I disagree. I expect 128-bit registers just like AltiVec and the PS2's vector units. Each vector core will have a set of at least 32 of them. Consider, for example, that the latest ATI & nVidia graphics chips have large numbers of vertex and pixel shader engines with 12+ 128-bit 4-way floating point registers each.
So are we talking a graphics processor here or a advance CPU? My impression was that Cell would still be supproted by a GPU of some sort. Due to the highly varied nature of the rest of the processing to be done it would seem to be more effiecent to build many smaller units that can be marshalled together if required. Actually I was under the impression that this was a key element of Cell. In other words say you have one or more audio streams needing processing where each stream does not need the capability of a wide vector unit, would this not be a waste to process these streams on multiple vector units of the side type suggested?
Quote:
Number of transistors is far less of a problem than you think... especially on the 65nm process. Also, many of AltiVec's transistors goes to certain kinds of functionality that may not be present in the vector cores.
Well transistors impact price. considering the market for these devices effiecent use of transitors is important.
Quote:
You've heard all the grousing about how hard the PS3 will be to program for, right...? Sony isn't terribly interested in making it easy on the software guys.
Actually I almost never watch TV and I can't even remember when I last played a video game. In any event I would imagine that Sony would have to eventualy take some interest in developers. The risk is that straight multi processor systems such as the one from MS will be far easier to build for.
Quote:
Word == Mum.
Word == Private E-Mail
Quote:
Yes, I expect to see 64-bit integer registers on the scalar core. Whether conventional 64-bit addressing is supported on all (or any) of the cores is another matter.
I would expect to see 64 bit addressing, especially for something that is still a ways off. Of course the big issue is memory to go along with all of that addressing capability.
The auxilliary cores are something else. Since my knowledge of Cell is limited, I can't see why they wouldn't have the same addressing range as the rest of the system.
Quote:
Keep in mind what they actually said -- conventional scaling has hit the wall.
See this is where I'm the complete skeptic. IBM's scaling has hit the wall, but I'm not willing to saddle the rest of the industry with IBM's problems. The final chapter on 90nm has not been written yet. We will soon see product from other manufactures at this feature size that may tell a different story. When Freescale, Chartered, Fujitsu, TI and the other players mature their 90nm processes then we will have a more complete picture of this alleged wall.
Quote:
They (and everyone else) still have avenues to pursue, but the rapid clock rate scaling of the past is over. They don't have a trailing edge process, and everybody else is having troubles too. I don't see where you get the impression that PPC doesn't mean much to them, aside from being impatient for the next new thing.
Ok let me count the ways:
1. Dropped the 32 bit embedded line like a lead ball.
2. Rushed a 90nm process into production for the 970 series
3. Have been less than fourth coming with respect to the 970FX. This may be very well an Apple issue but it does leave potential customers with a bad impression.
4. The slow release of a Power5 Derived PPC. Actually no release at the moment and there are rumbling that there may never be a Power 5 derived PPC.
5. Nothing it the way of affordable motherboards for any of their PPC chips. If nothing else one has to respect Intel for building mother boards and reference platforms for their hardware and making sure that these boards are affordable. Maybe not cheap but atleast affordable.
I could go on but at the moment it looks like PPC is a method of milking Apple for a few profits. There does not seem to be commitment to building a market for PPC.
So are we talking a graphics processor here or a advance CPU? My impression was that Cell would still be supproted by a GPU of some sort. Due to the highly varied nature of the rest of the processing to be done it would seem to be more effiecent to build many smaller units that can be marshalled together if required. Actually I was under the impression that this was a key element of Cell. In other words say you have one or more audio streams needing processing where each stream does not need the capability of a wide vector unit, would this not be a waste to process these streams on multiple vector units of the side type suggested?
Whether the PS3 has a dedicated GPU or not isn't relevant to the discussion at hand. I was drawing a parallel, that's all -- GPUs these days are gangs of specialized programmable vector cores with unconventional execution models and memory accessing schemes.
A "wide" vector unit in the old supercomputer days was thousands of elements. A 4 element vector is not particularly wide, and is an awfully convenient size for working on things like 3D graphics. The incremental cost of a 4-way vector core compared to a scalar core is small. It would be much cheaper to have 8 4-way units instead of 32 1-way units, and you can still do the same kind of work partitioning that you suggest.
