Noise cancelation should work if the Zune could accurately determine the position of it's headphones with it's microphone, since it knows what the Zune is playing. Knowing that, it knows the distance of the headphones. With two microphones it would be able to triangulate the headphones in 3d space.
You've got a bad picture of what science is capable of (been watching too much TV?)
Triangulation requires two things:
* at least two DIFFERENT measurement points. * a measured line-of-bearing to the signal at each measurment point (this is provided by your microphone array, using phase or amplitude deltas).
You take the intersection between multiple line-of-bearing measurments to triangulate a signal source. This means you would have to make sure that the microphone array "moved" in relation to the sound source, or you would get nothing more than a constant line-of-bearing measurement. Last time I checked, most of the time your head moves very little relative to your waist. See Wikipedia, as always, for further details.
Another problem with your concept: an array of two microphones just measures the line-of-bearing in the 2 dimensions along whatever plane is formed by the two microphones. This means your triangulation location solution is only 2D, and in a 3D world that doesn't work.
For eample, let's say you used a truck with an antenna to triangulate a radio signal. After driving around, taking two or more measurements, you know where the radio is on the plane of the earth's surface...but you have NO IDEA what the altitude is. It could be 30,000 feet in the air, or 10,000 feet below sea level, and you wouldn't know - the reason triangulation WORKS in that case is you make an assumption about the altitude - perhaps you assume the emitter is on the ground, and that the ground is at sea-level.
The solution to this problem? You'd have to add another array perpendicular to the previous one, and combine measurments to generate a 3D location. This raises the cost of the unit significantly, since triangulating in 3 dimensions requires twice as much receiving hardware, and beefier processing. And since sound cancellation is touchy, and sounds crappy with inaccurate hardware or timing, you'd probably want a more accurate 3D location, and that requires more array elements. See where this is going?
They also had trouble meeting that 5w spec, as Van's Hardware made painfully clear. The processor throttled down to 533 MHz under load, even though the real parts were advertised as running at 833 MHz or 1GHz without need for active cooling. The fact of the matter was, the processor used a LOT more than 5w under full load...and when cooling wasn't available to handle that heat, it throttled to maintain that 5w celing.
But this was late in the lifetime of Transmetta "the CPU manufacturer," and they were in dire need of sales. Had they been unable to hit those power numbers, they would have sold zero product. Banias LV, by that time, was seriously challenging Transmetta's hold on the low-power market, and could run circles around anything Transmetta could deliver.
Another thing to note: the Efficeon couldn't match the performance of a Pentium 4 1 GHz. No, it's more like a Pentium 4 CELERON 128k 1GHz. The Pentium 4, with 512k or 1MB L2 cache, is about %30 faster clock-for-clock than a Celeron 128k.
It was a nice idea...code morphing + VLIW processor sounded so cool, and Crusoe practially introduced the idea of low-power design to the industry. But the fact is the implementation ate cache for breakfast (and more cache means MUCH more die area and power consumption), and the performance just wasn't there even with the increased cache. Now that the low-power design concepts have been adopted by the entire x86 industry, Transmetta is no-longer important.
What I don't get is why DU gets all the bad press, and white phosphorous, lead and napalm don't. Hell, if you want to look at the really nasty stuff left over after a war ends, landmines beat all of the above. Why does it only become "nasty" when it's got the slightest hint of radioactivity? Oh right, because it's that evil nucular stuff, so it must be worse... somehow.
I think it's because your average person doesn't understand the science behind nuclear reactions, so it's mostly "magic" to them. They lack the mathematics and physics background to contemplate how unstable atoms behave, and why they might give off alpha, beta or gamma particles.
Magical power is something to be feared. Since the media know they can't educate people, they instead use that powerful fear as a ratings tool.
The rest of the things you listed: napalm, landmines, etc...these are chemical and mechanical things. Even a country bumpkin who makes moonshine or lights his farts recognizes basic chemistry, and probably has mechanical aptitude. With that background, these other "nasty" weapons (even in great quantities) are not nearly as powerful as magic.
What's wrong with making copies of the kid's DVDs?
Step 1: install DVDShrink.
Step 2: insert Disney DVD
Step 3: 20-cent DVD-R
Step 4: relax, and let your kids destroy yet another 20-cent disc.
You could even teach them to value the discs they break by refusing to replace the broken one for a week/month/year/decade. I wonder if they could learn the same value lessons from something as intangible as a fileserver?
The AWESOM-O 4000! Guaranteed to do everything from sweeping the floors to writing new movie scripts.
Soon to be released:
Star Trek 11, starting Adam Sandler. Star Trek 12, starting Adam Sandler. Star Trek 13, starting Adam Sandler. Star Trek 14, starting Adam Sandler. Star Trek 15, starting Adam Sandler.
I wonder wat the negative side of this is, there has to be.
The negative side is, you're competing with companies that not only make the engine, they also publish their own games using the cutting-edge engine before anyone else.
ID Software is one of the big culprits when it comes to this. Each major engine release is almost completely different from the previous, and includes many new features. If ID releases a new engine (game), games in development (using the previous engine) are instantly outclassed.
Take, for example, when ID software released Quake II. It was a much more advanced engine than Quake, and was released less than two years after Quake. The Quake II engine was also structured much differently from Quake, making it harder for developement teams to transition. Development teams currently working on Quake-engine products were screwed - they knew they HAD to upgrade, because Quake II (the game) would make their game look like crap. In the end, they either extended the capabilities of the Quake engine (Half-Life, for example), or wasted months moving the entire project to the Quake II engine (Daikatana, for example). Obviously, unless you have talented engine programmers, your options are limited to the painful transition.
Now, today engine design seems to have slowed consideraby, with 3 or more years in-between major ID engine releases. Epic has also somewhat remedied this problem with incremental engine upgrades, but there's still some lag time on feature releases.
That's just because they're the "pick of the litter." With arare execption (I knew one), only the smartest people in China get the chance to go to US colleges. The people who don't make the cut attend Chinese colleges. This is as-opposed to the US applicant pool for colleges, where some less-capable students are admitted for many possible reasons (sports, "legacy" child of alum, racial quotas, less applicants than previous year, etc.)
