Two weeks ago, in a test that focused on memory throughput, cache, and latency, we compared the results of the desktop version of APU Renoir / Ryzen 7 4700G and upcoming Cezanne / Ryzen 7 5700G tested with comparable memories. The novelty sample performed better in 12 tests out of a total of 16. In one it was a felt and in three it turned out worse. One of those unsuccessful results was the write speed to memory (DDR4). Now, more and more recent results have appeared. We can’t compare 1: 1 at all, because slightly faster memories were used (DDR4-4733 instead of DDR4-4400), but it is clear that this problem has been fixed. Writing to memory is not ~ 20% slower than on Renoir, but ~ 10% faster (current results in bold):
| read | write | copy | latency | ||
|---|---|---|---|---|---|
| DDR4 | 4700G | 64215 MB / s | 63228 MB / s | 58845 MB / s | 62,8 ns |
| 5700G #1 | 67234 MB / s | 49694 MB / s | 60345 MB / s | 54,9 ns | |
| – | +4,7 % | -21,4 % | +2,5 % | -12,6 % | |
| 5700G #2 | 69578 MB / s | 69380 MB / s | 60780 MB / s | 44,0 ns | |
| – | +8,4 % | +9,7 % | +3,3 % | -29,9 % | |
| L1 | 4700G | 2132,8 GB/s | 1070,7 GB/s | 2131,7 GB/s | 0,9 ns |
| 5700G | 2317,8 GB/s | 1171,6 GB/s | 2336,3 GB/s | 0,9 ns | |
| – | +8,7 % | +9,4 % | +9,6 % | ±0 % | |
| L2 | 4700G | 1072,0 GB/s | 998,75 GB/s | 1072,7 GB/s | 2,7 ns |
| 5700G | 1173,4 GB/s | 1153,4 GB/s | 1130,3 GB/s | 2,6 ns | |
| – | +9,5 % | +15,5 % | +5,4 % | -3,7 % | |
| L3 | 4700G | 668,11 GB/s | 747,85 GB/s | 650,38 GB/s | 9,5 ns |
| 5700G | 740,80 GB/s | 683,83 GB/s | 737,93 GB/s | 11,2 ns | |
| – | +10,9 % | -8,6 % | +13,5 % | +17,9 % |
New results for the cache are not available, but it can be assumed that the problem of writing to RAM was related to slower writing to the L3 cache and its higher latencies, so one problem that affected the three results was probably eliminated.
The hardware revision remained the same (A0), so the repair is due to the newer BIOS. The test sample ran on a boost over 4.7 GHz, the memory accesses are fine, so it looks like the platform is in a state where it could go on the market. However, limited production capacity will be a problem.
But I can offer one positive consideration: AMD is expected to release this year Zen 3+. Although this generation (in an older) roadmap was described as 7nm, there are currently rumors that it could be produced on a 6nm process. However, these two pieces of information do not substantially contradict each other. For Zen 3+ because the transition to the 7nm EUV process could have been originally planned, but the TSMC failed to bring it to a state with which it would be satisfied, so it tried again and marked a more successful version of the process as 6nm (also uses EUV, has very similar parameters as 7nm EUV, but probably better recovery). This seems to be a situation similar to when AMD originally stated that Zen+ will be manufactured on a 14nm ++ process, but GlobalFoundries eventually decided to name it 12nm. But basically the same thing.
Where am I going: If AMD moves at least part of the production of stand-alone desktop processors to 6nm lines (they use EUV lithography, so these are different lines than for the 7nm immersion process), more 7nm production capacity will be freed and AMD will have space to produce and release more 7nm products . With a little luck, the Ryzen 7 5750G could be one of them.
–