Quote:
Well transistors impact price. considering the market for these devices effiecent use of transitors is important.
Not as important as it used to be. At the 65nm node there are a lot of transistors in the "ideal" chip size for a consumer product, and the issue is how to maximize their computational efficiency. As I pointed out in my previous paragraph, having SIMD is actually an efficiency improvement... it increases your computational capability relative to the per-core overhead.
Quote:
Actually I almost never watch TV and I can't even remember when I last played a video game. In any event I would imagine that Sony would have to eventualy take some interest in developers. The risk is that straight multi processor systems such as the one from MS will be far easier to build for.
Yes, that is the risk. It seems that Sony will try to solve the development difficulties with tools after building their ideal hardware, and rely on the value of the PlayStation and Sony brands carrying them farther than MS' attempt. Time will tell if this and the supposed superior performance of the Cell architecture will win out over the MS approach.
Quote:
Word == Private E-Mail
Word == No.
Quote:
I would expect to see 64 bit addressing, especially for something that is still a ways off. Of course the big issue is memory to go along with all of that addressing capability.
Maybe -- there are downsides to straight 64-bit addressing. Its also not as far off as you seem to imply. And memory doesn't have to go along with all of that addressing capability.
Quote:
The auxilliary cores are something else. Since my knowledge of Cell is limited, I can't see why they wouldn't have the same addressing range as the rest of the system.
Well, for one, it is complex to support a flat memory model like we've gotten used to.
Quote:
See this is where I'm the complete skeptic. IBM's scaling has hit the wall, but I'm not willing to saddle the rest of the industry with IBM's problems. The final chapter on 90nm has not been written yet. We will soon see product from other manufactures at this feature size that may tell a different story. When Freescale, Chartered, Fujitsu, TI and the other players mature their 90nm processes then we will have a more complete picture of this alleged wall.
Hmmm. All the existing 90nm products are having troubles, and all the coming 90nm may not. The guys getting there soon are optimistic just like IBM and Intel were before they actually got there. You can draw the conclusions you want, but I know what it looks like to me.
Quote:
1. Dropped the 32 bit embedded line like a lead ball.
2. Rushed a 90nm process into production for the 970 series
3. Have been less than fourth coming with respect to the 970FX. This may be very well an Apple issue but it does leave potential customers with a bad impression.
4. The slow release of a Power5 Derived PPC. Actually no release at the moment and there are rumbling that there may never be a Power 5 derived PPC.
5. Nothing it the way of affordable motherboards for any of their PPC chips. If nothing else one has to respect Intel for building mother boards and reference platforms for their hardware and making sure that these boards are affordable. Maybe not cheap but atleast affordable.
1. The 32-bit products weren't differentiated enough from Freescale's, and they hadn't had serious investment in years. When IBM announced consolidation on Power they clearly meant on their PPC64 lineup. I don't see why pruning the weak part of the lineup is a negative.
2. IBM is trying to sit on the leading edge of process, which is a tough place to be. Their 90nm troubles only seemed to materialize when they went into production, and this seems to be the same for Intel. This doesn't reflect on their commitment to Power/PowerPC.
3. They haven't been forthcoming on lots of things, but that's probably because of trade secrets, NDAs, customer confidentiality, etc. Doesn't seem like much of a pattern change from before, and does it really matter what the general public is told? The guys and Apple, nVidia, AMD, Sony, Toshiba, and Microsoft seem to know a lot... and are willing to invest very large sums of money based on that knowledge.
4. Slow release? the 970 arrived, what, 3 years after the POWER4? The POWER5 shipped in, what, May? That's 6 months ago. Sorry, but engineering doesn't happen according to the impatience level of rumor forum contributors.
5. That's not the business model IBM has chosen (yet). I don't think they are in position to do that right now, and they have more than enough to keep them busy. That they are going to be in every major game console in the next generation, plus supplying Apple and their own server lines is going to be quite enough for now.
Quote:
I could go on but at the moment it looks like PPC is a method of milking Apple for a few profits. There does not seem to be commitment to building a market for PPC.
