AMD Zen 2 + Radeon RX 5700 Series For Linux Expectations

[starshipeleven]

The f**k that made me think that gaming hardware is unstable is: it isn't designed for 100% reliable operation (and especially 24/7). When I first bought this motherboard it would hard-freeze after a few hours of usage, and not even SysRq would help rebooting it. After a firmware update the freezes stopped. However, in November 2017 I found out it may still freeze (especially when playing Dolphin) (although rarely), which means, no, it isn't 100% reliable/stable.

So, because one board or board model is a lemon (or sucks on Linux, which is something the OEMs never usually test for consumer boards) then all boards of the same category are shit? Nice fucking thinking right there.

FYI: it's not normal for any board to just hard-freeze randomly. All PC boards are designed for 100% reliable operation and 24/7 usage. Everyone gets upset when their PC randomly shuts down, not just "workstation users".

User of 2 graphics cards here, but the problem was that my PSU couldn't take it and every damn hour it'd randomly turn off the AMD card, so had to go back to 1 card.

What prevented you from upgrading the PSU instead? You thought that consumer PSUs aren't designed for 100% reliable and 24/7 operation too?

Also, random crashing could be very well caused by unstable or overloaded PSU (it does not deliver stable voltage to the system), if suddenly a 12v rail goes down to 10 or 9v then shit is going to happen.

[oiaohm]

OK, thanks. Seems like 7nm indeed is very cost-effective.

To be correct we are talking in TSMC nm.

https://semiwiki.com/semiconductor/intel/7602-semicon-west-intel-10nm-and-gf-7nm-update/

TSMC 7nm is close to Intel 10nm that Intel is having hell with.

Note the AMD cpu chiplets are 7nm. But the chiplet with memory controller and all the stuff that talked to the motherboard is TSMC 12nm process or roughly 14nm++ intel that most of the current intel chips are made in.

Remember Intel has been attempting single chip in their 10nm process. If this worked out for intel this would have been more cost effective for intel because they own the fab and it would have been less silicon.

I do suspect we might have hit a limit. 10nm intel /7nm tsmc may not in fact be able to support driving a motherboard directly with a decent margin of safety.

AMD had a reason to go chiplet based on cost. Making chips in the older 12nm process is cheaper than using the new 7nm process. Also splitting chip in two basically cuts your failure rate in half even if you don't reduce nm. So less cost making a 2 chip chip than making a single chip. On top being able to get complete product line out of 2 wafers is also a big help.

https://hothardware.com/news/tsmc-5nm-node-doubles-density-amd-ryzen-3000-7nm

So it is possible that AMD for the ryzen 4000 goes down to the 5nm tsmc process for the cpu chiplets.

7nm is cost effective but just as critical on cost chiplet route to reduce failure waste.

Its been awhile since Intel has had to compete from the nm behind position.

[tildearrow]

at 7nm it's same area as 17.5 at 14nm

OK, thanks. Seems like 7nm indeed is very cost-effective.

[castlefox]

Thanks for this long write up phoronix . It mades me feel guilty I am not spending more money to support this website.

[oiaohm]

at 7nm it's same area as 17.5 at 14nm

You also need to divide that in 2 So 8.75 per chiplet if they were 14nm.

Lets do a 14nm Ryzen 7 2700 The half a meg per core of L2 has not changed. So you have 4 megs in that. Plus 16 megs Cache.
8.75Meg!=20Meg Remember those old 14nm Ryzen 7 2700 are single chip construction. So less than half the area in silicon decanted to Cache per core even that it increased in storage a lot.

So in Silicon area 70Meg of cache is still less area than the 20Meg in the prior generations.

Interesting point compared to prior generations of Rizen the Rizen 3000 are running around 2 megs of L3 per cpu thread where the old Rizen were only running 1 meg of L3 per cpu thread. If you look at Intel chips they were running about 1-1.5 per thread.

Its going to be interesting to see how these new chip benchmark out. The prior generation of Rizen for performance were being very dependant on having high speed ram and when you consider running programs built for intel the programs could have been optimised for a larger L3 this could explain this problem.

The ryzen 5 3500G and ryzen 3 3200G chips L3 at 6 being 2 megs of L2 and 4 megs of L3 has the major sniff of performance trouble.

[Dieter]

Wondering why the PSP (spy cpu) firmware hasn't been reverse engineered by now? Would the NSA Ghidra tool be helpful here?

[abott]

I think AMD will enjoy about 2 months of "glory" with this release and then Intel will smack them right back to reality with either Ice Lake or Comet Lake.

That's actually hilarious.

We know they won't have desktop 7nm chips until next year, at the earliest. It's also moronic to think this will be a flop. If AMD's results are even 50% of what they are saying (~13% IPC average, 4.7GHz clocking achievable) Intel is going to hurt. Badly. They will lose the only thing that saved them, having better single thread.
Intel definitely need to be getting any new technology and infrastructure they've been working on out the door ASAP, and on 7nm. They will be hurting badly until the day the mid and high end desktop chips arrive on 7nm.

[pal666]

Exactly, but then please explain, how can a $500 processor have 70MB of cache

at 7nm it's same area as 17.5 at 14nm

[wizard69]

Ok, I guess you're technically correct. We get they've doubled the L3 cache it to help reduce memory latency and therefore help framerates. But the term "gamecache" is still dumb, as if the cache only exists for game workloads.

“Gamecache” will eventually be seen as a marketing mistake. Marketing = a bunch of drunk guys throwing crap at the wall! I still maintain that it sounds like a lot of cache but it really isn’t in this day and age, you really need to think about the number of cores that cache serves.

[Danny3]

FYI: windows 7 will EOL next year, and within a couple more years they will stop supporting it. So prepare to jump ship.

I've been running Windows 7 with disabled updates for 3 years I think, actually from when Windows 10 was released and Microsoft started pushing junk updates to intentionally break Windows 7 so Windows 10 had a chance to beat it at something.
So actually I don't care what Microsoft says or does about Windows 7 support, because I don't trust them and I don't install their updates for a long time anyway.
But still I want good drivers from AMD for it for things like games and video playback with Kodi and MPC-HC, which can use hardware decoding for codecs like H.264 and H.265.
Linux still has to catch up in the gaming and video playback areas, so until then I plan to use Windows 7 as my favorite Windows version for so many reasons in a dual-boot mode with Linux.