It's Official — AMD Will Retire the ATI Brand 324
Posted
by
timothy
from the rose-by-any-other-name dept.
from the rose-by-any-other-name dept.
J. Dzhugashvili writes "A little over four years have passed since AMD purchased ATI. In May of last year, AMD took the remains of the Canadian graphics company and melded them into a monolithic products group, which combined processors, graphics, and platforms. Now, AMD is about to take the next step: kill the ATI brand altogether. The company has officially announced the move, saying it plans to label its next generation of graphics cards 'AMD Radeon' and 'AMD FirePro,' with new logos to match. The move has a lot to do with the incoming arrival of products like Ontario and Llano, which will combine AMD processing and graphics in single slabs of silicon."
Re:Wil this affect open source drivers (Score:5, Informative)
ATI really only started doing that after they were acquired by AMD so I wouldn't worry too much.
Re:Great news (Score:5, Informative)
With a 3 GHz clock, a signal at the speed of light travels 10 cm during one clock cycle. This means that if a chip needs data from another and there's a distance of five centimeters or more between both chips the data will not arrive in the same clock cycle.
Re:Great news (Score:3, Informative)
moving the gpu on-die will fix the latency associated with the pci-e bus, but it's not because of the reasons you seem to believe
Re:Let The Confustion Begin (Score:3, Informative)
Re:That's retarded. (Score:4, Informative)
Classic example of not reading the article... (Score:5, Informative)
It's not light speed (Score:4, Informative)
Re:Great news (Score:5, Informative)
Yeah, 3ghz doesn't come close to the light speed barrier. i think the issue is more from heat dissipation and electron bleed...
moving the gpu on-die will fix the latency associated with the pci-e bus, but it's not because of the reasons you seem to believe
Want to bet?
At 3 GHz, light moves just 7.2 cm [wolframalpha.com], given a typical upper range for the velocity factor of copper of 0.72. Silicon and fibre optics are usually worse, with a VF between 0.4 and 0.6, or between 4 and 6cm per clock. That's barely enough to traverse a CPU die, let alone the motherboard. Moving parts physically closer together has a lot to do with the speed of light!
Re:Great news (Score:1, Informative)
an unshielded copper conductor it is about 96% of the speed of light. [wikipedia.org]
Re:Great news (Score:5, Informative)
Assuming the signals travel in a straight line. If you look at current motherboards and video cards, you'll notice that many of the copper traces are "wiggly", not straight. That is done in order to get bits in parallel buses to arrive at the same time, and conductor traces on the chips must be designed similarly, it's the longest distance that any of the bits must travel that limits the others.
Besides, there are capacitance and inductance effects to be considered. Transitions from one to zero and vice-versa aren't instantaneous and that must be taken into account.
One could say that 9 GHz would be the absolute physical limit for a 1.7 cm chip and the technical limit is somewhat lower than that.
For a set of chips on a board, the absolute physical limit is much lower, and that's the reason why on-chip cache memory has become so important lately.
Re:That's fine and dandy (Score:1, Informative)
nVidia's drivers suck pretty bad too. The real problem is that the high-end graphics card companies will prioritize "getting a couple of extra FPS in a benchmark" over "not crashing all the goddamn time"
At least the Linux open-source drivers tend to be stable, when a card finally gets supported (a generation late, at least).
Re:Great news (Score:2, Informative)
In each case, its always about the heat.
Pretty much all CPU's sold today (even "2.x ghz" chips) can go over 4ghz with proper air cooling. The reason they dont sell 4ghz+ chips is because chips have warranties and require a proper cooling setup in order to not fail at those speeds. Most important of course is heat sink and cpu fan which Intel and AMD do have some control over, but also of considerable importance is case fans and case ventilation, which they do not have control over.
Just moving my case fan from the stock front position (intake) to the back (exhaust) gave me 10 degrees C more headroom at load, allowing my AMD 1055T to go from 2.8ghz to 4.1ghz (before moving the case fan, I was only stable up to 3.36ghz)
Re:Great news (Score:3, Informative)
An XFI-SFI interconnect runs up to 10.3 Gbps on a single serial link. It is double-pumped (bit on each end of the clock) so the clock rate is half that. This is the connection that links a 10Gbps phy to the transceiver module. You do have to keep the interconnects pretty short though.
http://www.altera.com/technology/high_speed/protocols/10gb-ethernet-xfi-sfi/pro-xfi-sfi.html [altera.com]
XDR ram can transmit 8 bits per clock on a serial line: http://en.wikipedia.org/wiki/XDR_DRAM [wikipedia.org]
Re:Great news (Score:5, Informative)
Reading comprehension fail. Nobody said that you can't go above 3 GHz for the CPU, but that if you do, if a chip needs data from another and there's a distance of five centimeters or more between both chips the data will not arrive in the same clock cycle [slashdot.org]
Re:Opportunity knocking for AMD here... (Score:3, Informative)
Intel's CPUs and integrated graphics have long had great support in the Linux kernel. Because Intel controls the tech, they can actually provide the correct and full source for the graphics drivers. The problem is that Intel integrated graphics aren't ever anything special.
Methinks you might are being a bit generous with Intel. I went with an Intel integrated chipset a number of years back because the alternatives weren't very well supported on FreeBSD, but the graphics weren't just not special, they were bad. Sufficiently bad that I've stayed away from them ever since. Which for Intel is just dumb, I have a very hard time believing that Intel couldn't do any better than what they've been doing. Hopefully with AMD owning ATI that'll kick a bit of sand in Intel's collective face so that they actually do something about it.
Re:Justice Department on vacation since 1980 (Score:3, Informative)
Which manufacturing workers exactly?!
