Mp3, as it turns out, is a lot easier to decode than wma and other later-generation formats. The fact that you have to use mpeg4 or wma with your DRMed purchase is just an unwanted side-effect.
That said, it is one reason I only play mp3s on my portable player. LAME has brought a level of quality to the mp3 format that none thought possible, and it keeps up suprisingly well with "more advanced" codecs. I see no reason to use anything else...it plays everywhere, and uses less battery life.
It's thanks to lovely things like ITAR that I can't even CALL CANADA to get tech support.
Like many defense companies, we make use of many Dy-4 products (now owned by Curtiss-Wright). The only problem? Dy-4 is a CANADIAN company.
So, in order to get tech support with Dy-4, I have to go through a specially-designated contact who has an export license for just this sort of thing. This wouldn't be a problem, except he is the only person with said export license, and has to serve all sorts of people.
I can't even EMAIL these people about a technical issue without someone holding my hand, even if I know it's not critical information.
Agreed. People who don't believe this need a touch of history.
Back in the 1980s, most TVs were much smaller than what they sell today, and much lower quality. You had a lucky few with big screens (greater than 30"), a significantly larger slice with screens in the 20-30" range, and the majority of the population with screens ranging from 12-20".
It's absolutely amazing how times change. My computer monitors even going back to the mid-1990s dwarf my family's old 14" TV, and have much better sharpness, bandwidth and color fidelity. You can buy 27" TVs at Walmart for less than 150 bucks that have visual quality most of us could only wish for in the 80s.
So, it's no surprise that VHS was considered good in the 1980s, when you consider that the average TV of the era was tiny and crappy. The 1990s started the wave of cheaper, larger, higher-fidelity televisions that showed the limitations of VHS, and by the end of the decade the market was hungry for DVD.
Unfortunately, the market hasn't changed much since DVDs were introduced. Screen sizes are up a bit, but HDTV technology is still too expensive for your average Walmart buyer. Even if HDTV suddenly dropped to the price range of your average Walmart buyer, you'd still have to wait 5 years or so for a potential market to build up.
HD-DVD and Bluray are in a very similar position to Laserdisc...not a huge improvement over the cheaper alternative, and selling to a very limited market. When only 5-10% of the market even owns equipment good enough to tell the difference, and they have to shell out $500+ to experience it, you've got limited room for growth.
* 16 stage pipeline (just like C7 Esther) * SSE support * Full-speed FPU
You see, the only real improvement here is that Nehemiah is getting a facelift. It is moving to a 0.09 micron SOI process (about 2 years behind AMD and 3 years behind Intel), is getting a new bus interface, double the L2 cache, and is getting SSE2 support. Other than that, it is the same old Nehemiah.
The bus and cache upgrade are the most important part. People have been stating for years that the bus speed of 133 MHz plus the pathetic 64K L2 cache has held the C3 back. All the SSE2 units in the world won't help you perform if you can't move enough data.
Think of the Pentium MMX. New process and dual voltage (2.8v) meant lower power than the previous Pentium (3.3v). Double the L1 cache and MMX support meant better real and potental performance. Now, imagine the Pentium MMX also included an upgrade to a 133 MHz bus, and you've got the equivilant of Esther. It should perform 30-40% better clock-for-clock over the C3, and should be more scalable, but it is nothing amazing.
Funny, my THREE PROCESSOR 12MB Creative 3D Blaster Voodoo2 was purely passive cooling - no fans, not even a heatsink. When did that come out... 1995?? Same with my Voodoo3 2000 PCI. ATi Rage/Rage Pro/RageIIC (digging them up as I dig thru my old hardware box here.) Same thing. Those are pretty old, as well. Matrox G400 - no fan or heatsink, either.
You need to learn how to troll properly. The Voodoo 2 chipset, consisting of one PixelFX2 and two TexelFX2 chips, was designed as a multi-chip solution for several reasons. Three independent chips meant three independent 64-bit memory busses instead of one shared 128-bit bus, reduced complexity per core meant higher yields, and the modular design allowed for many different configurations.
The side effect was distributed power consumption, which made it easier to use the chip package and PCB as its own heatsink.
Here's where you messed up: when 3dfx combined all the girth of a Voodoo 2 into a single chip, clocked it at 143MHz, and called it the Voodoo 3 2000, they added a huge passive heatsink. Suddenly, with the entire solution on a single chip, 3dfx couldn't just use the board and package for cooling...they needed a heatsink to extract the higher-density heat.
The Matrox G400 also shipped with a passive heatsink. Believe me, I own one.
The reasons manufacturers moved to active cooling is because the enthusiast market DIDN'T YET EXIST. Heatpipes and large custom heatsink assemblies were still VERY expensive, so most manufacturers went with simple, bulk passive heatsinks, or simple active-cooled heatsinks for high-end boards (like the G400 MAX, TNT2 Ultra and GeForce, for example). People back then balked at paying $300 for a video card, and most of the midrange sales were in the $100-150 range, so price was key. So, the industry went with loud fans on tiny heatsinks for high-end cards.
NOW, the enthusiast market is in full-swing, and the huge increase in power densities (and the need to remove said power) has improved prices on custom heatsinks and heatpipes. Now high-end buyers pay $500+ for a card, while even most midrange buyers shell out $150-300 for a card. Suddenly, price has a little room for play, and manufacturers can sell on features like passive or low-noise cooling. In fact, specialist companies have sprung up to make this even easier for manufacturers...they don't even have to design their own low-noise cooling solutions anymore.
So yes, your THREE PROCESSOR Voodoo 2 could be cooled passively...but so can a 7600 GT, which is and has tons more features. I'd call that impressive.