Comments
Originally posted by Programmer
I disagree. I expect 128-bit registers just like AltiVec and the PS2's vector units. Each vector core will have a set of at least 32 of them. Consider, for example, that the latest ATI & nVidia graphics chips have large numbers of vertex and pixel shader engines with 12+ 128-bit 4-way floating point registers each.
So are we talking a graphics processor here or a advance CPU? My impression was that Cell would still be supproted by a GPU of some sort. Due to the highly varied nature of the rest of the processing to be done it would seem to be more effiecent to build many smaller units that can be marshalled together if required. Actually I was under the impression that this was a key element of Cell. In other words say you have one or more audio streams needing processing where each stream does not need the capability of a wide vector unit, would this not be a waste to process these streams on multiple vector units of the side type suggested?
Number of transistors is far less of a problem than you think... especially on the 65nm process. Also, many of AltiVec's transistors goes to certain kinds of functionality that may not be present in the vector cores.
Well transistors impact price. considering the market for these devices effiecent use of transitors is important.
You've heard all the grousing about how hard the PS3 will be to program for, right...? Sony isn't terribly interested in making it easy on the software guys.
Actually I almost never watch TV and I can't even remember when I last played a video game. In any event I would imagine that Sony would have to eventualy take some interest in developers. The risk is that straight multi processor systems such as the one from MS will be far easier to build for.
Word == Mum.
Word == Private E-Mail
Yes, I expect to see 64-bit integer registers on the scalar core. Whether conventional 64-bit addressing is supported on all (or any) of the cores is another matter.
I would expect to see 64 bit addressing, especially for something that is still a ways off. Of course the big issue is memory to go along with all of that addressing capability.
The auxilliary cores are something else. Since my knowledge of Cell is limited, I can't see why they wouldn't have the same addressing range as the rest of the system.
Keep in mind what they actually said -- conventional scaling has hit the wall.
See this is where I'm the complete skeptic. IBM's scaling has hit the wall, but I'm not willing to saddle the rest of the industry with IBM's problems. The final chapter on 90nm has not been written yet. We will soon see product from other manufactures at this feature size that may tell a different story. When Freescale, Chartered, Fujitsu, TI and the other players mature their 90nm processes then we will have a more complete picture of this alleged wall.
They (and everyone else) still have avenues to pursue, but the rapid clock rate scaling of the past is over. They don't have a trailing edge process, and everybody else is having troubles too. I don't see where you get the impression that PPC doesn't mean much to them, aside from being impatient for the next new thing.
Ok let me count the ways:
1. Dropped the 32 bit embedded line like a lead ball.
2. Rushed a 90nm process into production for the 970 series
3. Have been less than fourth coming with respect to the 970FX. This may be very well an Apple issue but it does leave potential customers with a bad impression.
4. The slow release of a Power5 Derived PPC. Actually no release at the moment and there are rumbling that there may never be a Power 5 derived PPC.
5. Nothing it the way of affordable motherboards for any of their PPC chips. If nothing else one has to respect Intel for building mother boards and reference platforms for their hardware and making sure that these boards are affordable. Maybe not cheap but atleast affordable.
I could go on but at the moment it looks like PPC is a method of milking Apple for a few profits. There does not seem to be commitment to building a market for PPC.
Dave
Originally posted by wizard69
So are we talking a graphics processor here or a advance CPU? My impression was that Cell would still be supproted by a GPU of some sort. Due to the highly varied nature of the rest of the processing to be done it would seem to be more effiecent to build many smaller units that can be marshalled together if required. Actually I was under the impression that this was a key element of Cell. In other words say you have one or more audio streams needing processing where each stream does not need the capability of a wide vector unit, would this not be a waste to process these streams on multiple vector units of the side type suggested?
Whether the PS3 has a dedicated GPU or not isn't relevant to the discussion at hand. I was drawing a parallel, that's all -- GPUs these days are gangs of specialized programmable vector cores with unconventional execution models and memory accessing schemes.
