The Ryzen "ThreadRipper"... 16 cores of awesome

[dnavas]

Yeah these are enthusiast desktop units, not commercial servers with dedicated hardware raid, battery backup etc.

True, but great products can be made by bringing features down-market. Seems like there's an awful lot of room on that TR package, so unless they have some kind of huge L4 cache in there, there are a lot of potential possibilities that could really make it a special little device (more than just a 16-core wunderkind). I don't think AMD has the time or energy or money to explore them, but I'd love it if they started....

[That said, I don't personally need RAID support, as I've already got a NAS, but given that Ryzen is actually being put in qnap servers, any support that made raid-6 computationally cheaper ought to be well-received. Oh, oh, and anything that made it easier for qnap to support pcie-based nvme would be beyond awesome. Seriously.]

 

[formulav8]

As you're apparently not allowed

It's typical these days, like Intel wanting you to get a new z270 chipsets this year and someone calling out to the doctor when someone is not feeling quite right about CL possibly not being compatible with your new setup and a new Intel chipset not meeting Coffee Lake standards. Where you expect to do your own thing within moral reason, and its still not considered compatible with this years Kaby Lake platform at this time at least.

 

[SolMiester]

Yeah these are enthusiast desktop units, not commercial servers with dedicated hardware raid, battery backup etc.

Well, that is a fair point, however AMD are not going to make any headway against Intel, if their CPUs/platforms have less features.

 

[beginner99]

Well, that is a fair point, however AMD are not going to make any headway against Intel, if their CPUs/platforms have less features.

Why do you need RAID? It's hardly ever worth the hassle. For storage and backup, there is always software RAID (basically all NAS devices use software RAID and it works just fine).

 

[SolMiester]

Why do you need RAID? It's hardly ever worth the hassle. For storage and backup, there is always software RAID (basically all NAS devices use software RAID and it works just fine).

I would never use software raid, you have to rely on an O/S for it to be available Have you ever tried to recover an array from a dead NAS, or replace a failed drive on one?. I have a media library, hence the Raid requirement.

 

[wahdangun]

I would never use software raid, you have to rely on an O/S for it to be available Have you ever tried to recover an array from a dead NAS, or replace a failed drive on one?. I have a media library, hence the Raid requirement.

yeah that is why i'm still using raid 1 or 10. its simple and more reliable.


I'm just hope TR have RAID 1 NVME available and not locking it behind paywal like intel, considering they have much more pcie line.

 

[itsmydamnation]

Doesnt have that either and not everyone can afford server grade controllers.

I consider 1 or 10 both superior to 6, but seriously no 5 for your data's sake. Why no zfs or btrfs

 

[Schmide]

I have a media library, hence the Raid requirement.

Can you explain why raid 5 instead of 10. 10 seems better in all respects.

 

[Dresdenboy]

https://www.youtube.com/watch?v=QJNRtGo5IMc

some of the Youtube folks are not enjoying Intel's response to Threadripper

As a part time perfectionist I can understand the itch caused by a forced compromise like getting the highest ST performance with also a good MT performance esp. with 256b SIMD together with some limited I/O capabilities.

 

[beginner99]

I would never use software raid, you have to rely on an O/S for it to be available Have you ever tried to recover an array from a dead NAS, or replace a failed drive on one?. I have a media library, hence the Raid requirement.

Have you ever restored a large RAID5? RAID is about availability not backup. It's IMHO not worth the hassle in fact it's more expensive than going just single-drives but with proper backup because you need the backup anyway. That is what I do. Media/documents on single drive. Use synctoy to sync changes to USB drive(s). Store drives at work. Sync can be done once a month. If I loose some stuff in between it doesn't matter much. Can be re-ripped or re-downloaded. If it is important or unique stuff (like photographs), then do the backup immediately.

 

[imported_ats]

All Ryzen cpu's has ECC capabilities. It's up to mobo makers to implement. There are a few that does and persons have verified it's working.

There are ZERO Ryzen CPUs with official ECC support. Any business buying Ryzen and expecting ECC is being extremely short sighted. AMD does not support nor warranty ECC support on any Ryzen.

 

[Puffnstuff]

I can't wait until these new cpu's hit the market and make it into reviewers hands. I really want to see how they perform doing normal things and no matter what their price points should undercut Intel by a large margin. Any one of these high core count models would be a worthy replacement for my old 4790k and with Intel chopping pci-e lanes off their cheaper x299 cpu's that are still overpriced they will be a hard sell.

 

[SolMiester]

Can you explain why raid 5 instead of 10. 10 seems better in all respects.

Of course stripped mirror is better but there are couple of factors in there, 50% space loss, available sata ports, chassis space, and when you build on budget, you cant go out and buy new storage in the one go. Married with kids, you build up your storage. I have backups of course but it is tedious to backup over several drives, and I share my library with family.

I know lots of people bag R5, but in NZ, most SMBs use R5 due to cost of disks, and incremental shadow protect to NAS boxes.

 

[formulav8]

There are ZERO Ryzen CPUs with official ECC support.

That's been known from the beginning.

 

[Atari2600]

That's been known from the beginning.

Indeed.

Sometimes I wonder about what goes on in people's heads.