As of March 2006 here are your choices (using the newest technology available in each category, and cards are ranged in order of typical performance in their category):
Budget (ie: you really shouldn't spend this little):
GeForce 6200 256MB ($49)
Radeon x300 256MB ($55)
Upper budget (cards that will actually play new games):
Radeon x1300 Pro 256MB ($95)
GeForce 6600 256MB ($85)
Lower-midrange gamer:
GeForce 6600 GT 128MB ($120)
Radeon x1600 Pro 256MB ($125)
ATI has responded to the ultra-low prices of the 7900 series by bringing down prices across the board, something they should have done for a long time. Already in only 6 months of life, ATI has cut $100 or more off their upper-midrange (x1600 XT) and lower-hardcore (x1800 XL) parts, just scambling to keep up with Nvidia.
It might take a few weeks to filter through the channel, but those price cuts would make ATI a whole lot more competitive on the mid to high-end.
Just because you THINK there is some purpose in Nvidia's naming scheme doesn't mean there actually IS any purpose.
Take the GeForce 6 series, for example:
Within the first six months of release, Nvidia had laid-out a very simple set of cards (in performance order):
6800 Ultra 6800 GT 6800
6600 GT 6600
6200 6200 TC
Now, they had this great arrangement of performance levels, where all the cards within a lower numbered range were slower than the cards in the next higher numbered range. but like any company they had to deal with inefficiencies in their production processes, and try to keep their brands fresh. Thus, many cards were added to fit small but profitible niche or OEM markets.
So, by the end of 2005, you had a whole mess of cards. Some of them were added to compete with ATI, others were added to deal with yields (and had disabled pipes), while still others were introduced to replace a product that was "old" with something easier to make.
The mapping, in true performance, of all GeForce 6 chips, end of 2005:
6800 Ultra 6800 GT 6800 GS (Added as a reduced-cost replacement to 6800 GT) 6800 GTO (Added in response to ATI's x800 GTO) 6800 6600 GT 6800 XT 6800 LE 6600 DDR2 6600 6500 6600 LE 6200 6200 TC
See how confusing that became? It's just a natural progression, and ATI does the same thing. The 7 series is already beginning to see the effects of the naming scheme madness. Once Nvidia transitions fully over to the 7 series, expect the same product fragmentation to occur.
Oh, and I must correct you on this:
x extension (gtx, fx) was for a while PCI-X, but they've since dropped it.
Do you mean PCIe?
Incorrect. The FX series (GeForce 5) was entirely AGP. Board makers later released versions of the FX series with PCI-e bridge chips so they could dump their stock as "PCIe" cards.
The 7800 GTX is the only card EVER MADE by Nvidia to wear the "GTX" monkier. The 7800 GTX is PCIe, but so is the 7800 GT...see the problem with your assertion?
Yes, most reviews don't typically test maximum transfer speed on CONSUMER cards because it's not that important for CONSUMERs.
You are a professional. You are feeding your card texture and vertex data via a high-bandwidth storage system. Most consumers, on the other hand, are waiting on their 50MB/s hard disks as the choke point.
I have read one or two reviews that have tested the raw transfer rates of these PCIe 16x video cards, and most of them have topped out at the 900-1000 MB/s range (load). In other words, nowhere near the theoretical maximum, but certainly not limited by the onboard video bandwidth. The G70 and R500-series cards were at the top of the list, however.
If you want the best upload (to video card memory) performance you can possibly get, pick up one of the new NForce SLI boards with 2x 16x PCIe slots, and pick up a G70 or G71 card to take advantage of the second-generation PCIe controller.
You mean the 6800 GS, right? They both perform about the same.
The 7600 GT should be somewhat faster with no features turned on, both due to the faster core clock and the improvements in G71. With AA enabled, the performance should be closer, because the 7600 GT will be limited by its memory bandwidth.
You can't ignore the fact that the 7600 GT comes with support for transparency MSAA and SSAA, very nice in today's games that create multi-layered terrain effects using detail textures with transparency.
There is nothing "innovative" about Cell. Cell is basically multiple vector processors on a chip, and is a very predictable path for Sony after the release of the Playstation 2.
Each Cell SPE is simply a highly-optimized vector unit with 128-bit registers. It is capable of operating on 4 32-bit operands per cycle, just like SSE2/3 and Altivec. The difference is, eache SPE runs an independent task, while the Altivec / SSE units execute vector instructions in parellel with normal operations. However, the SPE is cut down: it has no branch prediction hardware or out-of-order execution, and depends on the main processor filling and emptying its Load/Store memory.
If you think the speed makes it innovative, think again: neither the Cell SPE nor its predecessor, the Emotion Engine, are IEEE754 compliant for 32-bit floating-point operations (for speed reasons). Cell can do IEEE754 compliant 64-bit floating point, but at an estimated speed hit of 10x, which makes it just "competitive" with existing solutions.
Sony / IBM actually inflate the performance numbers of the SPE, advertising it as 25.6 GFLOPS. But this doesn't take into account that the two pipelines of the SPE are NOT flexible, and can only perform certain types of instructions. The "Even" pipe can do arithmetic, and the "Odd" pipe can do Load / Store / Permute / Branch. Thus, the maximum arithmetic thoroughput per SPE is cut in half, to 12.8 billion arithmetic operations per second, and the double-percision performance is just 1.28 billion arithmetic instructions per second.
It's a nice idea for a media processor, but the complexity guarantees it will have a hard time finding buyers, and programmers for the Playstation 3 will be slow on optimization.
I mean, really, can you really break down a game into more than a few concurrent tasks without going crazy trying to synchronize it all? In addition, how many of those independent tasks can be designed with few or no branches? In your average code, branches make up about 20% of instructions. With an 18-cycle penalty per-branch, you'd have to keep those SPE branches under 1% of all instructions to avoid a serious performance loss.
So, get a job that doesn't require constant connectivity and 60 hour weeks. You might make a little less, but you'll be contented.
If you went the system admin route, you knew the responsibilities. 24/7 business means they need you available 24/7. It is cheaper for them to pay you time and a half (if you're that lucky) if something comes up than pay a second guy full-time to be there after hours.
Of course, if you're not a system admin, and your boss is deciding when you should work overtime, then you REALLY need a new job. I mean, if I do overtime, I do it WHEN I WANT TO. I coordinate with any people I need to work with in-advance, and we agree on a time. The boss should have zero say in this matter, so long as you deliver on-time.