You just THINK foreigners are statistically smaller, but the fact is you're not seeing a statistically random sample group.
DDR2 also gives AMD the same power consumption improvements Intel has enjoyed. The lower signaling and operational voltage of DDR2 means less power consumed by the memory. The lower signaling voltage also means less power consumed by the memory controller.
DDR2 was an essential move for AMD to cut power consumption on their mobile chipsets.
Are there things I need to do to improve the performance?
Yes, and they are easy to do. You need to turn off all the transition crap, this makes Win2k MUCH more responsive on slower machines (for reference, I once ran Win2k on a P1660 with 48MB ram. It was slow but usable...but these day's I'd try for 64-128MB just to handle some of the richer websites.
1. Do NOT run a virus scanner. They typically tie up 32-64MB of ram, and with only 128 your machine is going to stall.
Just do the occasional online virus scan.
2. Disable Fade Effect: right-click on the desktop, Properties->Effects Tab->Change Fade Effect to Scroll Effect (Win98
effect), or disable entirely (Win95 effect).
3. Disable Active Desktop in explorer. Start->Programs->Accessories->Windows Explorer. On Tools->Folder Options, under
the Active Desktop section choose Windows Classic Folders.
4. Disable mouse pointer shadow (yet another software transparency effect). Start->Settings->Control Panel->Mouse. Under
the Pointers tab, uncheck Enable Pointer Shadow.
Now you have the equivilant to a Windows 98 desktop with Active Desktop disabled (or Windows 95 desktop if you removed all transitions). There's more you can do, but these are the easy ones.
What's so hard to believe? Onboard graphics have had performance to run Quake III since Nvidia released the NForce chipset. With an onboard GeForce 2 MX core and dual-channel DDR, the NForce could run Quake III almost as fast as the standalone card.
The GMA 900 and 950 series are a little higher in performance than the GeForce 2 MX, and they also have pixel shaders. What really holds the GMA 9x0 back is lack of hardware vertex shaders or TnL. It can play older games just fine without hardware (just like cards like the Radeon 7000VE and the Kyro II used to), but newer games crawl without hardware vertex shaders.
Even in 3dmark 05 the "software vertex shader" GMA 950 beats the much beefier GMA 3000, and in the test with Quake 3 you can see how much the switch back to TWO ROPs (GMA 950 had four) hurts overall performance. GMA 900 is a flop under DX9, and it will continue to be a flop under DX10.
And, as always, never forget that the latest onboard graphics chipsets from ATI and Nvidia kick the crap out of GMA 950. Intel has a LONG way to go to produce a good performance embedded graphics solution.
The performance problems of the G part of the G965 aren't going to improve, even with the C2 stepping.
I wouldn't hold your breath. It took Intel three tries to make a modern embedded chipset with "decent" low-end performance, with the GMA 900, and that wasn't even very advanced. With hardware shaders, this is an entirely new concept for Intel to tackle, so give them till next chipset to beat the performance of the GMA 950.
Of course, the Inq could be full of shit, but when it comes to Intel and graphics I'd be skeptical if OEMs DIDN'T call it crap.
I'm surprised you didn't make the comment, but your figures are more proof that OS X is treading water.
Take a look at the downward trend or Mac OS, and how it almost matches the upward trend of Macintel. There's some lag between the two (on a per-month basis), but over the course of the year MacIntel increases 0.62 and Mac OS decreases 0.64. What this indicates to me is that, for the most part, OS X is just barely treading water. Most new purchases are from existing Mac users (which, from my experience, would be consistent with their loyalty).
Yes, I agree with most of the posters here that we should ignore the "total marketshare" numbers, but the relative marketshare of Mac OS versus Macintel has much more meaning, and may indicate a trend of zero growth. This is not to say new customers AREN'T buying mew Macintels (I bought one, and I never owned a Mac in my life), it's just that they've also lost some customers with the move to Intel.
My ass. You can buy a 27inch HDTV for $500 or less and a LCD HDTV of 26 inches for $600ish.
But let's take your point into consideration. When you move up to a 16:9 screen with the same number of (diagonal) inches, you lose vertical screen size. Now, this may not be an issue with 16:9 feeds, but for 4:3 feeds (still the majority of content), you really notice the loss of screen real-estate.
Let's say you have a choice between a 27" SDTV ($200) and a 27" HDTV ($500). Sounds good, right? Only $200-300 more for a much better picture, right?
The only problem is, the move to the 27" 16:9 HDTV will display all your 4:3 shows MUCH smaller. Take a 27" SDTV: it has a vertical screen size of 16". Now take your 27" 16:9 HDTV: it has a verical screen size of only 11.5". So, the diagonal screen size stays the same, but you trade vertical inches for horizontal inches. On a 27" wide-screen HDTV, the effective screen size for 4:3 content will be the same as a 19 inch 4:3 SDTV! What a downgrade!
The fact is if you want your SDTV programs to look as good as they did on your old 4:3 set, you have to STEP UP when you buy an HDTV, and this is what people always seem to miss. In order to get the same 4:3 area as your 27" SDTV (same vertical screen size), you have to buy a 38" wide-screen monster! That will set you back at least a thousand bucks, probably more.
But what kicked ass was you could link 4 voodoo 2's together and destroy nearly any game made back then
Somebody has a bad memory. You can link TWO (2) Voodoo 2 cards together in SLI. It is possible to link more together, as 3DFX did in their arcade boards, but this requires custom glue logic.
The Voodoo2 was great in that it could use memory from other graphics cards. It's funny to run GLQuake and see 268 megs of memory pop up on the console while loading.
That's just not true. The only successful card to use memory from other graphics cards was the PowerVR. The card had onboard memory, but this was simply a texture cache. The chip processed tiles using the on-chip tile cache, and after a tile was rendered it was transfered via PCI bus to your 2D card's framebuffer.
If GL Quake is reporting the combined framebuffer size, that's a bug in Quake, or it is being misreported by Windows.