ATI does not have a plant. It's all TMSC and the other one I forgot how's it called.
Re:Great news (Score:5, Informative)
Want to bet?
At 3 GHz, light moves just 7.2 cm, given a typical upper range for the velocity factor of copper of 0.72. Silicon and fibre optics are usually worse, with a VF between 0.4 and 0.6, or between 4 and 6cm per clock. That's barely enough to traverse a CPU die, let alone the motherboard. Moving parts physically closer together has a lot to do with the speed of light!
I really would mod this informative, since I was about to make a similar point. I think a lot of the confusion is that people hear things like the Speed of Light in terms of Kilometers per second, and it gets filed away by the brain as inconsequential for scales which are measured in centimeters and MUCH smaller.
But when you realize that that scale which is only a factor measured in millions meters per second is being divided into segments that are fractions of billionths of a second, the speed of light manifests in a much more physically understandable term.
Re:Not terribly surprising (Score:3, Informative)
ATI graphics cards work just fine with Intel processors. I don't believe there's any move to stop them doing so when they rebrand.
Re:Wil this affect open source drivers (Score:3, Informative)
Other way around; AMD has always released specs and started releasing ATI specs after ATI was acquired. You may notice that http://www.x.org/docs/AMD/ [x.org] is lacking docs for the r200 and earlier; that's because AMD made the acquisition during the r400 era, and the docs for older chipsets were more or less lost forever at that point.
Right now, the open-source drivers are called radeon, r300, r600, etc.; one developer committed his code as "amd" instead at one point. (It got changed to avoid end-user confusion.)
Re:fglrx (Score:5, Informative)
fglrx support for r500 and earlier (anything before the HD lines) is already delegated to the open-source drivers. We're working on getting r800 (redwood) support for acceleration together, and r600 support is getting better by the day.
Re:WTF how is this offtopic? (Score:2, Informative)
To counter your posts: I've never had a single driver problem or failure with ATI cards, even in Linux (several distributions across several major release revisions). To top it, they have always had superior image quality compared to Nvidia, regardless of the back-and-forth of performance lead.
Re:Justice Department on vacation since 1980 (Score:3, Informative)
Four companies became three. ATI, Nvidia, Intel, AMD. AMD bought ATI.
Intel has been trying to buy Nvidia for years, saying that they need to merge in order to "compete". Nvidia resists, but it'll happen eventually.
Re:WTF how is this offtopic? (Score:4, Informative)
Re:Great news (Score:3, Informative)
Re:Justice Department on vacation since 1980 (Score:3, Informative)
Sure, that's why air fares doubled in many major routes last year. And foreign carriers raised prices even more because of the mergers going on over there.
And maybe you didn't hear now that American wants to buy Southwest so they can "compete" with the new United merger. If Southwest is out of the market, how much do you think fares will go up? The airlines need more regulation. We already have a situation where you can fly coast to coast for $100 but it costs $800 to go 300 miles.
If you've ever done business in a city where you have to fly through Cleveland, say, Akron, you know that when you fly there you get fleeced.
We're down to 3 carriers. The "regional players" (which are not really regional) are getting snapped up.
It's not only the consolidation, which is terrible for the US economy and for consumers, it's also the fact that there are these huge bidding wars where companies use their war chests to pay outrageous sums for failing companies, engaging in corporate pissing contests where the money never sees the rest of the economy, just going from one mattress to another, instead of paying their shareholders dividends.
Further, everytime there's one of these mergers, another few thousand American workers get thrown off the back of the gravy train so the CEO and chairman can get another $100 million in bonuses.
Re:It's not light speed (Score:4, Informative)
If the propagation speed of an electrical signal is .96C in an uninsulated chunk of copper and only .66C in a coaxial cable, what is it reduced to in an on-chip environment? On a computer bus? I have seen the figure .33C, but I can't find any primary source for this.
Let's assume the 0.33C for the moment, and consider what this means. A CPU contains some fairly large functional units that need to be run synchronously - meaning that all transistors within the unit switch are synchronized by a master clock signal. If this is to work, the propagation delay across the unit must be significantly less than 1/2 of a clock cycle. Taking .33C figure as correct, and limiting delay to 1/4 of a clock cycle, the maximum size of a functional unit is about 8mm. This is not far removed from the size of structure on modern CPU chips. You can make functional units accept larger delays (that's one application of pipelines), but this carries the price of complexity.
The point: power consumption is an important problem, but signal propagation is also very relevant. If 3GHz isn't the limit, from a signal propagation point of view, it is not so far away from that limit...
Here's a chart showing how the race to ever-faster processors came to a screeching halt [tomshardware.com] a few years ago.
Re:Great news (Score:3, Informative)
Your calculation assumes light is traveling in a vacuum. The velocity of light is always slower in a medium than in a vacuum. Our computers use copper and silicon (and other materials), in which propagation is by electrons, not light. Anyways, light speed would be slower in fibre optics than in a vacuum.
The propagation of electrons in copper is about 2/3 that of light speed in a vacuum, which on the time and length scales we're using in computers, is quite significant.
Re:That's retarded. (Score:3, Informative)
AMD was founded 16 years before ATI and was producing branded processors before ATI existed.
Re:That's retarded. (Score:3, Informative)
Matrox [wikipedia.org] is older.
We? No, X86 vendors hit it, RISC is at 4.25 Ghz (Score:3, Informative)
http://en.wikipedia.org/wiki/POWER7 [wikipedia.org]
IBM settled around 4.25 Ghz now. Their original promise (which seems to be very expensive) is around 5+ Ghz speeds.
Don't get me wrong, that is a high end/enterprise UNIX server chip, I don't say Apple should be shipping POWER7 now.
If they just... took consumer desktop&portable CPU business serious...