I agree, if you buy the new hardware in advance, the game is guaranteed to be delayed.
Conversely, if you buy the hardware at the same time as the game is released, the hardware will take longer than normal to get working. Such was the case with my upgrade for Half-Life 2, bought the upgrade the day of release, and didn't get to play the game for a week and a half:D
And, as I have no dignity, I call this maxim Default Luser's Law (C) (R) (TM), and expect bountiful royalties every time it is invoked.
But, it's not like the N64 was the first time Nintendo had done a fairly dry launch. According to Wikipedia's Launch Titles page, the SNES had the same exact launch titles, plus F-Zero. Hell, the Genesis launched in the US with only 5 titles, and half of them were just arcade ports, and the Sega Saturn had only Virtua Fighter.
This is probably due to the fact that the console wars were in full-swing, and Sega and Nintendo figured getting the new console out was top priority. I mean, it took Sega a couple years to get their shit together and release Sonic, along with a slew of excellent first and third-party games. I don't recall the SNES being anything amazing until christmas 1991 and 1992, when we started seeing games like FF2 and Super Ghouls 'n Ghosts.
But console makers have learned, and have improved launch titles. Since the N64, not a single home console has launched with less than a dozen titles, not even the Gamecube. Now that most desireable games are no longer unique to one console, and over half of all games are sequels, they have to provide quantity to please early adopters. Unfortunately, even this is getting old and tired.
If the dozen and a half games from the Xbox 360 launch are not enough to spark interest, then we very well may see problems in this generation's transition.
The Cell is a new CPU and needs time to evolve a little.
No, the Cell is a games CPU that IBM is trying to hock as a general-purpose DSP. The only reason IBM is hyping it is because they already designed it for Sony. Any extra sales beyond Sony is pure profit.
And I do say this: Cell is a gaming processor. Like most gaming processors, it doesn't put much value on 64-bit floating-point, and instead it is very fast at 32-bit floating-point. For example, Gekko (Gamecube) can execute two 32-bit FP instructions or 1 64-bit FP instruction each cycle using the same hardware. The PS2 (Emotion Engine) doesn't support 64-bit FP. The VMX in Xenon (Xbox 360) is also limited to 32-bit precision. That's just the way the gaming world works, for 99% of the time, 32 bits is enough.
On the Intel platform, definitely. Thanks to their external memory controller, they already suffer from high latency. A little extra latency over DDR1 does not hurt them.
For AMD, it may yet be better. Keep in mind, these platforms being tested today are not yet optimized, and aren't using DDR2 800. I expect some performance improvement with the final product using DDR2 800. The key point is, even with DDR2 667, the release product should perform about the same as DDR1 400. This allows you to reap the power benefits of DDR2 (discussed below) without taking a performance hit.
For power:
No contest, DDR2 is miles ahead of DDR1. DDR2 runs on 1.8v versus 2.5v. As power usage runs with the square of the voltage, the reduction is (1 - 1.8^2 / 2.5^2) * 100 = 49%. That's a huge chunk of power.
In addition, DDR2 uses an internal buffer per memory cell double the size of DDR1 to get its increased data rates. This means the internal speed of DDR2 is half that of DDR1 for the same data rate (at the cost of more latency). Thus, even though power usage is proportional to frequency, DDR2 doesn't take a power hit over DDR1 400 until you move beyond DDR2 800.
Basically, AMD is moving to DDR2 because their Turions need lower-power memory to compete with Intel in notebooks. Turions are cherry-picked desktop chips that run at very low voltages, and they need their entire range of desktop processors to support DDR2 to allow them to keep doing this.
So:
* DDR2 800 should use 50% less power than DDR1 400. * DDR2 667 (most likely to be used on Turion platforms, much like Intel is using with Yonah) should use 60% less power than DDR1 400.
On copper, the speed is probably limited to somewhere in the 10 GHz range even for short drops, much lower if you have a parallel bus. But given serial interfaces like HyperTransport, you can just keep adding lanes (to a reasonable limit) to increase bandwidth.
When we reach the limit of copper serial busses, we branch out to optical serial busses, which have the potential to run as fast as hundreds of GHz. Probably won't see these for at least a decade.
There is no advantage over ML cores, and hell, we've known the Athlon 64 desktop can operate under 30w full-load since the release of the Winchester core. In fact, I bought a Winchester core when they were released because of the low power consumption.
The MT cores have a slight advantage over desktop processors, but that also comes at a hefty price. See my previous posts for more information.
Really, I think they started this review with the idea of providing a cheap and low-power upgrade path for Socket 754 users, but when you consider the price premium of Turions over desktop processors, and the incompatibility with older mainboards, this just isn't a good idea at all. I figure they just finished the review for the sake of finishing it.
Right, this "test" was pointless. Turion ML cores use the exact same voltage as desktop processors at load, and use about 1 watt less at idle (but for that, you also only get 800 MHz operation at idle, versus the desktop Athlon 64 at 1 GHz).
The lack of compatibility with older Socket 754 boards means this is of limited utility for upgraders. I mean, what's the point of "upgrading" when your board can't activate PowerNOW! on the Turion? It will use less power under load, but use more under idle than a Newcastle core with Cool 'N Quiet enabled.
I mean, sure, the Turion MT is a nice low-power platform, but this comes with a price. A Socket 939 Athlon 64 3200+ goes for around $140 online. The equivilant Turion MT goes for around $270, and it still can't be easily passively cooled (25w under load).
I mean, if you need a low-power, quiet server processor, just buy a desktop Athlon 64 and clock it down to 1 GHz with 1.1v core (4w idle, 10w load, can be easily cooled passively). All Athlon 64 desktop chips are certified to run at this speed and voltage, as it is a part of Cool 'n Quiet.
This already exists. Modern Nvidia cards reduce their core clcok speed and voltage when not doing 3D, and ATI cards reduce both their core and memory clock & voltages when not doing 3D.