3DFX was the last holdout on combining all their chips into a single core logic for mainstream product lines.
Example: in 1996 when 3dlabs designed the Permedia, it was a multi-chip solution (just like their workstation products) consisting of a pixel and vertex processor. In 1997, 3dlabs combined the multi-chip Permedia into the single-chip Permedia 2. Despite being priced mucn cheaper than the Permedia, the Permedia 2 made 3dlabs much more money due to the low-cost, single-chip design.
3DFX designed the Voodoo Graphics as a multi-chip solution (just like their arcade boards), and they were high-priced due to the cost of a multi-chip solution. Even worse were the Voodoo Rush cards, which required 3 chips, and didn't work properly. 3DFX raised that cost even higher with the Voodoo 2, which required THREE chips for a 3D-only solution. They also increased the PCB complexity by requiring THREE 64-bit independent busses.
What they should have done after the Voodoo Graphics got them recognition was release something like the Voodoo 3 (with reduced clocks), but they put that off in favor of the Voodoo 2 because they could release it earlier. Later, they released the cut-down Banshee, and they made the mistake of marketing it (and pricing it) as a performance product, instead of a midrange part designed to entice OEMs.
Near the end of the year, other competitors released chips that were much better than single Voodoo 2 cards for the same price. The Banshee barely kept up in the price war with the TNT and Savage 3D, and the Voodoo2's price plummeted as a result of that price war. The market for Voodoo 2 cards saturated, and because 3DFX had no way to reduce the build cost (thanks to their multi-chip design), they took losses.
So, by the end of 1998, consumers were left confused by all the inconsistency. All 3DFX fans had to purchase were overpriced Voodoo 2 cards that required a 2D card, and all they had to look forward to was the Voodoo 3 (same performance as Voodoo 2 SLI, 6 months down the road, big deal). The only impressive card 3DFX released in 1998 was the Voodoo 2, but it was only impressive for the first half of the year. 3dFX never saw the "big picture" that was the single-chip 2D/3D card until it was too late.
There used to be the high end the mid range and the low end
No, even back in the "goold old days" that you are obviously not remembering correctly, ATI and Nvidia were guilty of confusing tactics and far too many "models". Let's take your two examples:
The NV25-based GeForce 4 series of cards were most known for the confusing release of the GeForce 4 MX, which was really a GeForce 2 renamed. Without the DirectX 8 features of the true Geforce 4, the GeForce 4 MX had no business carrying that name. The cards were also confusing for people purchasing them because Nvidia released confusing refresh models with AGP 8x.
*** Also, the Ti4800SE (clocked at 275 / 550) was actually SLOWER than the Ti4600 (clocked at 300 / 650), another annoying and confusing product release.
Just like today, there are TONS of confusing model names with often contradictory performance.
The R300 (Radeon 9500-9800) generation used R300 for mid-range and high-end, and left the low-end to old R200 variants. This created confusion, because even though the Radeon 9000-9250 were in the same "thousand," they had a totally different feature set.
As you can see, the name game took over following the refresh, just like it does for ANY product set (including Nvidia's latest).
Cards (in performance order):
DirectX 8.1
Radeon 9000, 9000 Pro
DirectX 9
Radeon 9500, 9500 Pro, 9700, 9700 Pro
LATER REFRESH:
DirectX 8.1
Radeon 9100, 9200, 9250
DirectX 9
Radeon 9550 SE, 9550, 9600 SE, 9600, 9600 Pro, 9600 XT, 9800 SE, 9800, 9800 Pro, 9800 XT
Kinda confusing, ain't it?
To card manufacturers, we're sorry about the press leaks but you better match last generations top model with a mid price card or you're going to get destroyed eventually.
They DID. Or were you not paying attention when Nvidia released the 7600 GT 256MB for $200 (same price as the 6600 GT 128MB on release), and amazed the crowd with the performance potential of fast memory on a 128-bit bus?
Were you just incapacitated when review sites across the planet proclaimed the 7600 GT to stomp all over the 6800 GT, and even slightly outperform the 6800 Ultra? The only last-generation card with better performance is the x850 XT series, and it's not by much.
NOW, you have the 7900 GS released in the same price point as the 7600 GT was 6 months ago ($200). For that increase in price, you get 20 pipes instead of 12, and a 256-bit bus, GUARANTEED to best a 7600 GT by a huge margin. How can you NOT love this?
Maybe you find the number of models released confusing, but rest-assured, they are releasing better and better cards at the $150-200 price-point. On the Nvidia side you have the 7900 GS, and on the ATI side you have the x1900 GT, and they're both excellent cards.
If you actually look at the link, you'll notice that the model there has a reference fan and heatsink. Both the listed specs and the picture mention this.
The passive model you are looking for is not sold on Newegg...in fact, I can't find it for sale anywhere. Tiger Direct no longer carries the item, and it's out-of-stock at Mwave. At $330, it has a sizeable price premium over other 7950 GTs.
See, that's the problem when buying from XFX - they have TOO MANY models, so you can easily get confused and buy the wrong one.
As Sadko posted, the fan mounts are standardized, so it is easy these days to purchase a "one size fits all" fan upgrade.
Furthermore, by upgrading to a better heatsink with a larger fan, one can run the fan at extremely low RPMs, making it virtually silent.
For example, I replaced the fan on my 7900 GT (REALLY loud) with a Zalman vf900. The heatpipe design makes this cooler incredibly efficient, and it also comes with a Fanmate to allow you to tune the speed to your tastes. I was able to run my card overclocked on the LOWEST Fanmate setting (inaudible), and still see a 10C reduction over the stock cooler.
Note that I have a fairly quiet PC case, based on a Sonata II with Scythe 120mm S-FLEX FDB fans, so you can rest assured that when I say "inaudible" over the rest of the case, this actually means something.
