DDR4 Memory Scaling Performance On AMD Raven Ridge / Ryzen 5 2400G

[phoronix]

Phoronix: DDR4 Memory Scaling Performance On AMD Raven Ridge / Ryzen 5 2400G

While we all know that APUs crave as fast as system memory as possible, with DDR4 memory kits these days easily costing more than the Ryzen 3 2200G and even the Ryzen 5 2400G, here are some reference results when testing the Ryzen 5 2400G under Linux with memory speeds from DDR4-2133MHz to DDR4-3600MHz...

http://www.phoronix.com/scan.php?page=news_item&px=Ryzen-2400G-Memory-Scaling

[davidbepo]

a shame that launch support for graphics is so poor
the test that were interesting were the graphics ones but they couldnt be performed

[schmidtbag]

Huh... It's a lot less starved for bandwidth than I expected. I figured this was going to be like the FM2 days where the increase of frequency yielded a nearly directly proportionate performance improvement.

It'll be interesting to see if driver updates after a half year from now affect this.

[torsionbar28]

Huh... It's a lot less starved for bandwidth than I expected.

Looks like the synthetic benchmarks are the only place where it makes a big difference. If real world software only shows ~7% difference between 2133 vs 3600, I agree it doesn't seem very bandwidth starved that's for sure.

[duby229]

And I was looking at those results thinking they looked badly bandwidth starved. But it's not at the memory controller, it's at the L3 cache. Some more specific benchmarking would need to be done to show that though.

[lucasbekker]

Looks like the synthetic benchmarks are the only place where it makes a big difference. If real world software only shows ~7% difference between 2133 vs 3600, I agree it doesn't seem very bandwidth starved that's for sure.

Faster memory for APU's is most relevant for graphics. These benchmarks don't include any...

[schmidtbag]

Faster memory for APU's is most relevant for graphics. These benchmarks don't include any...

Oh whoops I somehow totally missed that point, which is kinda a big one. Yeah graphics benchmarks will be the real test for memory bandwidth.

[andre30correia]

Amd need to improve the drivers at launch, windows drivers look a nightmare in all reviews I see

[dungeon]

Amd need to improve the drivers at launch, windows drivers look a nightmare in all reviews I see

Which drivers ? They support Vega only on Windows 10 and only 64bit, basically one and only Windows OS and nothing else... i somehow expect supporting one target would be as easiest as possible to support, but it seems it isn't

[birdie]

7-Zip results don't seem to be valid:

2666 - 16732
2800 - 16816 (+5.0% RAM speed, +0.5% performance)
3000 - 17036 (+7.1% RAM speed, +1.3% performance)
3200 - 17567 (+6.7% RAM speed, +3.1% performance)

This all doesn't look even remotely real and needs to be retested with e.g. CPU governor set to maximum performance, 'cause otherwise Michael is testing God knows what and shows God knows what results.