For example, my 6600 GT runs at 500 MHz during 3D games, but it clocks down to 300 MHz and a reduced voltage in 2D operation. The power consumption is reduced by half in 2D.
These are technologies adapted from their mobile chips, although they're not as aggressive as the power saving modes of those mobile chips. I do wonder, however, what we are going to do once Vista starts stressing the 3D engine full-time.
What I don't understand with this Turion story is why AMD are not pushing desktop usage for this processor themselves.
They already are. Did you even read the review? Turion processors are low-voltage Athlon 64 processors. They run at 1.35v for the ML line at full speed, 1.2v for the MT line at full speed, and 0.9v at 800 MHz idle.
The "Newcastle" core they compared the Turion to is VERY OLD, 0.13 micron, with an operating voltage of 1.5v. The Turions are based on a second-generation, highly refined 0.09 micron process. So are all of AMD's current desktop processors.
Modern desktop Socket 939 Athlon 64 single-core processors use the Venice and San Diego cores, which are based on the same process as the Turions, and are VERY LOW POWER. Venice chips run at 1.35v, the same as the Turion ML, and with Cool 'n Quiet enabled, they idle at 1GHz with 1.1v, for a power usage of ~ 4w. So, expect equivilantly-clocked Athlon 64s to use the same power as Turion MLs.
I thought this review was a stupid waste of time, and here is why:
1. They originally stated they did the review because Socket 754 motherboards could be found cheap, and thus you could make a cheap, powerful and low-power box. But the MSI RS482M-IL they settled on sells for more in the $70-80 range. Socket 939 boards can be had for that.
2. The Turion ML is no lower-power than its desktop counterparts (except in idle, but the difference is so small it only matters to a notebook), but it has a price premium of about $80 for the same performance level. For example (from Pricewatch), the Socket 939 3200+ sells for around $140, and the Turion ML-40 (its performance equivilant) sells for around $230. The MT-40 has an even higher premium, costing an additional $40 over the ML-40.
Oh, and a few quick answers to your questions:
They can't just sell everyone Turion MTs. Those cores are cherry-picked for low-voltage operation, and they are in much shorter supply than the ML / desktop voltage chips.
And the single memory channel on the Turion was the obvious choice. Dual memory channels would require every Turion notebook to ship with two sticks of DDR1. As I have mentioned earlier, the DDR2 used on the Pentium M platform is a lot lower power than DDR1. Thus, with a single channel, AMD has encouraged manufacturers to use only a single stick. Later this year, when AMD moves to DDR2, expect to see dual-channel memory on the Turions.
Online game service fees will hardly make a dent in these kind of losses, assuming these loss estimates are correct.
Let's say that Sony does release their console at $500, at a $400 loss per console. Let's also assume that Sony will cut the sale price in half within 3 years ($250), and reduce the component costs within 3 years by 2/3 as they did with the Playstation 2 ($300).
Assuming robust sales of say, 50 million consoles, I'm eyeing the average loss per console over 5 years in the range of $200-300, yielding a total loss over five years of 10-15 billion dollars.
Now, assume Sony can get away with charging slightly more than Microsoft for their online service (say, 60 bucks a year), and that Sony can maintain an average (over those 5 years) of 10 million customers paying online fees. That yields 600 million gross a year, probably more in the range of 200-300 million net.
So, over 5 years, that's a paltry 1.5 billion (best-case) to be applied toward a huge 10-15 billion hole. Game licensing and sales fees will still have to make up the lion's share of those losses.
I've always considered FF1 to be a clunky game. Cheesy black backgrounds in dungeons and battle scenes because they didn't want to bother creating rich enviroments. Agonizingly slow battle system. Enemies with the same 16 colors because the colors of the party members locked them into a single pallette.
Phantasy Star was a much better game, and at least Dragon Warrior (from 2 on up) took greater advantage of the colors available to make their enemies more varied and vivid.
Sure, when they moved the game to the SNES, it looked brilliant, but the FF "format" looked clumsy on limited 8-bit consoles.
The problem was simple: as you started taking out the Z-95s, you simultaneously had to deal with wings of rogue T/A attacking you after they kill your wingmen.
The solution? Blast in toward the Victory Star Destroyer as fast as you can in the early part of the mission, take out as mnay T/A as you can, then break off when the cruiser appears with its load of Z-95s.
The best armament is Adv Torpedos, because you get 6 of them, and they can kill a Z95 or T/A in one shot. Your other alternative is to shoot down stray rockets with your Adv Missiles, but that never worked well for me.
And like I said, this mission was dumbed down for every re-release. The CD edition, and the Totally Games 3D-accelerated edition both have the easier mission.
I remember it like it was yesterday. You had to perform nothing short of a miracle in a Tie Advanced (without the tractor beam upgrade, so it was short on power).
You had to fly to one Star Destroyer, inspect some containers, kill some ships, then get over to the other end of the battle area to protect your Interdictor from about a dozen wings of Z-95s with heavy rockets. Just as you were taking out one wing, another would have already launched and started firing rockets. Keep in mind, this was on MEDIUM difficulty.
I must have flown that mission dozens of times before I actually beat it. It would have been easy with a modified Tie Advanced because of all the extra speed, or maybe if they had given me a 16 missile load instead of the usual 8, but with a regular T/A it was a pain chasing down those Z-95s and trying to take out a few of the rockets before they moved out of range.
Needless to say, in the CD version this mision was toned down considerably. And, not surprisingly, no other mission in the game came close to this level of difficulty.
Slashdot Mirror
Synthetic muscle that runs better when it's hot...
Damn, looks like the Clans have caputured our Triple-Strength Myomer technlolgy!
What's next, the C3 Computer?
Mp3, as it turns out, is a lot easier to decode than wma and other later-generation formats. The fact that you have to use mpeg4 or wma with your DRMed purchase is just an unwanted side-effect.
That said, it is one reason I only play mp3s on my portable player. LAME has brought a level of quality to the mp3 format that none thought possible, and it keeps up suprisingly well with "more advanced" codecs. I see no reason to use anything else...it plays everywhere, and uses less battery life.
