[SOLVED] Cell CPU vs. modern day CPU

Pero_2

Reputable
Dec 9, 2016
6
0
4,510
I know this comparison isn't easy to do because of different architectures used but I'm wondering which cpu today can rival and even surpass a Cell cpu in terms of doing vector and floating point math? AFAIK any modern cpu can blow cell out of water but I'm interested in particular area I mentioned. I'd really like to know how far we got in this regard with modern cpu's-namely with gaming/workstation cpu's? The whole idea of cpu that's 10+ years old competing with modern ones seems a bit far stretched if I may say, comic. It would be also helpful to hear an answer from someone who works in field but any answer is most welcome. I apologize if I haven't been perfectly clear, I have little experience with it.
 
Last edited:
Solution
According to Wikipedia, the Cell was capable of 25.6 GFLOPS of FP32 per SPE at 3.2GHz, which is about 8 FP32 operations per clock cycle per SPE. If we look at the table on Wikipedia's FLOPS page, Intel's Core microarchiteture with SSE4 was capable of doing 8 FP32 ops per core, with AMD implementing the feature in K10. So on a per-clock performance, the Cell wasn't really anything special. Where it was special was it had more "cores" to do those operations. The PS3's Cell was essentially a 1+7 core processor in a time where dual core processors were becoming the mainstream and quad-core processors were on the top-end in PC land.

Today, using Intel's or AMD's most recent designs, they have 4/8 times the per-cycle performance. So...
According to Wikipedia, the Cell was capable of 25.6 GFLOPS of FP32 per SPE at 3.2GHz, which is about 8 FP32 operations per clock cycle per SPE. If we look at the table on Wikipedia's FLOPS page, Intel's Core microarchiteture with SSE4 was capable of doing 8 FP32 ops per core, with AMD implementing the feature in K10. So on a per-clock performance, the Cell wasn't really anything special. Where it was special was it had more "cores" to do those operations. The PS3's Cell was essentially a 1+7 core processor in a time where dual core processors were becoming the mainstream and quad-core processors were on the top-end in PC land.

Today, using Intel's or AMD's most recent designs, they have 4/8 times the per-cycle performance. So even a single/dual core processor running a 3.2GHz should match what the Cell could do.
 
  • Like
Reactions: ragnarok0274
Solution

Pero_2

Reputable
Dec 9, 2016
6
0
4,510
According to Wikipedia, the Cell was capable of 25.6 GFLOPS of FP32 per SPE at 3.2GHz, which is about 8 FP32 operations per clock cycle per SPE. If we look at the table on Wikipedia's FLOPS page, Intel's Core microarchiteture with SSE4 was capable of doing 8 FP32 ops per core, with AMD implementing the feature in K10. So on a per-clock performance, the Cell wasn't really anything special. Where it was special was it had more "cores" to do those operations. The PS3's Cell was essentially a 1+7 core processor in a time where dual core processors were becoming the mainstream and quad-core processors were on the top-end in PC land.

Today, using Intel's or AMD's most recent designs, they have 4/8 times the per-cycle performance. So even a single/dual core processor running a 3.2GHz should match what the Cell could do.
I see. Have any ideas why such higher core count weren't implemented in computer market first? What was first CPU that surpassed it in that regard ( if you know by chance )?