So, you get better cooling than any "passive" solution, and no noticeable increase in noise. That's why it may be a better idea to go with Zalman. I've heard of low-airflow, "quiet" or "silent" cases having issues with "passive" heatsink performance video cards (which REALLY require medium airflow to be cooled well), so I stick with low-noise active coolers. With the excessive premium you pay for passively-cooled video cards, the extra cash for a Zalman cooler isn't that much.
do real-world comparison tests. Run similar tasks over a period of time on machines built on the 2 architectures and compare the actual power demand.
Two reasons I can't:
1. I don't own a C7 system.
2. Regardless of how thoroughly I set up the test, there would be constant naysayers, and people with other ideas of how to erform the test. In other words, it wouldn't solve anything. Remember how it took almost a year for the computing community to realize just how much reduce the power consumption of the A64 90nm parts were over the 130nm A64s, not to mention Prescott? It mostly had to do with poor testing with incomplete information, plus conflicting results due to this.
So, let's do the next best thing, and take real-world numbers and try piecing them together. Unfortunately, there's a real dearth of actual REVIEWS. If you put C7 and review into Google, you actually get a lot of PREviews, but no solid benchmarking.
The 256k cache plus the single channel memory combine to really hamper the Sempron's true performance (As a representative of the Athlon 64) in some benchmarks, but it still managed to come out shining in benchmarks that don't depend wholly on memory bandwidth and cache size. Thanks to the massive AMD price drop in July (after this review), you can now buy an Athlon 64 AM2 plus a 3200+ for that price, so I'd say pump up the benchmark results %10-20 across the board to get a feel for what you could buy for that money today.
Even with all the talk about low-power states and fast state switching, even though the Sempron is clocked %33 faster, the Sempron with the trusty old Cool 'n Quiet enabled managed to consume less power. How embarassing. If we can trust these test results, not only will the Sempron system finish processing ages earlier, it will also use less power.
Yes, AMD's pressure has pushed Intel to make a lot of changes for the better.
AMD's 386DX 40 Mhz pushed Intel to release faster 486 chips...otherwise Intel would have ridden their overpriced 486DX 33 forever.
AMD and Cyrix produced Pentium clones which pushed Intel and forced them to reduce prices.
AMD's push to revive Socket 7 (Super 7) with the introduction of the 100 MHz bus and the K6-2 forced Intel to release the Mendocino Celeron. With on-die cache, it was one of the best budget gaming processors ever released. The K6-2's "3D-Now!" instruction set forced Intel to finally admit that MMX was a failure, and release an addentum in 1999 with SSE.
AMD saved us from the hot, expensive, high-latency world that is Rambus. Without AMD pushing the industry-standard PC-133 SDRAM and later DDR SDRAM, Intel would have made Rambus the defacto desktop memory standard.
Fast-forward to today: AMD has been at parity or better for years, and it has slowed their improvements. After the cancelation of the K9 project, AMD sat on its ass, just releasing small improvements for K8. But now, with Conroe wiping the floor, AMD has to get back in gear, and is releasing K8L next year.
It's the way of the business. When you're on-top, you tend to stagnate.
While it is true that you can buy any chip you can imagine in a server, the original poster gave me the impression that he/she wanted a cheap solution with a simple chip-on-a-board (ala PegasOS). Unfortunately, the money is in complete systems tied in with services, so that's the last thing you'll catch IBM selling.
And sure, IBM's chips are popular in consoles, but that's mostly because IBM is the only major chip house that will offer to develop custom chip designs. The game console companies help fund the R&D bill and sign-on for some minimum initial shipment, and IBM makes some money. IBM takes existing core logic and pieces it together to optimize performance for consoles, and out pops a new chip design.
But there's a good reason IBM has time for game consoles: IBM doesn't have NEARLY the sales volume of AMD, let alone Intel. AMD doesn't have time for custom chips because they have enough trouble just meeting market demand for their x86 products. They are shooting for 100 million chips next year once their new fab ramps up, and Intel is shooting for 300 million. IBM, by contrast, will sell only 3-20 million chips per-console, per-year, depending on popularity, and those sales rise and fall periodically as consoles get introuced and mature. Even with all three consoles in their pockets, don't expect IBM to sell more then 30 million per-year.
I don't really see what the attraction is to small-market chips like CELL anyway. My prediction on the subject has already come true: way back when IBM announced the CELL, I predicted that by the time it was released, the PC market would have a similar-performance chip on-offer.
With Intel's release of the quad-core Core2 products later this year (Clovertown), my prediction becomes reality. With 2x128-bit SIMD units per core (for a total of 8 128-bit units per chip), Clovertown is capable of the same performance levels as Cell. AMD's K8L, due out next year, is expected to boast similar performance levels.
It's sad how few people realize their "efficient" Via CPU is not so efficient?
Clock-for-clock, the optimized WinChip core (yes, even the C7 uses the very same core) can only process one integer and one floating-point instruction in parallel. This makes it 2-3x slower per-clock than modern CPUs. So, while you're still waiting on your Via C7 to crunch those numbers (at 20w), a Core2 Duo or A64 X2 system can do it in 1/4-1/6 the time (at 35w), and clock down to low-power state (3-5w).
So, I hope you feel good about how much carbon Via saves building the chip, because not only does it uses more power than competing processors to do the same amount of work, it takes longer too:D
Via's day in the sun is over. They were faced with the poor performance of the Winchip core, and instead of redesigning it, they touted the low power (which is true) and efficiency (which is not true). Intel and AMD responded with innovations like real-time voltage and frequency adjustment, and all of a sudden Via is scrambling just to try and keep up.
Slashdot Mirror
Noise cancelation should work if the Zune could accurately determine the position of it's headphones with it's microphone, since it knows what the Zune is playing. Knowing that, it knows the distance of the headphones. With two microphones it would be able to triangulate the headphones in 3d space.
You've got a bad picture of what science is capable of (been watching too much TV?)
Triangulation requires two things:
* at least two DIFFERENT measurement points.
* a measured line-of-bearing to the signal at each measurment point (this is provided by your microphone array, using phase or amplitude deltas).
You take the intersection between multiple line-of-bearing measurments to triangulate a signal source. This means you would have to make sure that the microphone array "moved" in relation to the sound source, or you would get nothing more than a constant line-of-bearing measurement. Last time I checked, most of the time your head moves very little relative to your waist. See Wikipedia, as always, for further details.