It's thanks to lovely things like ITAR that I can't even CALL CANADA to get tech support.
Like many defense companies, we make use of many Dy-4 products (now owned by Curtiss-Wright). The only problem? Dy-4 is a CANADIAN company.
So, in order to get tech support with Dy-4, I have to go through a specially-designated contact who has an export license for just this sort of thing. This wouldn't be a problem, except he is the only person with said export license, and has to serve all sorts of people.
I can't even EMAIL these people about a technical issue without someone holding my hand, even if I know it's not critical information.
Agreed. People who don't believe this need a touch of history.
Back in the 1980s, most TVs were much smaller than what they sell today, and much lower quality. You had a lucky few with big screens (greater than 30"), a significantly larger slice with screens in the 20-30" range, and the majority of the population with screens ranging from 12-20".
It's absolutely amazing how times change. My computer monitors even going back to the mid-1990s dwarf my family's old 14" TV, and have much better sharpness, bandwidth and color fidelity. You can buy 27" TVs at Walmart for less than 150 bucks that have visual quality most of us could only wish for in the 80s.
So, it's no surprise that VHS was considered good in the 1980s, when you consider that the average TV of the era was tiny and crappy. The 1990s started the wave of cheaper, larger, higher-fidelity televisions that showed the limitations of VHS, and by the end of the decade the market was hungry for DVD.
Unfortunately, the market hasn't changed much since DVDs were introduced. Screen sizes are up a bit, but HDTV technology is still too expensive for your average Walmart buyer. Even if HDTV suddenly dropped to the price range of your average Walmart buyer, you'd still have to wait 5 years or so for a potential market to build up.
HD-DVD and Bluray are in a very similar position to Laserdisc...not a huge improvement over the cheaper alternative, and selling to a very limited market. When only 5-10% of the market even owns equipment good enough to tell the difference, and they have to shell out $500+ to experience it, you've got limited room for growth.
Via is the king of "rebranding" existing technologies.
Nehemiah C3 has:
* 16 stage pipeline (just like C7 Esther)
* SSE support
* Full-speed FPU
You see, the only real improvement here is that Nehemiah is getting a facelift. It is moving to a 0.09 micron SOI process (about 2 years behind AMD and 3 years behind Intel), is getting a new bus interface, double the L2 cache, and is getting SSE2 support. Other than that, it is the same old Nehemiah.
The bus and cache upgrade are the most important part. People have been stating for years that the bus speed of 133 MHz plus the pathetic 64K L2 cache has held the C3 back. All the SSE2 units in the world won't help you perform if you can't move enough data.
Think of the Pentium MMX. New process and dual voltage (2.8v) meant lower power than the previous Pentium (3.3v). Double the L1 cache and MMX support meant better real and potental performance. Now, imagine the Pentium MMX also included an upgrade to a 133 MHz bus, and you've got the equivilant of Esther. It should perform 30-40% better clock-for-clock over the C3, and should be more scalable, but it is nothing amazing.
Funny, my THREE PROCESSOR 12MB Creative 3D Blaster Voodoo2 was purely passive cooling - no fans, not even a heatsink. When did that come out... 1995?? Same with my Voodoo3 2000 PCI. ATi Rage/Rage Pro/RageIIC (digging them up as I dig thru my old hardware box here.) Same thing. Those are pretty old, as well. Matrox G400 - no fan or heatsink, either.
You need to learn how to troll properly. The Voodoo 2 chipset, consisting of one PixelFX2 and two TexelFX2 chips, was designed as a multi-chip solution for several reasons. Three independent chips meant three independent 64-bit memory busses instead of one shared 128-bit bus, reduced complexity per core meant higher yields, and the modular design allowed for many different configurations.
The side effect was distributed power consumption, which made it easier to use the chip package and PCB as its own heatsink.
Here's where you messed up: when 3dfx combined all the girth of a Voodoo 2 into a single chip, clocked it at 143MHz, and called it the Voodoo 3 2000, they added a huge passive heatsink. Suddenly, with the entire solution on a single chip, 3dfx couldn't just use the board and package for cooling...they needed a heatsink to extract the higher-density heat.
The Matrox G400 also shipped with a passive heatsink. Believe me, I own one.
The reasons manufacturers moved to active cooling is because the enthusiast market DIDN'T YET EXIST. Heatpipes and large custom heatsink assemblies were still VERY expensive, so most manufacturers went with simple, bulk passive heatsinks, or simple active-cooled heatsinks for high-end boards (like the G400 MAX, TNT2 Ultra and GeForce, for example). People back then balked at paying $300 for a video card, and most of the midrange sales were in the $100-150 range, so price was key. So, the industry went with loud fans on tiny heatsinks for high-end cards.
NOW, the enthusiast market is in full-swing, and the huge increase in power densities (and the need to remove said power) has improved prices on custom heatsinks and heatpipes. Now high-end buyers pay $500+ for a card, while even most midrange buyers shell out $150-300 for a card. Suddenly, price has a little room for play, and manufacturers can sell on features like passive or low-noise cooling. In fact, specialist companies have sprung up to make this even easier for manufacturers...they don't even have to design their own low-noise cooling solutions anymore.
So yes, your THREE PROCESSOR Voodoo 2 could be cooled passively...but so can a 7600 GT, which is and has tons more features. I'd call that impressive.
You got it.
As of March 2006 here are your choices (using the newest technology available in each category, and cards are ranged in order of typical performance in their category):
Budget (ie: you really shouldn't spend this little):
GeForce 6200 256MB ($49)
Radeon x300 256MB ($55)
Upper budget (cards that will actually play new games):
Radeon x1300 Pro 256MB ($95)
GeForce 6600 256MB ($85)
Lower-midrange gamer:
GeForce 6600 GT 128MB ($120)
Radeon x1600 Pro 256MB ($125)
Upper-midrange gamer:
GeForce 6800 GS 256MB ($185)
Radeon x1600 XT 256MB ($200)
The 7600 GT 256MB card should be in this price range,
and should have slightly better performance than the 6800 GS.
