AMD Ryzen 7 5700G Linux Performance

[zerothruster]

I bought a 4750G a few weeks ago for compiling. Not as nice as zen3, obviously, but works well (only PCIe3 SSD but speeds of that seem similar to SATA and "fast enough"). The important things for me were that it was available (found somewhere in GB building to order with scope for variations), avoided needing to sell a body part to buy a GPU, and uses a sensible amount of power.

Very pleased, I'm sure people with the 57nn parts will be equally pleased. Only drawback is that mine needs secure boot, but I guess that is now common.

[gukin]

Michael.

If you can, could you benchmark some different memory speeds, ie. 2400, 3200, 3600, 4000 etc.?

Ryzen APUs have always been sensitive to memory bandwidth and it would certainly be helpful to someone purchasing a system to know what memory speed would be best bang for the buck.

[foobaz]

AMD seem to finally enable ECC support on non-pro APUs.

Please be careful, there has been bad information circulated regarding ECC support on APUs. In the past, people have posted that it works because their system operates with ECC UDIMMs installed, not realizing the RAM was operating in non-ECC mode. I myself bought a Ryzen 2400G and ECC RAM but dmidecode showed it running as non-ECC.

It is very possible AMD has fixed this issue but I recommend waiting for a dmidecode screenshot to be posted before making purchasing decisions.

[domih]

Michael.

If you can, could you benchmark some different memory speeds, ie. 2400, 3200, 3600, 4000 etc.?

Ryzen APUs have always been sensitive to memory bandwidth and it would certainly be helpful to someone purchasing a system to know what memory speed would be best bang for the buck.

I ran the testing suites pts/compilation, pts/compression, pts/cryptography, pts/java, pts/python, pts/productivity, pts/unigine, pts/video-encoding and pts/cassandra using the 3400g with 2666 MT/s, 4750g with 2666 MT/s, 3200 MT/s and 3600 MT/s and 5700g with 3200 MT/s and 3600 MT/s all other things being equal (e.g. same mobo).

What I saw:

Yes there is a significant improvement from 3400g (Zen+) to 4750g (Zen 2) using identical memory.

Yes there is a significant improvement from 4750g (Zen 2) and 5700g (Zen 3) using identical memory.

Yes there is improvement from 2400/2666 to 3200 MT/s using identical CPU.

No it is a diminishing return story above 3200 MT/s. The 3200 and 3600 exchanged barbs in both directions in all the testing suites and mostly within margin of error.

CONCLUSION

Very specifically memory focus benchmark can show high return on the dollar but with more application level benchmark the difference is within margin of error once you reached the recommended memory, 3200 MT/s in this example.

Go with 3200 MT/s and rather spend the additional $$$ you could spend on a super-duper-bester-faster memory kit on the CPU unless you are super reach and you want a trophee PC with trophee components to go with your trophee wife.

In games this translate in a few more frames here and there as shown there: https://www.youtube.com/watch?v=mi4eAWWWxGc

IMHO these stories of high-end memory are borderline urban legends and I'm sure vendors are willing to propagate them so that they can sell higher priced kits.

HTH,

Disclosure: I'm not a gamer not a graphics oriented user. I like CPU with iGPU because I can use them even in mini-itx boards that offers x8 + x8 bifurcation on the single x16 PCIe slot and then plug in a fast network card (10 GbE or higher) and a true hardware RAID card. Such a DIY system SFF is then cheaper than these expensive QNAP, Synology, Terra Master turn-key appliances.

Best.

[George99]

Only drawback is that mine needs secure boot, but I guess that is now common.

I built my PC around a Ryzen 4650G too. No need to use secure boot here in fact I boot from a legacy MBR partitioned SSD. (I only switched motherboard, CPU and (ECC)-DRAM.
Motherboard is ASUS Prime B450M-A, both very affordable and good firmware update support.

[loganj]

what happen to GE variants of 5000 series?
will they only be available for OEM builds?

[domih]

The 3200 and 3600 exchanged barbs in both directions in all the testing suites and mostly within margin of error.

This kind of situation always make me think about what a great new feature the Phoronix Test Suite could implement:

Add 3 columns to the Test Results graphs:
- %: the difference expressed in percentage. Personally anything under 5% is not worth $$$.
- Comparison: based on the estimated margin of error: Significant, Borderline or Non-Significant difference.
- Who will notice: Anybody, Power User, Professional, Expert

I'm not a gamer, so I do not know how one feels about Frame Rates. But for a beginner gamer, is there really a difference of user experience between 125 and 150 FPS? or 60 and 90? I don't know. What I do know is that even for a 3 FPS difference while we are in the stratospheric 100 or above FPS there are Volga-river-long online discussion threads about which hardware is the "best of the best" based on the result. Really?

Not talking about Phoronix here but how many times one can see online stuff like "hardware A" rates at 2387, while "hardware B" rates at 2490 and therefore B is so mucho much better, while in fact the difference is 4.3% and a lambda user will probably never see the difference.

I don't even want to talk about charts where the origin is not zero in order to "highlight" the difference.

Phoronix could show the way to the pseudo-science statisticians from a few major tech news sites (I'm not giving names) and restore some sanity (...)

[willupowers]

Is the CPU package power reporting ok? It seems there might be something weird going on since the 11600K is trending higher than the 11900K on the light load/idle states. Any possibility of some power test results at the wall or at the PSU out for idle and non-idle?

[Paradigm Shifter]

I'd like to see a performance comparison between 2133/2666/3200/4000, as the overclocking forums I'm active on swear by 4000MHz RAM, but I'm still chugging along with 3200MHz on one system (it originally had 2666 and I noticed basically no difference in performance in my daily workload) and even 2133MHz in one box! But my daily workload is a mix of CPU, RAM, GPU and I/O load (not all maximum at the same time, but each will be fully loaded at some point during a normal workload) so there are many factors which come into play. Faster RAM is nice, but frankly more cores and faster I/O will give me larger gains than faster RAM.

It basically boils down to buy the fastest stuff you can afford at the capacity you require. If the price difference between 2666/2933 and 3200 is minimal (it was when I upgraded from 64GB 2666 to 128GB 3200) then why not go faster? But with 3600/4000/4400 RAM the prices go crazy pretty quickly.

edit: ...and I forgot to comment on the CPU.

I've been able to find them in stock here (finally) so am weighing up buying one to replace the Ryzen 1700 in my home PC. A 65w TDP is perfect as it's the same as the 1700, but should be a big jump in CPU work. Only problem is do I try to update the BIOS of my current board (if it will support it) or just replace that too? And suddenly what wasn't so expensive gets that way quickly. Bah.

[ddriver]

Please be careful, there has been bad information circulated regarding ECC support on APUs. In the past, people have posted that it works because their system operates with ECC UDIMMs installed, not realizing the RAM was operating in non-ECC mode. I myself bought a Ryzen 2400G and ECC RAM but dmidecode showed it running as non-ECC.

It is very possible AMD has fixed this issue but I recommend waiting for a dmidecode screenshot to be posted before making purchasing decisions.

In order to get ecc running you need cpu support and motherboard support, for starters the mobo has to have the extra traces, which many don't, and also firmware support.

Asrock have always explicitly stated that ecc is not supported on NON PRO APUs. So it is no surprise that the 2400g didn't. They are my goto for ecc systems, almost all of their boards support it.

They have updated their info to:

For Ryzen Series APUs (Picasso and Renoir), ECC is only supported with PRO CPUs.

This no longer puts apus under the same umbrella, leading me to presume that the 5000 series cezanne apus indeed support it.

Also, Wendel from L1T claims it will be supported this time.