Another problem with your concept: an array of two microphones just measures the line-of-bearing in the 2 dimensions along whatever plane is formed by the two microphones. This means your triangulation location solution is only 2D, and in a 3D world that doesn't work.
For eample, let's say you used a truck with an antenna to triangulate a radio signal. After driving around, taking two or more measurements, you know where the radio is on the plane of the earth's surface...but you have NO IDEA what the altitude is. It could be 30,000 feet in the air, or 10,000 feet below sea level, and you wouldn't know - the reason triangulation WORKS in that case is you make an assumption about the altitude - perhaps you assume the emitter is on the ground, and that the ground is at sea-level.
The solution to this problem? You'd have to add another array perpendicular to the previous one, and combine measurments to generate a 3D location. This raises the cost of the unit significantly, since triangulating in 3 dimensions requires twice as much receiving hardware, and beefier processing. And since sound cancellation is touchy, and sounds crappy with inaccurate hardware or timing, you'd probably want a more accurate 3D location, and that requires more array elements. See where this is going?
3D passive location is non-trivial.
They had the same performance as a 1ghz pentium 4 with 3 watts power usage.
No, they didn't. Advertised power envelope for the Efficeon at 1GHz was 5w.
They also had trouble meeting that 5w spec, as Van's Hardware made painfully clear. The processor throttled down to 533 MHz under load, even though the real parts were advertised as running at 833 MHz or 1GHz without need for active cooling. The fact of the matter was, the processor used a LOT more than 5w under full load...and when cooling wasn't available to handle that heat, it throttled to maintain that 5w celing.
But this was late in the lifetime of Transmetta "the CPU manufacturer," and they were in dire need of sales. Had they been unable to hit those power numbers, they would have sold zero product. Banias LV, by that time, was seriously challenging Transmetta's hold on the low-power market, and could run circles around anything Transmetta could deliver.
Another thing to note: the Efficeon couldn't match the performance of a Pentium 4 1 GHz. No, it's more like a Pentium 4 CELERON 128k 1GHz. The Pentium 4, with 512k or 1MB L2 cache, is about %30 faster clock-for-clock than a Celeron 128k.
It was a nice idea...code morphing + VLIW processor sounded so cool, and Crusoe practially introduced the idea of low-power design to the industry. But the fact is the implementation ate cache for breakfast (and more cache means MUCH more die area and power consumption), and the performance just wasn't there even with the increased cache. Now that the low-power design concepts have been adopted by the entire x86 industry, Transmetta is no-longer important.
What I don't get is why DU gets all the bad press, and white phosphorous, lead and napalm don't. Hell, if you want to look at the really nasty stuff left over after a war ends, landmines beat all of the above. Why does it only become "nasty" when it's got the slightest hint of radioactivity? Oh right, because it's that evil nucular stuff, so it must be worse... somehow.
I think it's because your average person doesn't understand the science behind nuclear reactions, so it's mostly "magic" to them. They lack the mathematics and physics background to contemplate how unstable atoms behave, and why they might give off alpha, beta or gamma particles.
Magical power is something to be feared. Since the media know they can't educate people, they instead use that powerful fear as a ratings tool.
The rest of the things you listed: napalm, landmines, etc...these are chemical and mechanical things. Even a country bumpkin who makes moonshine or lights his farts recognizes basic chemistry, and probably has mechanical aptitude. With that background, these other "nasty" weapons (even in great quantities) are not nearly as powerful as magic.
What's wrong with making copies of the kid's DVDs?
Step 1: install DVDShrink.
Step 2: insert Disney DVD
Step 3: 20-cent DVD-R
Step 4: relax, and let your kids destroy yet another 20-cent disc.
You could even teach them to value the discs they break by refusing to replace the broken one for a week/month/year/decade. I wonder if they could learn the same value lessons from something as intangible as a fileserver?
What are your accomplishments in the CS field?
I can cook an egg for breakfast!
Yes, they've all ben replaced by...
The AWESOM-O 4000! Guaranteed to do everything from sweeping the floors to writing new movie scripts.
Soon to be released:
Star Trek 11, starting Adam Sandler.
Star Trek 12, starting Adam Sandler.
Star Trek 13, starting Adam Sandler.
Star Trek 14, starting Adam Sandler.
Star Trek 15, starting Adam Sandler.
Isn't technology wonderful?
I wonder wat the negative side of this is, there has to be.
The negative side is, you're competing with companies that not only make the engine, they also publish their own games using the cutting-edge engine before anyone else.
ID Software is one of the big culprits when it comes to this. Each major engine release is almost completely different from the previous, and includes many new features. If ID releases a new engine (game), games in development (using the previous engine) are instantly outclassed.
Take, for example, when ID software released Quake II. It was a much more advanced engine than Quake, and was released less than two years after Quake. The Quake II engine was also structured much differently from Quake, making it harder for developement teams to transition. Development teams currently working on Quake-engine products were screwed - they knew they HAD to upgrade, because Quake II (the game) would make their game look like crap. In the end, they either extended the capabilities of the Quake engine (Half-Life, for example), or wasted months moving the entire project to the Quake II engine (Daikatana, for example). Obviously, unless you have talented engine programmers, your options are limited to the painful transition.
Now, today engine design seems to have slowed consideraby, with 3 or more years in-between major ID engine releases. Epic has also somewhat remedied this problem with incremental engine upgrades, but there's still some lag time on feature releases.
That's just because they're the "pick of the litter." With arare execption (I knew one), only the smartest people in China get the chance to go to US colleges. The people who don't make the cut attend Chinese colleges. This is as-opposed to the US applicant pool for colleges, where some less-capable students are admitted for many possible reasons (sports, "legacy" child of alum, racial quotas, less applicants than previous year, etc.)
You just THINK foreigners are statistically smaller, but the fact is you're not seeing a statistically random sample group.