Lower-hardcore gamer:
Radeon x1800 XL ($350)
GeForce 7800 GT ($285)
The 7900 GT should be in this price range,
and if it is it will be the fastest card under $300
Upper-hardcore gamer:
Radeon x1900 XT 512MB ($496)
Radeon x1800 XT 512MB ($479)
GeForce 7800 GTX 256MB ($415)
The 7900 GTX should be in this price range
Crazy gamer:
Radeon x1900 XTX 512MB ($575)
ATI has responded to the ultra-low prices of the 7900 series by bringing down prices across the board, something they should have done for a long time. Already in only 6 months of life, ATI has cut $100 or more off their upper-midrange (x1600 XT) and lower-hardcore (x1800 XL) parts, just scambling to keep up with Nvidia.
It might take a few weeks to filter through the channel, but those price cuts would make ATI a whole lot more competitive on the mid to high-end.
Just because you THINK there is some purpose in Nvidia's naming scheme doesn't mean there actually IS any purpose.
Take the GeForce 6 series, for example:
Within the first six months of release, Nvidia had laid-out a very simple set of cards (in performance order):
6800 Ultra
6800 GT
6800
6600 GT
6600
6200
6200 TC
Now, they had this great arrangement of performance levels, where all the cards within a lower numbered range were slower than the cards in the next higher numbered range. but like any company they had to deal with inefficiencies in their production processes, and try to keep their brands fresh. Thus, many cards were added to fit small but profitible niche or OEM markets.
So, by the end of 2005, you had a whole mess of cards. Some of them were added to compete with ATI, others were added to deal with yields (and had disabled pipes), while still others were introduced to replace a product that was "old" with something easier to make.
The mapping, in true performance, of all GeForce 6 chips, end of 2005:
6800 Ultra
6800 GT
6800 GS (Added as a reduced-cost replacement to 6800 GT)
6800 GTO (Added in response to ATI's x800 GTO)
6800
6600 GT
6800 XT
6800 LE
6600 DDR2
6600
6500
6600 LE
6200
6200 TC
See how confusing that became? It's just a natural progression, and ATI does the same thing. The 7 series is already beginning to see the effects of the naming scheme madness. Once Nvidia transitions fully over to the 7 series, expect the same product fragmentation to occur.
Oh, and I must correct you on this:
x extension (gtx, fx) was for a while PCI-X, but they've since dropped it.
Do you mean PCIe?
Incorrect. The FX series (GeForce 5) was entirely AGP. Board makers later released versions of the FX series with PCI-e bridge chips so they could dump their stock as "PCIe" cards.
The 7800 GTX is the only card EVER MADE by Nvidia to wear the "GTX" monkier. The 7800 GTX is PCIe, but so is the 7800 GT...see the problem with your assertion?
Yes, most reviews don't typically test maximum transfer speed on CONSUMER cards because it's not that important for CONSUMERs.
You are a professional. You are feeding your card texture and vertex data via a high-bandwidth storage system. Most consumers, on the other hand, are waiting on their 50MB/s hard disks as the choke point.
I have read one or two reviews that have tested the raw transfer rates of these PCIe 16x video cards, and most of them have topped out at the 900-1000 MB/s range (load). In other words, nowhere near the theoretical maximum, but certainly not limited by the onboard video bandwidth. The G70 and R500-series cards were at the top of the list, however.
If you want the best upload (to video card memory) performance you can possibly get, pick up one of the new NForce SLI boards with 2x 16x PCIe slots, and pick up a G70 or G71 card to take advantage of the second-generation PCIe controller.
You mean the 6800 GS, right? They both perform about the same.
The 7600 GT should be somewhat faster with no features turned on, both due to the faster core clock and the improvements in G71. With AA enabled, the performance should be closer, because the 7600 GT will be limited by its memory bandwidth.
You can't ignore the fact that the 7600 GT comes with support for transparency MSAA and SSAA, very nice in today's games that create multi-layered terrain effects using detail textures with transparency.
There is nothing "innovative" about Cell. Cell is basically multiple vector processors on a chip, and is a very predictable path for Sony after the release of the Playstation 2.
/Store memory.
Each Cell SPE is simply a highly-optimized vector unit with 128-bit registers. It is capable of operating on 4 32-bit operands per cycle, just like SSE2/3 and Altivec. The difference is, eache SPE runs an independent task, while the Altivec / SSE units execute vector instructions in parellel with normal operations. However, the SPE is cut down: it has no branch prediction hardware or out-of-order execution, and depends on the main processor filling and emptying its Load
If you think the speed makes it innovative, think again: neither the Cell SPE nor its predecessor, the Emotion Engine, are IEEE754 compliant for 32-bit floating-point operations (for speed reasons). Cell can do IEEE754 compliant 64-bit floating point, but at an estimated speed hit of 10x, which makes it just "competitive" with existing solutions.
Sony / IBM actually inflate the performance numbers of the SPE, advertising it as 25.6 GFLOPS. But this doesn't take into account that the two pipelines of the SPE are NOT flexible, and can only perform certain types of instructions. The "Even" pipe can do arithmetic, and the "Odd" pipe can do Load / Store / Permute / Branch. Thus, the maximum arithmetic thoroughput per SPE is cut in half, to 12.8 billion arithmetic operations per second, and the double-percision performance is just 1.28 billion arithmetic instructions per second.
It's a nice idea for a media processor, but the complexity guarantees it will have a hard time finding buyers, and programmers for the Playstation 3 will be slow on optimization.
I mean, really, can you really break down a game into more than a few concurrent tasks without going crazy trying to synchronize it all? In addition, how many of those independent tasks can be designed with few or no branches? In your average code, branches make up about 20% of instructions. With an 18-cycle penalty per-branch, you'd have to keep those SPE branches under 1% of all instructions to avoid a serious performance loss.
So, get a job that doesn't require constant connectivity and 60 hour weeks. You might make a little less, but you'll be contented.