If AMD were to release an all-singing, all-dancing Ryzen with full ECC support, quad-channel memory of massive capacities, 60 PCIe lanes, fully supported virtualisation and all the other bells and whistles.....

Why would any company ever consider buying the higher priced server chip?

Some folks want to have their cake and eat it too. Never happy.

 

[nathanddrews]

Why would any company ever consider buying the higher priced server chip?

Server density.

As I understand it, though, some motherboards do allow for ECC with Ryzen...? Or is that no longer the case?

 

[Atari2600]

Server density.

Yes, well, I suppose they could restrict blade mobo's to a slight variation of socket.


But, for workstations etc, it'd be cutting yourself off at the knees!

Server
As I understand it, though, some motherboards do allow for ECC with Ryzen...? Or is that no longer the case?

Yes, ECC exists on the CPU and runs on some boards - but is not officially supported in any way shape or form.

 

[Topweasel]

Indeed.

Sometimes I wonder about what goes on in people's heads.

If AMD were to release an all-singing, all-dancing Ryzen with full ECC support, quad-channel memory of massive capacities, 60 PCIe lanes, fully supported virtualisation and all the other bells and whistles.....

Why would any company ever consider buying the higher priced server chip?

Some folks want to have their cake and eat it too. Never happy.

Look at the lineup. I don't think AMD has any intention of overlapping Ryzen, Ryzen ThreadRipper, or EPYC. So realistically they can all share the same tech including ECC and it wouldn't matter. If you need a 8 core basic CPU, get Ryzen, if you want high end workstation/desktop or low cost server, get ThreadRipper, if you want the balls to walls best 2P configuration you get EPYC.

Zens greatest asset is greatly scalable cores off of a single die and AMD is riding it all the way up the ladder. At this point, maybe not in the future, a 1 EPYC is as good as selling 2 ThreadRippers, 1 ThreadRipper is as good as selling 2 Ryzen's. So I am not AMD cares as much if one feature set gets overused in a market it wasn't really intended if it gets them to buy Zeppelin dies.

 

[kalmquist]

T1beriu said: ↑
Here you go. Ryzen and ECC.

LOL have you actually read it? I see this:

(quoted text deleted)

That official BIOSTAR website proves nothing. What is interesting if somebody did investigate it on BIOSTAR and found it to be fully functional unlike on the X370 Taichi.

You read the linked article, but not the comments. The writer of that article was mistaken about one thing. He was expecting the system to halt when an uncorrectable error occurs, but that is overkill unless the error happens when reading kernel code or data structures that cannot be reconstructed. If it happens inside a user process, the operating system should, at most, terminate that one process.

In summary. ECC RAM appears to work just fine under Linux. Windows has problems with reporting on ECC memory. (This is not limited to Ryzen; it also applies to Intel Xeon E3 v4 and E3 v5 processors.) The BIOS options for ECC memory are quite limited (or at least were in March), but most users don't have any reason to tweak these anyway. So the linked article doesn't identify anything that I would consider a serious problem with ECC memory on Ryzen.

 

[wildhorse2k]

You read the linked article, but not the comments. The writer of that article was mistaken about one thing. He was expecting the system to halt when an uncorrectable error occurs, but that is overkill unless the error happens when reading kernel code or data structures that cannot be reconstructed. If it happens inside a user process, the operating system should, at most, terminate that one process.

In summary. ECC RAM appears to work just fine under Linux.

How did you come to that conclusion? Is there any article that proves it? That linked article doesn't. No process termination was seen. Unless you have an article that demonstrates that.

 

[itsmydamnation]

How did you come to that conclusion? Is there any article that proves it? That linked article doesn't. No process termination was seen. Unless you have an article that demonstrates that.

Yesterday i had unrecoverable ECC errors on a Xeon V4 (in production), nothing got terminated there either...........

edit: You should go and find Linus rants about the lack of ECC consistency on RWT, it was a few years ago now but it was quite interesting.

 

[wildhorse2k]

Yesterday i had unrecoverable ECC errors on a Xeon V4 (in production), nothing got terminated there either...........

And you consider that to be working ECC?

 

[itsmydamnation]

And you consider that to be working ECC?

Yes,

You have to understand what ECC buys you and what you want out of it. Just knowing about a corruption normally makes its fine for the vast majority of use cases.

 

[Ajay]

Yes,

You have to understand what ECC buys you and what you want out of it. Just knowing about a corruption normally makes its fine for the vast majority of use cases.

I have something like 10K work hours of use with systems with ECC Dram (plus, we never shut them off)

The only time I had an un-recoverable ECC error, was on a Sparc workstation a long time ago (I actually collected four over time). Just shutdown my system (damn, only 180 days of up-time), replaced one of the sticks and booted back up. When you are checking code in an out on multi-million line mission critical firmware - memory errors can't be tolerated. Also, all the Sparcs/Ultrasparcs in our group were used to build our releases as part of a networked build system - another reason for ECC.

Some of those systems had huge amounts of Dram (like 1/2 GB in 8 slots - LOL). I wonder what kind of error rates exists today in workstations with 64 GB+ of Dram?

 

[formulav8]

And you consider that to be working ECC?

Many seems to buy ECC for the soft single-bit error recovery feature. Which Ryzen has shown to do fine even on non-official boards.