DDR2 also gives AMD the same power consumption improvements Intel has enjoyed. The lower signaling and operational voltage of DDR2 means less power consumed by the memory. The lower signaling voltage also means less power consumed by the memory controller.
DDR2 was an essential move for AMD to cut power consumption on their mobile chipsets.
Are there things I need to do to improve the performance?
Yes, and they are easy to do. You need to turn off all the transition crap, this makes Win2k MUCH more responsive on slower machines (for reference, I once ran Win2k on a P1660 with 48MB ram. It was slow but usable...but these day's I'd try for 64-128MB just to handle some of the richer websites.
1. Do NOT run a virus scanner. They typically tie up 32-64MB of ram, and with only 128 your machine is going to stall.
Just do the occasional online virus scan.
2. Disable Fade Effect: right-click on the desktop, Properties->Effects Tab->Change Fade Effect to Scroll Effect (Win98
effect), or disable entirely (Win95 effect).
3. Disable Active Desktop in explorer. Start->Programs->Accessories->Windows Explorer. On Tools->Folder Options, under
the Active Desktop section choose Windows Classic Folders.
4. Disable mouse pointer shadow (yet another software transparency effect). Start->Settings->Control Panel->Mouse. Under
the Pointers tab, uncheck Enable Pointer Shadow.
Now you have the equivilant to a Windows 98 desktop with Active Desktop disabled (or Windows 95 desktop if you removed all transitions). There's more you can do, but these are the easy ones.
What's so hard to believe? Onboard graphics have had performance to run Quake III since Nvidia released the NForce chipset. With an onboard GeForce 2 MX core and dual-channel DDR, the NForce could run Quake III almost as fast as the standalone card.
The GMA 900 and 950 series are a little higher in performance than the GeForce 2 MX, and they also have pixel shaders. What really holds the GMA 9x0 back is lack of hardware vertex shaders or TnL. It can play older games just fine without hardware (just like cards like the Radeon 7000VE and the Kyro II used to), but newer games crawl without hardware vertex shaders.
You havn't been paying attention. Go sit in the corner!
Seriously, the G965 has terrible performance compared to the G945 (GMA 950), even when you take into account the new DirectX9/10 hardware vertex shaders.
Even in 3dmark 05 the "software vertex shader" GMA 950 beats the much beefier GMA 3000, and in the test with Quake 3 you can see how much the switch back to TWO ROPs (GMA 950 had four) hurts overall performance. GMA 900 is a flop under DX9, and it will continue to be a flop under DX10.
And, as always, never forget that the latest onboard graphics chipsets from ATI and Nvidia kick the crap out of GMA 950. Intel has a LONG way to go to produce a good performance embedded graphics solution.
The performance problems of the G part of the G965 aren't going to improve, even with the C2 stepping.
I wouldn't hold your breath. It took Intel three tries to make a modern embedded chipset with "decent" low-end performance, with the GMA 900, and that wasn't even very advanced. With hardware shaders, this is an entirely new concept for Intel to tackle, so give them till next chipset to beat the performance of the GMA 950.
Of course, the Inq could be full of shit, but when it comes to Intel and graphics I'd be skeptical if OEMs DIDN'T call it crap.
I'm surprised you didn't make the comment, but your figures are more proof that OS X is treading water.
Take a look at the downward trend or Mac OS, and how it almost matches the upward trend of Macintel. There's some lag between the two (on a per-month basis), but over the course of the year MacIntel increases 0.62 and Mac OS decreases 0.64. What this indicates to me is that, for the most part, OS X is just barely treading water. Most new purchases are from existing Mac users (which, from my experience, would be consistent with their loyalty).
Yes, I agree with most of the posters here that we should ignore the "total marketshare" numbers, but the relative marketshare of Mac OS versus Macintel has much more meaning, and may indicate a trend of zero growth. This is not to say new customers AREN'T buying mew Macintels (I bought one, and I never owned a Mac in my life), it's just that they've also lost some customers with the move to Intel.
My ass. You can buy a 27inch HDTV for $500 or less and a LCD HDTV of 26 inches for $600ish.
But let's take your point into consideration. When you move up to a 16:9 screen with the same number of (diagonal) inches, you lose vertical screen size. Now, this may not be an issue with 16:9 feeds, but for 4:3 feeds (still the majority of content), you really notice the loss of screen real-estate.
Let's say you have a choice between a 27" SDTV ($200) and a 27" HDTV ($500). Sounds good, right? Only $200-300 more for a much better picture, right?
The only problem is, the move to the 27" 16:9 HDTV will display all your 4:3 shows MUCH smaller. Take a 27" SDTV: it has a vertical screen size of 16". Now take your 27" 16:9 HDTV: it has a verical screen size of only 11.5". So, the diagonal screen size stays the same, but you trade vertical inches for horizontal inches. On a 27" wide-screen HDTV, the effective screen size for 4:3 content will be the same as a 19 inch 4:3 SDTV! What a downgrade!
The fact is if you want your SDTV programs to look as good as they did on your old 4:3 set, you have to STEP UP when you buy an HDTV, and this is what people always seem to miss. In order to get the same 4:3 area as your 27" SDTV (same vertical screen size), you have to buy a 38" wide-screen monster! That will set you back at least a thousand bucks, probably more.
But what kicked ass was you could link 4 voodoo 2's together and destroy nearly any game made back then
Somebody has a bad memory. You can link TWO (2) Voodoo 2 cards together in SLI. It is possible to link more together, as 3DFX did in their arcade boards, but this requires custom glue logic.
The Voodoo2 was great in that it could use memory from other graphics cards. It's funny to run GLQuake and see 268 megs of memory pop up on the console while loading.
That's just not true. The only successful card to use memory from other graphics cards was the PowerVR. The card had onboard memory, but this was simply a texture cache. The chip processed tiles using the on-chip tile cache, and after a tile was rendered it was transfered via PCI bus to your 2D card's framebuffer.
If GL Quake is reporting the combined framebuffer size, that's a bug in Quake, or it is being misreported by Windows.
3DFX was the last holdout on combining all their chips into a single core logic for mainstream product lines.
