peering at their monitors and exchanging puzzled IMs about what's what.

I started forming some theories eventually, and after poking around online, I was pleased to see that some folks in the B3D discussion thread had some similar ideas. We don't really know much about the particular chip shown in the die shot, but given what we know about the architecture from Larry Seiler's Siggraph paper and Michael Abrash's overview of the instruction set, some possibilities become apparent.

If you look closely at this high-res version of the die shot, you'll see that the chip is laid out in three rows. The design of the chip looks to be fairly modular, with repeating areas of uniform structures of several types. The most common unit of the chip is most likely the x86-compatible Larrabee shader core, and the dark areas at the ends of its long, rectangular shape are probably cache of some sort, either L1, L2, or both. We know that each core has L1 data and instruction caches, plus 256KB of L2 cache. By my count, there are a total of 32 cores on the chip—10 on the top row, 12 in the middle, and 10 in the bottom row.

Along with the cores are two other types of regular blocks on the chip. The larger of these two is a little narrower than a core and has a lot of dark area, which suggests cache or other storage. I count eight of those. There's also one other block type, a narrow column, of which there are four total, two in the top row and two in the bottom. (After I had sorted all of this out myself, I saw this B3D post with an excellent visual aid. Worth a look if you can't identify what's what.)

My best guess is that the eight larger, dark-and-light blocks are texture sampling and filtering hardware. Larrabee doesn't have as much dedicated hardware as most GPUs, but it does have that.

After spending some quality time with the color-coded RV770 die shot at the bottom of this page and noodling it around with David Kanter, who bears no responsibility for any of this mess, I'm betting the logic bits running along the upper and and lower edges of the die, outside of the cores and such, are the memory pads. I see four repeating patterns there. Kanter notes that the four narrow columns on the interior of the chip are perpendicular to the memory pads. They are relatively evenly spaced, protrude from the edge of the chip into the center, and thus could be memory interfaces and other assorted logic that participates on the bus and talks to the I/O pads. So the magic number for memory interfaces would appear to be four.

David also suggests it might be fun to play "Where's Waldo?" with the fuses, analog nest, and any other logic we'd expect to find in a GPU. We're guessing the PCIe interface logic is along the right edge of the chip. Some other unidentified, non-repeating bits are on that side of the die.

I spent wasted some time trying to figure out the relationships between the cores and these other bits of hardware, but there don't appear to be any clear groupings of blocks or physical alignments between cores and texture units. More than likely, each of these resources is just a client on Larrabee's ring bus.

Happily, with no more information than that, we can tentatively pretend to start handicapping this chip's possible graphics power. We know Larrabee cores have 16-wide vector processing units, so 32 of them would yield a total of 512 operations per clock. The RV770/790 has 160 five-wide execution units for 800 ops per clock, and the GT200/b has 240 scalar units, for 240 ops/clock. Of course, that's not the whole story. The GT200/b is designed to run at higher clock frequencies than the RV770/790, and its scalar execution units should be more fully utilized, to name two of several considerations. Also, Larrabee cores are dual-issue capable, with a separate scalar execution unit.

If I'm right about the identity of the texture and memory blocks, and if they are done in the usual way for today's GPUs (quite an assumption, I admit), then this chip should have eight texture units capable of filtering four texels per clock, for a total of 32 tpc, along with four 64-bit memory interfaces. I'd assume we're looking at GDDR5 memory, which would mean four transfers per clock over that 256-bit (aggregate) memory interface.

All of which brings us closer to some additional guessing about likely clock speeds. Today's GPUs range from around 700 to 1500MHz, if you count GT200/b shader clocks. G92 shader clocks range up to nearly 1.9GHz. But Larrabee is expected to be produced on Intel's 45nm fab process, which offers higher switching speeds than the usual 55/65nm TSMC process used by Nvidia and AMD. Penryn and Nehalem chips have made it to ~3.4GHz on Intel's 45nm tech. At the other end of the spectrum, the low-power Atom tends to run comfortably at 1.6GHz. I'd expect Larrabee to fall somewhere in between.

Where, exactly? Tough to say. I've got to think we're looking at somewhere between 1.5 and 2.5GHz. Assuming we were somehow magically right about everything, and counting on a MADD instruction to enable a peak of two FLOPS per clock, that would mean the Larrabee chip in this die shot could line up something like this: 

In the numbers above, I'm betting that GDDR5 memory will make it up to 1GHz by the time this GPU is released, and I'm counting on Intel's texture filtering logic to work at half the rate on FP16 texture formats. We can't determine the pixel fill rate because Larrabee will use its x86 cores to do rasterization in software rather than dedicated hardware. I'm just working my way through Michael Abrash's write-up of the default Larrabee rasterizer now, but I don't think we can assume a certain rate per clock given how it all works.

Obviously, clock speed makes a tremendous difference in this whole picture. Nonetheless, we're looking at a potentially rather powerful graphics chip, at least in terms of raw, peak arithmetic. If the tile-based approach to rasterization is as fast and efficient as purported, then the relatively pedestrian memory bandwidth quoted above might not be as much of an obstacle as it would be for a conventional GPU, either.

That's my first crack at this, anyhow. Would be cool if I turned out to be more right than wrong, but it's all guesswork for now. At the very least, one can begin to see the potential for Larrabee to compete with today's best DX10 GPUs. Whether or not it will be effective enough to contend with tomorrow's DX11 parts, well, that's another story.

82 comments —
Last by FroBozz_Inc at 3:08 PM on 05/20/09