A "wide" vector unit in the old supercomputer days was thousands of elements. A 4 element vector is not particularly wide, and is an awfully convenient size for working on things like 3D graphics. The incremental cost of a 4-way vector core compared to a scalar core is small. It would be much cheaper to have 8 4-way units instead of 32 1-way units, and you can still do the same kind of work partitioning that you suggest.
Well transistors impact price. considering the market for these devices effiecent use of transitors is important.
Not as important as it used to be. At the 65nm node there are a lot of transistors in the "ideal" chip size for a consumer product, and the issue is how to maximize their computational efficiency. As I pointed out in my previous paragraph, having SIMD is actually an efficiency improvement... it increases your computational capability relative to the per-core overhead.
Actually I almost never watch TV and I can't even remember when I last played a video game. In any event I would imagine that Sony would have to eventualy take some interest in developers. The risk is that straight multi processor systems such as the one from MS will be far easier to build for.
Yes, that is the risk. It seems that Sony will try to solve the development difficulties with tools after building their ideal hardware, and rely on the value of the PlayStation and Sony brands carrying them farther than MS' attempt. Time will tell if this and the supposed superior performance of the Cell architecture will win out over the MS approach.
Word == Private E-Mail
Word == No.
I would expect to see 64 bit addressing, especially for something that is still a ways off. Of course the big issue is memory to go along with all of that addressing capability.
Maybe -- there are downsides to straight 64-bit addressing. Its also not as far off as you seem to imply. And memory doesn't have to go along with all of that addressing capability.
The auxilliary cores are something else. Since my knowledge of Cell is limited, I can't see why they wouldn't have the same addressing range as the rest of the system.
Well, for one, it is complex to support a flat memory model like we've gotten used to.
See this is where I'm the complete skeptic. IBM's scaling has hit the wall, but I'm not willing to saddle the rest of the industry with IBM's problems. The final chapter on 90nm has not been written yet. We will soon see product from other manufactures at this feature size that may tell a different story. When Freescale, Chartered, Fujitsu, TI and the other players mature their 90nm processes then we will have a more complete picture of this alleged wall.
Hmmm. All the existing 90nm products are having troubles, and all the coming 90nm may not. The guys getting there soon are optimistic just like IBM and Intel were before they actually got there. You can draw the conclusions you want, but I know what it looks like to me.
1. Dropped the 32 bit embedded line like a lead ball.
2. Rushed a 90nm process into production for the 970 series
3. Have been less than fourth coming with respect to the 970FX. This may be very well an Apple issue but it does leave potential customers with a bad impression.
4. The slow release of a Power5 Derived PPC. Actually no release at the moment and there are rumbling that there may never be a Power 5 derived PPC.
5. Nothing it the way of affordable motherboards for any of their PPC chips. If nothing else one has to respect Intel for building mother boards and reference platforms for their hardware and making sure that these boards are affordable. Maybe not cheap but atleast affordable.
1. The 32-bit products weren't differentiated enough from Freescale's, and they hadn't had serious investment in years. When IBM announced consolidation on Power they clearly meant on their PPC64 lineup. I don't see why pruning the weak part of the lineup is a negative.
2. IBM is trying to sit on the leading edge of process, which is a tough place to be. Their 90nm troubles only seemed to materialize when they went into production, and this seems to be the same for Intel. This doesn't reflect on their commitment to Power/PowerPC.
3. They haven't been forthcoming on lots of things, but that's probably because of trade secrets, NDAs, customer confidentiality, etc. Doesn't seem like much of a pattern change from before, and does it really matter what the general public is told? The guys and Apple, nVidia, AMD, Sony, Toshiba, and Microsoft seem to know a lot... and are willing to invest very large sums of money based on that knowledge.
4. Slow release? the 970 arrived, what, 3 years after the POWER4? The POWER5 shipped in, what, May? That's 6 months ago. Sorry, but engineering doesn't happen according to the impatience level of rumor forum contributors.
5. That's not the business model IBM has chosen (yet). I don't think they are in position to do that right now, and they have more than enough to keep them busy. That they are going to be in every major game console in the next generation, plus supplying Apple and their own server lines is going to be quite enough for now.
I could go on but at the moment it looks like PPC is a method of milking Apple for a few profits. There does not seem to be commitment to building a market for PPC.
Heh. Okay, I'll take your word for it.