Example: in 1996 when 3dlabs designed the Permedia, it was a multi-chip solution (just like their workstation products) consisting of a pixel and vertex processor. In 1997, 3dlabs combined the multi-chip Permedia into the single-chip Permedia 2. Despite being priced mucn cheaper than the Permedia, the Permedia 2 made 3dlabs much more money due to the low-cost, single-chip design.
3DFX designed the Voodoo Graphics as a multi-chip solution (just like their arcade boards), and they were high-priced due to the cost of a multi-chip solution. Even worse were the Voodoo Rush cards, which required 3 chips, and didn't work properly. 3DFX raised that cost even higher with the Voodoo 2, which required THREE chips for a 3D-only solution. They also increased the PCB complexity by requiring THREE 64-bit independent busses.
What they should have done after the Voodoo Graphics got them recognition was release something like the Voodoo 3 (with reduced clocks), but they put that off in favor of the Voodoo 2 because they could release it earlier. Later, they released the cut-down Banshee, and they made the mistake of marketing it (and pricing it) as a performance product, instead of a midrange part designed to entice OEMs.
Near the end of the year, other competitors released chips that were much better than single Voodoo 2 cards for the same price. The Banshee barely kept up in the price war with the TNT and Savage 3D, and the Voodoo2's price plummeted as a result of that price war. The market for Voodoo 2 cards saturated, and because 3DFX had no way to reduce the build cost (thanks to their multi-chip design), they took losses.
So, by the end of 1998, consumers were left confused by all the inconsistency. All 3DFX fans had to purchase were overpriced Voodoo 2 cards that required a 2D card, and all they had to look forward to was the Voodoo 3 (same performance as Voodoo 2 SLI, 6 months down the road, big deal). The only impressive card 3DFX released in 1998 was the Voodoo 2, but it was only impressive for the first half of the year. 3dFX never saw the "big picture" that was the single-chip 2D/3D card until it was too late.
There used to be the high end the mid range and the low end
No, even back in the "goold old days" that you are obviously not remembering correctly, ATI and Nvidia were guilty of confusing tactics and far too many "models". Let's take your two examples:
The NV25-based GeForce 4 series of cards were most known for the confusing release of the GeForce 4 MX, which was really a GeForce 2 renamed. Without the DirectX 8 features of the true Geforce 4, the GeForce 4 MX had no business carrying that name. The cards were also confusing for people purchasing them because Nvidia released confusing refresh models with AGP 8x.
*** Also, the Ti4800SE (clocked at 275 / 550) was actually SLOWER than the Ti4600 (clocked at 300 / 650), another annoying and confusing product release.
Just like today, there are TONS of confusing model names with often contradictory performance.
Cards (in performance order):
DirectX 7
GeForce 4 MX420, MX440SE, MX 4000, PCX 4300, MX440, MX440SE-8X, MX440-8X, MX460
DirectX 8
GeForce 4 Ti4200, Ti4200-8X, Ti4400, Ti4800SE, Ti4600, Ti4800
The R300 (Radeon 9500-9800) generation used R300 for mid-range and high-end, and left the low-end to old R200 variants. This created confusion, because even though the Radeon 9000-9250 were in the same "thousand," they had a totally different feature set.
As you can see, the name game took over following the refresh, just like it does for ANY product set (including Nvidia's latest).
Cards (in performance order):
DirectX 8.1
Radeon 9000, 9000 Pro
DirectX 9
Radeon 9500, 9500 Pro, 9700, 9700 Pro
LATER REFRESH:
DirectX 8.1
Radeon 9100, 9200, 9250
DirectX 9
Radeon 9550 SE, 9550, 9600 SE, 9600, 9600 Pro, 9600 XT, 9800 SE, 9800, 9800 Pro, 9800 XT
Kinda confusing, ain't it?
To card manufacturers, we're sorry about the press leaks but you better match last generations top model with a mid price card or you're going to get destroyed eventually.
They DID. Or were you not paying attention when Nvidia released the 7600 GT 256MB for $200 (same price as the 6600 GT 128MB on release), and amazed the crowd with the performance potential of fast memory on a 128-bit bus?
Were you just incapacitated when review sites across the planet proclaimed the 7600 GT to stomp all over the 6800 GT, and even slightly outperform the 6800 Ultra? The only last-generation card with better performance is the x850 XT series, and it's not by much.
NOW, you have the 7900 GS released in the same price point as the 7600 GT was 6 months ago ($200). For that increase in price, you get 20 pipes instead of 12, and a 256-bit bus, GUARANTEED to best a 7600 GT by a huge margin. How can you NOT love this?
Maybe you find the number of models released confusing, but rest-assured, they are releasing better and better cards at the $150-200 price-point. On the Nvidia side you have the 7900 GS, and on the ATI side you have the x1900 GT, and they're both excellent cards.
I've purchased two dual-DVI cards from XFX: first my 6600 GT, and later my 7900 GT.
Both came with 2 DVI->VGA adaptors. Despite the "digital out" leanings of the cards, both had excellent 2D quality reminiscent of my old G400 MAX.
If you actually look at the link, you'll notice that the model there has a reference fan and heatsink. Both the listed specs and the picture mention this.
The passive model you are looking for is not sold on Newegg...in fact, I can't find it for sale anywhere. Tiger Direct no longer carries the item, and it's out-of-stock at Mwave. At $330, it has a sizeable price premium over other 7950 GTs.
See, that's the problem when buying from XFX - they have TOO MANY models, so you can easily get confused and buy the wrong one.
As Sadko posted, the fan mounts are standardized, so it is easy these days to purchase a "one size fits all" fan upgrade.
Furthermore, by upgrading to a better heatsink with a larger fan, one can run the fan at extremely low RPMs, making it virtually silent.
For example, I replaced the fan on my 7900 GT (REALLY loud) with a Zalman vf900. The heatpipe design makes this cooler incredibly efficient, and it also comes with a Fanmate to allow you to tune the speed to your tastes. I was able to run my card overclocked on the LOWEST Fanmate setting (inaudible), and still see a 10C reduction over the stock cooler.
