Official PS3 Thread
- Thread starter Sonic
- Start date
- Status
qwerty2000
Newcomer
Paul said:
here where i got 512
32 gflops mutilpy by a apus will give you 128 gflops and multilpy by 4 controllers will give you 512 gflops
megadrive0088
Regular
BroadBand Engine CPU - 1TFLOPS - 32 APUs / 128 FPUs - 2-4 GHz?
GPU - 128-256 GFLOPs - 16 APUs / 64 FPUs 1-2 Ghz?
GPU probably 256 GFLOPs (if 2 Ghz)
GPU - 128-256 GFLOPs - 16 APUs / 64 FPUs 1-2 Ghz?
GPU probably 256 GFLOPs (if 2 Ghz)
qwerty2000
Newcomer
every fpu dose 32 gflops you'll get more than 128 glops
megadrive0088
Regular
The GPU/Visualizer in the patient shows it has 16 APUs total. there are 4 APUs per Processing Element (and 4 PEs) instead of 8 APUs per PE as in the CPU. there are usually 4 FPUs per APU. each FPU is NOT ment to produce 32 GFLOPs. it is the 4 FPUs together in each APU that produces the 32 GFLOPs, thus each APU produces 32 GFLOPs, but that is in the CPU only, from what I understood. since the APUs of the GPU would be clocked lower (tho i'm not sure of this) they would not produce 32GFLOPs, but 8-16 GFLOPs.
the GPU clockspeed was supposed to be 1-2 GHz. so each APU in the GPU should produce 8-16 GFLOPs depending on clockrate.
16 GFLOPs * 16 APUs is 256 GFLOPs
or
8 GFLOPs * 16 APUs is 128 GFLOPs
for the GPU.
All of this is based on my previous understanding of the APUs. if these things no longer apply obviously the all of the things I've said in my post goes out the window
the GPU clockspeed was supposed to be 1-2 GHz. so each APU in the GPU should produce 8-16 GFLOPs depending on clockrate.
16 GFLOPs * 16 APUs is 256 GFLOPs
or
8 GFLOPs * 16 APUs is 128 GFLOPs
for the GPU.
All of this is based on my previous understanding of the APUs. if these things no longer apply obviously the all of the things I've said in my post goes out the window
megadrive0088
Regular
if the APUs of the GPU are running at 4 Ghz then they would get
32 GFLOPs. then 32 GFLOPs * 16 APUs is 512 GFLOPs.
unless somehow the APUs of the GPU are each getting GFLOPs with a different clockspeed than 4 GHz. if each APU had more than FPUs
(say 8 FPUs per APU)
which is more than possible because....
Cell based CPUs and GPUs are modualar / scalable, so... you can get a given level of performace per APU by adding / removing FPUs.
32 GFLOPs. then 32 GFLOPs * 16 APUs is 512 GFLOPs.
unless somehow the APUs of the GPU are each getting GFLOPs with a different clockspeed than 4 GHz. if each APU had more than FPUs
(say 8 FPUs per APU)
which is more than possible because....
Cell based CPUs and GPUs are modualar / scalable, so... you can get a given level of performace per APU by adding / removing FPUs.
megadrive0088
Regular
now my question is, could the APUs of the GPU/Visualizer be running at a different clockspeed than the Pixel Engines?
megadrive0088
Regular
a 1 Ghz Cell-CPU / Broadband Engine would only get 256 GFLOPs then.
assuming 4 Processing Elements, each PE with 8 APUs, each APU with 4 FPUs and each APU getting 32 GFLOPs
unless they have 16 Processing Elements / 128 APUs / 512 FPUs
(8 GFLOPs per APU then) - then you could have 1 TFLOPs @ 1 Ghz
I mean that would be one way of getting 1 TFLOPs
IIRC, in 2001-2002, various tech reports mentioned 16 cores in a console version of Cell. but only 4 cores in a Cell for set-top boxes.
For PS3, assuming that each of the 16 cores is a Processing Element, each PE with it's own array of APUs [8] each APU with 4 FPUs, each APU getting 32 GFLOPs. then a 1 Ghz Cell with 16 PEs/cores could get 1TFLOPs.
this is so much fun.
assuming 4 Processing Elements, each PE with 8 APUs, each APU with 4 FPUs and each APU getting 32 GFLOPs
unless they have 16 Processing Elements / 128 APUs / 512 FPUs
(8 GFLOPs per APU then) - then you could have 1 TFLOPs @ 1 Ghz
I mean that would be one way of getting 1 TFLOPs
IIRC, in 2001-2002, various tech reports mentioned 16 cores in a console version of Cell. but only 4 cores in a Cell for set-top boxes.