If you went the system admin route, you knew the responsibilities. 24/7 business means they need you available 24/7. It is cheaper for them to pay you time and a half (if you're that lucky) if something comes up than pay a second guy full-time to be there after hours.
Of course, if you're not a system admin, and your boss is deciding when you should work overtime, then you REALLY need a new job. I mean, if I do overtime, I do it WHEN I WANT TO. I coordinate with any people I need to work with in-advance, and we agree on a time. The boss should have zero say in this matter, so long as you deliver on-time.
I agree, if you buy the new hardware in advance, the game is guaranteed to be delayed.
:D
Conversely, if you buy the hardware at the same time as the game is released, the hardware will take longer than normal to get working. Such was the case with my upgrade for Half-Life 2, bought the upgrade the day of release, and didn't get to play the game for a week and a half
And, as I have no dignity, I call this maxim Default Luser's Law (C) (R) (TM), and expect bountiful royalties every time it is invoked.
But, it's not like the N64 was the first time Nintendo had done a fairly dry launch. According to Wikipedia's Launch Titles page, the SNES had the same exact launch titles, plus F-Zero. Hell, the Genesis launched in the US with only 5 titles, and half of them were just arcade ports, and the Sega Saturn had only Virtua Fighter.
This is probably due to the fact that the console wars were in full-swing, and Sega and Nintendo figured getting the new console out was top priority. I mean, it took Sega a couple years to get their shit together and release Sonic, along with a slew of excellent first and third-party games. I don't recall the SNES being anything amazing until christmas 1991 and 1992, when we started seeing games like FF2 and Super Ghouls 'n Ghosts.
But console makers have learned, and have improved launch titles. Since the N64, not a single home console has launched with less than a dozen titles, not even the Gamecube. Now that most desireable games are no longer unique to one console, and over half of all games are sequels, they have to provide quantity to please early adopters. Unfortunately, even this is getting old and tired.
If the dozen and a half games from the Xbox 360 launch are not enough to spark interest, then we very well may see problems in this generation's transition.
The Cell is a new CPU and needs time to evolve a little.
No, the Cell is a games CPU that IBM is trying to hock as a general-purpose DSP. The only reason IBM is hyping it is because they already designed it for Sony. Any extra sales beyond Sony is pure profit.
And I do say this: Cell is a gaming processor. Like most gaming processors, it doesn't put much value on 64-bit floating-point, and instead it is very fast at 32-bit floating-point. For example, Gekko (Gamecube) can execute two 32-bit FP instructions or 1 64-bit FP instruction each cycle using the same hardware. The PS2 (Emotion Engine) doesn't support 64-bit FP. The VMX in Xenon (Xbox 360) is also limited to 32-bit precision. That's just the way the gaming world works, for 99% of the time, 32 bits is enough.
Is DDR2 really any better?
For performance:
On the Intel platform, definitely. Thanks to their external memory controller, they already suffer from high latency. A little extra latency over DDR1 does not hurt them.
For AMD, it may yet be better. Keep in mind, these platforms being tested today are not yet optimized, and aren't using DDR2 800. I expect some performance improvement with the final product using DDR2 800. The key point is, even with DDR2 667, the release product should perform about the same as DDR1 400. This allows you to reap the power benefits of DDR2 (discussed below) without taking a performance hit.
For power:
No contest, DDR2 is miles ahead of DDR1. DDR2 runs on 1.8v versus 2.5v. As power usage runs with the square of the voltage, the reduction is (1 - 1.8^2 / 2.5^2) * 100 = 49%. That's a huge chunk of power.
In addition, DDR2 uses an internal buffer per memory cell double the size of DDR1 to get its increased data rates. This means the internal speed of DDR2 is half that of DDR1 for the same data rate (at the cost of more latency). Thus, even though power usage is proportional to frequency, DDR2 doesn't take a power hit over DDR1 400 until you move beyond DDR2 800.
Basically, AMD is moving to DDR2 because their Turions need lower-power memory to compete with Intel in notebooks. Turions are cherry-picked desktop chips that run at very low voltages, and they need their entire range of desktop processors to support DDR2 to allow them to keep doing this.
So:
* DDR2 800 should use 50% less power than DDR1 400.
* DDR2 667 (most likely to be used on Turion platforms, much like Intel is using with Yonah) should use 60% less power than DDR1 400.
On copper, the speed is probably limited to somewhere in the 10 GHz range even for short drops, much lower if you have a parallel bus. But given serial interfaces like HyperTransport, you can just keep adding lanes (to a reasonable limit) to increase bandwidth.
When we reach the limit of copper serial busses, we branch out to optical serial busses, which have the potential to run as fast as hundreds of GHz. Probably won't see these for at least a decade.
There is no advantage over ML cores, and hell, we've known the Athlon 64 desktop can operate under 30w full-load since the release of the Winchester core. In fact, I bought a Winchester core when they were released because of the low power consumption.
Here is an early test of the Winchester's DC power consumption. Note that these cores have even lower power consumption than the Turions tested because they use a smaller cache (512k versus 1MB).
The MT cores have a slight advantage over desktop processors, but that also comes at a hefty price. See my previous posts for more information.
Really, I think they started this review with the idea of providing a cheap and low-power upgrade path for Socket 754 users, but when you consider the price premium of Turions over desktop processors, and the incompatibility with older mainboards, this just isn't a good idea at all. I figure they just finished the review for the sake of finishing it.
Right, this "test" was pointless. Turion ML cores use the exact same voltage as desktop processors at load, and use about 1 watt less at idle (but for that, you also only get 800 MHz operation at idle, versus the desktop Athlon 64 at 1 GHz).
The lack of compatibility with older Socket 754 boards means this is of limited utility for upgraders. I mean, what's the point of "upgrading" when your board can't activate PowerNOW! on the Turion? It will use less power under load, but use more under idle than a Newcastle core with Cool 'N Quiet enabled.
I mean, sure, the Turion MT is a nice low-power platform, but this comes with a price. A Socket 939 Athlon 64 3200+ goes for around $140 online. The equivilant Turion MT goes for around $270, and it still can't be easily passively cooled (25w under load).
