4 ms·

AMD 4700S Review: Defective Playstation 5 Chip Resurrected

8 points by ItsTotallyOn 5y ago

bcrl 5y ago
Gaming benchmarks seem like the worst possible way to evaluate / portray this kind of CPU since the GPU is completely throttled by the PCIe 2.0 x4 slot housing it. I'd be really curious to see how something like large parallel compiles ran on it since the additional memory bandwidth might well help on multicore workloads.
- johng 5y ago
  Uhhh, if you read the article you'll see they test it with discrete GPUs. In fact, this statement is in the opening paragraph: "rests on its ability to push a discrete GPU to high levels of performance. Today we'll put the chip through its paces in a full spate of tests, including gaming and applications. "
  - detaro 5y ago
    Yes, and said discrete GPU is limited by the "PCIe 2.0 x4 slot housing it", so parent obviously is talking about that? Also this > Uhhh, if you read the article is something the HN rules explicitly ask you to not do, and quite ironic given how badly you misread the comment you replied to.
  - bcrl 5y ago
    I read the entire article before posting, and that's how I knew that the discrete GPU is sitting in an electrically PCIe 2.0 x4 slot (even though it has an x16 connector on the motherboard) that makes it bandwidth limited at somewhat less than 2GB/s. Modern discreet GPUs are now typically benchmarked in PCIe 4.0 x16 slots with upwards of 32GB/s of bandwidth. Having 1/16th the amount of bandwidth to and from the GPU versus the other CPUs in the article have is going to result in it having a pretty massive performance handicap. The article had very few benchmarks relevant to typical software development workloads (a single LLVM compilation graph vs dozens of gaming benchmarks that we already knew it was going to lose is not very balanced in my opinion). If it were up to me, I'd do a few more single vs multithreaded benchmarks for developer workloads (ie Linux kernel compiles) to see how the GDDR6 impacts multicore scaling as compared to other CPUs. The few benchmarks that did have single threaded vs multithreaded results showed that it did fairly well on multithreaded workloads, but the latency hit of the GDDR6 controller was hard to overcome. A curious CPU, yes, but not really worth much more in the real world.

8 points by ItsTotallyOn 5y ago

bcrl 5y ago

Gaming benchmarks seem like the worst possible way to evaluate / portray this kind of CPU since the GPU is completely throttled by the PCIe 2.0 x4 slot housing it. I'd be really curious to see how something like large parallel compiles ran on it since the additional memory bandwidth might well help on multicore workloads.

johng 5y ago
Uhhh, if you read the article you'll see they test it with discrete GPUs. In fact, this statement is in the opening paragraph: "rests on its ability to push a discrete GPU to high levels of performance. Today we'll put the chip through its paces in a full spate of tests, including gaming and applications. "
- detaro 5y ago
  Yes, and said discrete GPU is limited by the "PCIe 2.0 x4 slot housing it", so parent obviously is talking about that? Also this > Uhhh, if you read the article is something the HN rules explicitly ask you to not do, and quite ironic given how badly you misread the comment you replied to.
- bcrl 5y ago
  I read the entire article before posting, and that's how I knew that the discrete GPU is sitting in an electrically PCIe 2.0 x4 slot (even though it has an x16 connector on the motherboard) that makes it bandwidth limited at somewhat less than 2GB/s. Modern discreet GPUs are now typically benchmarked in PCIe 4.0 x16 slots with upwards of 32GB/s of bandwidth. Having 1/16th the amount of bandwidth to and from the GPU versus the other CPUs in the article have is going to result in it having a pretty massive performance handicap. The article had very few benchmarks relevant to typical software development workloads (a single LLVM compilation graph vs dozens of gaming benchmarks that we already knew it was going to lose is not very balanced in my opinion). If it were up to me, I'd do a few more single vs multithreaded benchmarks for developer workloads (ie Linux kernel compiles) to see how the GDDR6 impacts multicore scaling as compared to other CPUs. The few benchmarks that did have single threaded vs multithreaded results showed that it did fairly well on multithreaded workloads, but the latency hit of the GDDR6 controller was hard to overcome. A curious CPU, yes, but not really worth much more in the real world.