Note that I have a fairly quiet PC case, based on a Sonata II with Scythe 120mm S-FLEX FDB fans, so you can rest assured that when I say "inaudible" over the rest of the case, this actually means something.
So, you get better cooling than any "passive" solution, and no noticeable increase in noise. That's why it may be a better idea to go with Zalman. I've heard of low-airflow, "quiet" or "silent" cases having issues with "passive" heatsink performance video cards (which REALLY require medium airflow to be cooled well), so I stick with low-noise active coolers. With the excessive premium you pay for passively-cooled video cards, the extra cash for a Zalman cooler isn't that much.
do real-world comparison tests. Run similar tasks over a period of time on machines built on the 2 architectures and compare the actual power demand.
Two reasons I can't:
1. I don't own a C7 system.
2. Regardless of how thoroughly I set up the test, there would be constant naysayers, and people with other ideas of how to erform the test. In other words, it wouldn't solve anything. Remember how it took almost a year for the computing community to realize just how much reduce the power consumption of the A64 90nm parts were over the 130nm A64s, not to mention Prescott? It mostly had to do with poor testing with incomplete information, plus conflicting results due to this.
So, let's do the next best thing, and take real-world numbers and try piecing them together. Unfortunately, there's a real dearth of actual REVIEWS. If you put C7 and review into Google, you actually get a lot of PREviews, but no solid benchmarking.
The only REVIEW I managed to find was this one, which shows the performance of the 1.5 GHz C7 versus a Sempron 3400+. The author notes that the cost of the Sempron plus board is the same as the cost of the Via system, so it's a good comparison.
The 256k cache plus the single channel memory combine to really hamper the Sempron's true performance (As a representative of the Athlon 64) in some benchmarks, but it still managed to come out shining in benchmarks that don't depend wholly on memory bandwidth and cache size. Thanks to the massive AMD price drop in July (after this review), you can now buy an Athlon 64 AM2 plus a 3200+ for that price, so I'd say pump up the benchmark results %10-20 across the board to get a feel for what you could buy for that money today.
Even with all the talk about low-power states and fast state switching, even though the Sempron is clocked %33 faster, the Sempron with the trusty old Cool 'n Quiet enabled managed to consume less power. How embarassing. If we can trust these test results, not only will the Sempron system finish processing ages earlier, it will also use less power.
Yes, AMD's pressure has pushed Intel to make a lot of changes for the better.
AMD's 386DX 40 Mhz pushed Intel to release faster 486 chips...otherwise Intel would have ridden their overpriced 486DX 33 forever.
AMD and Cyrix produced Pentium clones which pushed Intel and forced them to reduce prices.
AMD's push to revive Socket 7 (Super 7) with the introduction of the 100 MHz bus and the K6-2 forced Intel to release the Mendocino Celeron. With on-die cache, it was one of the best budget gaming processors ever released. The K6-2's "3D-Now!" instruction set forced Intel to finally admit that MMX was a failure, and release an addentum in 1999 with SSE.
AMD saved us from the hot, expensive, high-latency world that is Rambus. Without AMD pushing the industry-standard PC-133 SDRAM and later DDR SDRAM, Intel would have made Rambus the defacto desktop memory standard.
Fast-forward to today: AMD has been at parity or better for years, and it has slowed their improvements. After the cancelation of the K9 project, AMD sat on its ass, just releasing small improvements for K8. But now, with Conroe wiping the floor, AMD has to get back in gear, and is releasing K8L next year.
It's the way of the business. When you're on-top, you tend to stagnate.
While it is true that you can buy any chip you can imagine in a server, the original poster gave me the impression that he/she wanted a cheap solution with a simple chip-on-a-board (ala PegasOS). Unfortunately, the money is in complete systems tied in with services, so that's the last thing you'll catch IBM selling.
And sure, IBM's chips are popular in consoles, but that's mostly because IBM is the only major chip house that will offer to develop custom chip designs. The game console companies help fund the R&D bill and sign-on for some minimum initial shipment, and IBM makes some money. IBM takes existing core logic and pieces it together to optimize performance for consoles, and out pops a new chip design.
But there's a good reason IBM has time for game consoles: IBM doesn't have NEARLY the sales volume of AMD, let alone Intel. AMD doesn't have time for custom chips because they have enough trouble just meeting market demand for their x86 products. They are shooting for 100 million chips next year once their new fab ramps up, and Intel is shooting for 300 million. IBM, by contrast, will sell only 3-20 million chips per-console, per-year, depending on popularity, and those sales rise and fall periodically as consoles get introuced and mature. Even with all three consoles in their pockets, don't expect IBM to sell more then 30 million per-year.
I don't really see what the attraction is to small-market chips like CELL anyway. My prediction on the subject has already come true: way back when IBM announced the CELL, I predicted that by the time it was released, the PC market would have a similar-performance chip on-offer.
With Intel's release of the quad-core Core2 products later this year (Clovertown), my prediction becomes reality. With 2x128-bit SIMD units per core (for a total of 8 128-bit units per chip), Clovertown is capable of the same performance levels as Cell. AMD's K8L, due out next year, is expected to boast similar performance levels.
It's sad how few people realize their "efficient" Via CPU is not so efficient?
:D
Clock-for-clock, the optimized WinChip core (yes, even the C7 uses the very same core) can only process one integer and one floating-point instruction in parallel. This makes it 2-3x slower per-clock than modern CPUs. So, while you're still waiting on your Via C7 to crunch those numbers (at 20w), a Core2 Duo or A64 X2 system can do it in 1/4-1/6 the time (at 35w), and clock down to low-power state (3-5w).
So, I hope you feel good about how much carbon Via saves building the chip, because not only does it uses more power than competing processors to do the same amount of work, it takes longer too
Via's day in the sun is over. They were faced with the poor performance of the Winchip core, and instead of redesigning it, they touted the low power (which is true) and efficiency (which is not true). Intel and AMD responded with innovations like real-time voltage and frequency adjustment, and all of a sudden Via is scrambling just to try and keep up.