I mean, if you need a low-power, quiet server processor, just buy a desktop Athlon 64 and clock it down to 1 GHz with 1.1v core (4w idle, 10w load, can be easily cooled passively). All Athlon 64 desktop chips are certified to run at this speed and voltage, as it is a part of Cool 'n Quiet.
This already exists. Modern Nvidia cards reduce their core clcok speed and voltage when not doing 3D, and ATI cards reduce both their core and memory clock & voltages when not doing 3D.
For example, my 6600 GT runs at 500 MHz during 3D games, but it clocks down to 300 MHz and a reduced voltage in 2D operation. The power consumption is reduced by half in 2D.
These are technologies adapted from their mobile chips, although they're not as aggressive as the power saving modes of those mobile chips. I do wonder, however, what we are going to do once Vista starts stressing the 3D engine full-time.
What I don't understand with this Turion story is why AMD are not pushing desktop usage for this processor themselves.
They already are. Did you even read the review? Turion processors are low-voltage Athlon 64 processors. They run at 1.35v for the ML line at full speed, 1.2v for the MT line at full speed, and 0.9v at 800 MHz idle.
The "Newcastle" core they compared the Turion to is VERY OLD, 0.13 micron, with an operating voltage of 1.5v. The Turions are based on a second-generation, highly refined 0.09 micron process. So are all of AMD's current desktop processors.
Modern desktop Socket 939 Athlon 64 single-core processors use the Venice and San Diego cores, which are based on the same process as the Turions, and are VERY LOW POWER. Venice chips run at 1.35v, the same as the Turion ML, and with Cool 'n Quiet enabled, they idle at 1GHz with 1.1v, for a power usage of ~ 4w. So, expect equivilantly-clocked Athlon 64s to use the same power as Turion MLs.
I thought this review was a stupid waste of time, and here is why:
1. They originally stated they did the review because Socket 754 motherboards could be found cheap, and thus you could make a cheap, powerful and low-power box. But the MSI RS482M-IL they settled on sells for more in the $70-80 range. Socket 939 boards can be had for that.
2. The Turion ML is no lower-power than its desktop counterparts (except in idle, but the difference is so small it only matters to a notebook), but it has a price premium of about $80 for the same performance level. For example (from Pricewatch), the Socket 939 3200+ sells for around $140, and the Turion ML-40 (its performance equivilant) sells for around $230. The MT-40 has an even higher premium, costing an additional $40 over the ML-40.
Oh, and a few quick answers to your questions:
They can't just sell everyone Turion MTs. Those cores are cherry-picked for low-voltage operation, and they are in much shorter supply than the ML / desktop voltage chips.
And the single memory channel on the Turion was the obvious choice. Dual memory channels would require every Turion notebook to ship with two sticks of DDR1. As I have mentioned earlier, the DDR2 used on the Pentium M platform is a lot lower power than DDR1. Thus, with a single channel, AMD has encouraged manufacturers to use only a single stick. Later this year, when AMD moves to DDR2, expect to see dual-channel memory on the Turions.
I can't believe the above was modded Interesting.
Online game service fees will hardly make a dent in these kind of losses, assuming these loss estimates are correct.
Let's say that Sony does release their console at $500, at a $400 loss per console. Let's also assume that Sony will cut the sale price in half within 3 years ($250), and reduce the component costs within 3 years by 2/3 as they did with the Playstation 2 ($300).
Assuming robust sales of say, 50 million consoles, I'm eyeing the average loss per console over 5 years in the range of $200-300, yielding a total loss over five years of 10-15 billion dollars.
Now, assume Sony can get away with charging slightly more than Microsoft for their online service (say, 60 bucks a year), and that Sony can maintain an average (over those 5 years) of 10 million customers paying online fees. That yields 600 million gross a year, probably more in the range of 200-300 million net.
So, over 5 years, that's a paltry 1.5 billion (best-case) to be applied toward a huge 10-15 billion hole. Game licensing and sales fees will still have to make up the lion's share of those losses.
I've always considered FF1 to be a clunky game. Cheesy black backgrounds in dungeons and battle scenes because they didn't want to bother creating rich enviroments. Agonizingly slow battle system. Enemies with the same 16 colors because the colors of the party members locked them into a single pallette.
Phantasy Star was a much better game, and at least Dragon Warrior (from 2 on up) took greater advantage of the colors available to make their enemies more varied and vivid.
Sure, when they moved the game to the SNES, it looked brilliant, but the FF "format" looked clumsy on limited 8-bit consoles.
The problem was simple: as you started taking out the Z-95s, you simultaneously had to deal with wings of rogue T/A attacking you after they kill your wingmen.
The solution? Blast in toward the Victory Star Destroyer as fast as you can in the early part of the mission, take out as mnay T/A as you can, then break off when the cruiser appears with its load of Z-95s.
The best armament is Adv Torpedos, because you get 6 of them, and they can kill a Z95 or T/A in one shot. Your other alternative is to shoot down stray rockets with your Adv Missiles, but that never worked well for me.
And like I said, this mission was dumbed down for every re-release. The CD edition, and the Totally Games 3D-accelerated edition both have the easier mission.
I remember it like it was yesterday. You had to perform nothing short of a miracle in a Tie Advanced (without the tractor beam upgrade, so it was short on power).
You had to fly to one Star Destroyer, inspect some containers, kill some ships, then get over to the other end of the battle area to protect your Interdictor from about a dozen wings of Z-95s with heavy rockets. Just as you were taking out one wing, another would have already launched and started firing rockets. Keep in mind, this was on MEDIUM difficulty.
I must have flown that mission dozens of times before I actually beat it. It would have been easy with a modified Tie Advanced because of all the extra speed, or maybe if they had given me a 16 missile load instead of the usual 8, but with a regular T/A it was a pain chasing down those Z-95s and trying to take out a few of the rockets before they moved out of range.
Needless to say, in the CD version this mision was toned down considerably. And, not surprisingly, no other mission in the game came close to this level of difficulty.