For PS3, assuming that each of the 16 cores is a Processing Element, each PE with it's own array of APUs [8] each APU with 4 FPUs, each APU getting 32 GFLOPs. then a 1 Ghz Cell with 16 PEs/cores could get 1TFLOPs.
this is so much fun.
qwerty2000
Newcomer
that mean the gpu needs a clock speed of 4 ghz to get 512 gflops
megadrive0088
Regular
yes.
qwerty2000
Newcomer
i dont think the gpu will be ablie to achive 4ghz for the gpu so i'll take the 256 glops then runnig at 2hz
megadrive0088
Regular
yep. 256 GFlops for GPU is what was largely agreed opon by some of the forum goers a few months back when the Sony patient became public knowledge.
megadrive0088
Regular
yeah that is certainly a possibility. a good possibility even, if PS3 comes in at the absolute bare minimum of what Sony has said (1 TFLOPs)
however, recall that Sony at one time was saying Cell would be in the multi-teraflops range of performance. i'll have to find a quote for that, but.... so at the very least, I would think 1 TFLOP for the CPU alone would be needed if ppl are to concider Cell reaching its minimum stated goal.
It wouldn't look too good if the PS3 Cell CPU (one of the more powerful configurations of Cell at that) only reached 700-900 GFLOPs,
and that isnt even sustained power, its peak of course 900 GFLOPs or 1 TFLOP does not really make that much of a technical difference. but it could make a huge difference in perception overall, giving fuel to Sony's advisaries who will go, 'ha! Cell is a failure blah blah blah'
megadrive0088
Regular
ahhh, here is but one of the quotes I was looking for:
http://www.ibm.com/news/2001/03/12.phtml
quote:
Sony, Toshiba, IBM to develop supercomputer on a chip
Sony Computer Entertainment, Toshba Corp. and IBM are joining to to research and develop an advanced chip architecture for a new wave of devices in the emerging broadband era.
The companies will invest more than $400 million in the next five years to design a "supercomputer-on-a-chip" at a development center within an IBM facility in Austin, Texas.
Code-named "Cell," the new microchips will employ the world's most advanced research technologies and chip-making techniques, including copper wires, silicon-on-insulator transistors and low-K dielectric insulation, with features smaller than 0.10 microns -- 1,000 times thinner than a human hair.
The result will be consumer devices that are more powerful than IBM’s Deep Blue supercomputer, operate at low power and access the broadband Internet at ultra high speeds. ----->Cell will be designed to deliver "teraflops" of processing power<-----
Each company will manufacture the product for a variety of consumer applications.
"We're defining the next era of computing, providing the technology that will bring computer intelligence and network access to a wide array of consumer electronics," said Dr. John Kelly, senior vice president and group executive for the IBM Technology Group. "As a result, IBM’s advanced chip technologies are in more demand than ever. We expect a considerable portion of our new, state-of-the-art 300 mm wafer manufacturing facility in Fishkill, N.Y. , to be dedicated to this product."
The expansion of ultra high-speed broadband networks, coupled with advancements in semiconductor technology, is making possible a whole new range of products and services that use the Internet as a source of entertainment, information and communication. These new chips from Sony, IBM and Toshiba will enable global communication through the broadband network.
http://www.ibm.com/news/2001/03/12.phtml
quote:
Sony, Toshiba, IBM to develop supercomputer on a chip
Sony Computer Entertainment, Toshba Corp. and IBM are joining to to research and develop an advanced chip architecture for a new wave of devices in the emerging broadband era.
The companies will invest more than $400 million in the next five years to design a "supercomputer-on-a-chip" at a development center within an IBM facility in Austin, Texas.
Code-named "Cell," the new microchips will employ the world's most advanced research technologies and chip-making techniques, including copper wires, silicon-on-insulator transistors and low-K dielectric insulation, with features smaller than 0.10 microns -- 1,000 times thinner than a human hair.
The result will be consumer devices that are more powerful than IBM’s Deep Blue supercomputer, operate at low power and access the broadband Internet at ultra high speeds. ----->Cell will be designed to deliver "teraflops" of processing power<-----
Each company will manufacture the product for a variety of consumer applications.
"We're defining the next era of computing, providing the technology that will bring computer intelligence and network access to a wide array of consumer electronics," said Dr. John Kelly, senior vice president and group executive for the IBM Technology Group. "As a result, IBM’s advanced chip technologies are in more demand than ever. We expect a considerable portion of our new, state-of-the-art 300 mm wafer manufacturing facility in Fishkill, N.Y. , to be dedicated to this product."
The expansion of ultra high-speed broadband networks, coupled with advancements in semiconductor technology, is making possible a whole new range of products and services that use the Internet as a source of entertainment, information and communication. These new chips from Sony, IBM and Toshiba will enable global communication through the broadband network.
megadrive0088
Regular
some things said in this thread that are worth repeating:
"Why should they do such a thing ? The chip in Figure 6 might not be for PlayStation 3 as the Broadband Engine and Visualizer might be even better than that
Seriously, that patent doesn't have to cover the final and exact IMPLEMENTATION of the chips in PlayStation 3 as they can see modifications over the example that the patent was providing..."
+1 Panajev
"The 1tflop figure is based on Kutaragi promising that PS3 will 1000X as powerful as PS2. The flops for EE is 6.2Gflops, so the performance of PS3 should be 6.2Tflops, not 1Tflops. So you see, Pana, Paul, the actual performance might be even greater than the example given in the patent. That's what all those Cell chips in the patents were, examples of possible configurations, NOT actual final specifications for Cell chips."
bbot had a really good point there, regardless of what actually happens with PS3's CPU.
"You DONT multiply everything by 1000X. 1000X means the total power will be that great with everything put together, the cpu + rasterizer + greater efficiancy will all together do it."
Paul hits the nail right on the head.
"PlayStation 3 using the patent Broadband Engine would not only have lots of FP power, but also lots of Integer processing power as well ( 1 Tera OPS ) and much more bandwidth and Local Memory...
We have e-DRAM and a 1,024 bits bus to it and connecting all the APUs in each PE ( PE bus is 1,024 bits wide ), 4 MB of total Local Storage ( 128 KB of Local Storage, realized in SRAM, in each APU.. this is where the execution of code happens ) and in each APU we have thirty-two 128 bits registers...
We haven't just increased the FLOPS and OPS rating of the machine, we added more of those features which helps keep the efficiency high..."
Panajev hits a homerun out of the park
"Yes, that efficiency might add to the perf difference between it and the EE... What is the real world performance of the EE? The specs are 6.2Tflops, but the real world number is likely lower...
Anyway last time sony said 100x perf... and they showed 100x perf in one of the specs, accompanied by improvements in all others.... 350K low rez no effects poly peak for psone.... 75M raw GS for ps2....
100K ingame with barely any effects psone.... 10-30M+ for ps2 with lots of effects and higher rez...
So indeed, a 100X jump was done on one of the specs, and probably others too... the EE is probably 100x+, the vram bandwith, etc..."
zidane1strife backs Panajev, strengthning his case while making some points of his own
"Whether CELL or not but IBM/SONY/TOSHIBA are designing the CPU for PS3....thats for sure!
Could it be that they are working on two projects together....CELL for new network servers and something else for PS3......maybe our speculations are wrong.....PS3 CPU may not be CELL (I dont care) but it will still be more than 1TFLOPs (I am happy now )....
So no need to fight over whether CELL will be there or not.....SONY will make sure that they have a powerful CPU for PS3.."
Deepak's exellent post that makes moot the arguements that people make about Cell not being in PS3
"Why should they do such a thing ? The chip in Figure 6 might not be for PlayStation 3 as the Broadband Engine and Visualizer might be even better than that
Seriously, that patent doesn't have to cover the final and exact IMPLEMENTATION of the chips in PlayStation 3 as they can see modifications over the example that the patent was providing..."
+1 Panajev
"The 1tflop figure is based on Kutaragi promising that PS3 will 1000X as powerful as PS2. The flops for EE is 6.2Gflops, so the performance of PS3 should be 6.2Tflops, not 1Tflops. So you see, Pana, Paul, the actual performance might be even greater than the example given in the patent. That's what all those Cell chips in the patents were, examples of possible configurations, NOT actual final specifications for Cell chips."
bbot had a really good point there, regardless of what actually happens with PS3's CPU.
"You DONT multiply everything by 1000X. 1000X means the total power will be that great with everything put together, the cpu + rasterizer + greater efficiancy will all together do it."
Paul hits the nail right on the head.
"PlayStation 3 using the patent Broadband Engine would not only have lots of FP power, but also lots of Integer processing power as well ( 1 Tera OPS ) and much more bandwidth and Local Memory...
We have e-DRAM and a 1,024 bits bus to it and connecting all the APUs in each PE ( PE bus is 1,024 bits wide ), 4 MB of total Local Storage ( 128 KB of Local Storage, realized in SRAM, in each APU.. this is where the execution of code happens ) and in each APU we have thirty-two 128 bits registers...
We haven't just increased the FLOPS and OPS rating of the machine, we added more of those features which helps keep the efficiency high..."
Panajev hits a homerun out of the park
"Yes, that efficiency might add to the perf difference between it and the EE... What is the real world performance of the EE? The specs are 6.2Tflops, but the real world number is likely lower...
Anyway last time sony said 100x perf... and they showed 100x perf in one of the specs, accompanied by improvements in all others.... 350K low rez no effects poly peak for psone.... 75M raw GS for ps2....
100K ingame with barely any effects psone.... 10-30M+ for ps2 with lots of effects and higher rez...
So indeed, a 100X jump was done on one of the specs, and probably others too... the EE is probably 100x+, the vram bandwith, etc..."
zidane1strife backs Panajev, strengthning his case while making some points of his own
"Whether CELL or not but IBM/SONY/TOSHIBA are designing the CPU for PS3....thats for sure!
Could it be that they are working on two projects together....CELL for new network servers and something else for PS3......maybe our speculations are wrong.....PS3 CPU may not be CELL (I dont care) but it will still be more than 1TFLOPs (I am happy now )....
So no need to fight over whether CELL will be there or not.....SONY will make sure that they have a powerful CPU for PS3.."
Deepak's exellent post that makes moot the arguements that people make about Cell not being in PS3
- Status
Similar threads
- Locked
