Ryzen 5/7 as a Xeon E3 Competitor?

[matt_garman]

I know AMD is marketing their new Ryzen 7 and 5 chips mainly towards enthusiasts and gamers. But, given the price and performance, and ECC support, it seems like a good competitor for single-socket Xeons (E3 series) and arguably Xeon-D. I'm waiting for a motherboard vendor like Supermicro or ASRock Rack to release a motherboard with features similar to single processor Xeon or Xeon-D motherboards: onboard video, full-featured IPMI, quality Intel networking, at least six SATA ports, etc.

(Based on this excellent article, ECC Memory Tested Working On Ryzen...Sort of., I understand the Ryzen ECC support isn't quite fully baked. But it sounds like maybe only a bit of polish is needed to get the ECC support where it needs to be.)

I'm just thinking this would make a killer home server at a reasonable price. Where I'm coming from now: I've been using an ASRock E3C224D2I with an E3-1230v3 for a couple years now (build writeup w/pics). It's been working great, no complaints. But now I'm into video surveillance with Zoneminder, which (currently) a bit of a CPU hog. I'm currently running Zoneminder on a separate dedicated system. But I could consolidate ZM and my main server into one if I had an 8-core CPU. Also, the ASRock E3C224D2I motherboard I'm using currently has a memory limit of 16 GB. I'd like to at least double that if I consolidate systems. Xeon-D would be great here, but (1) it's expensive, and (2) looks like you're into warranty-voiding territory if you want to use a big aftermarket heatsink.

Mostly just thinking out loud here. Does anyone else have similar sentiments?

 

[ttabbal]

My thought is that the current Ryzen are not a good replacement. They could be, but not yet.

First is that ECC support is "kinda". AMD officially doesn't support it, but doesn't actually disable it either. I guess that's something, but I don't really think of that as competing with a server part.

The second is BIOS support is all over the place. There's no way to know, even between versions on the same board, if it will work. And because it's not officially supported, the motherboard manufacturers are under no obligation to keep it working.

I also don't like that the system doesn't seem to halt on an uncorrectable error. That's kind of a big deal... Sure, that might be fixable in software, but will it be?

And there's the fact that you need rarer, more expensive, unbuffered DIMMS. I don't want single use RAM. It's expensive enough that I want to be able to use it across all systems when I move to DDR4. I believe this might not be software fixable, I think you need hardware support for buffered RAM.

My first dedicated server build was AMD/ECC based in a similar situation, but with a little better RAM and ECC support as the platform had matured a bit before I did it. I used a gamer board and ECC UDIMM. It worked, but for my next build I went with server gear and Intel because I don't have to guess if it works, it's proven to work. AMD can get there, and the server oriented parts are coming soon.

Ryzen are good for what AMD is marketing them toward, the desktop performance and value crowd. I think the info we have on the server lineup is very interesting and I hope they do well.

 

[Patrick]

I think @ttabbal has the gist as well.

Here is my one differentiation. I put a different spin on this for a home server v. something in a remote data center.

If you have a Xeon E3 "server" that is essentially a heavy compute engine for your laptop, Ryzen will be fine if you can just walk over and hit reboot. If you get 2x simulation speeds or compile speeds at the cost of doing a monthly reboot that is fine. In the data center, it costs us about $300 per machine to get remote power and iKVM with remote media that functions somewhat OK.

Stability on server OSes, and support in some cases is not there. The server platform makers I expect are working on Naples as an AMD platform not Ryzen at this time. Ryzen BIOS code still needs work and Naples is the bigger server offering. ECC memory is not a consideration at this point since you are going to likely see a kernel panic many times before you actually catch an ECC error.

Finally, the fact that you need a GPU means it may be OK for machine learning, but by the time you add in GPU cost, server boards have more expandability and much lower power consumption.

I think there is clearly a market for Ryzen servers, I am just not ready to recommend that anyone puts these in the data center.

 

[matt_garman]

I agree with both of you, Ryzen is likely not quite ready for this role.

It's frustrating though, because it looks like all the pieces are there for a killer single-socket server chip. I suppose the necessary "a little more polish" is easier said than done. (And my colleague had to remind me, they've only been released for about a month, so maybe there's a chance they'll get there in time.)

I'm definitely excited about Naples, although 32 cores might be overkill for home. Of course, I also love overbuilding things.

 

[Patrick]

They will get there in a few more months.

BTW last 2 CPUs are being benchmarked then will have the entire E3-1200 v6 line complete.

I also bought the i7-7700k, i7-7700 and a few of the Kaby Lake i5s to build out the STH dataset.

 

[trippehh]

E3 competitor? If ECC gets sorted, which I think it probably will, it certainly could replace Xeon E3 and lower end S1 E5 workstations easily.

On the server side some vendor would have to come out with some server-grade boards, but the chip seems capable enough. Less likely to happen but would be interesting.

My thought is that the current Ryzen are not a good replacement. They could be, but not yet.

First is that ECC support is "kinda". AMD officially doesn't support it, but doesn't actually disable it either. I guess that's something, but I don't really think of that as competing with a server part.

The second is BIOS support is all over the place. There's no way to know, even between versions on the same board, if it will work. And because it's not officially supported, the motherboard manufacturers are under no obligation to keep it working.

Some of the vendors lists it as supported. But operating system support is not quite there yet so listing it now is a little.. lets say brave. At least on these boards it should remain not deactivated.

I also don't like that the system doesn't seem to halt on an uncorrectable error. That's kind of a big deal... Sure, that might be fixable in software, but will it be?

Halt on unrecoverable error is something that is controlled from the operating system/drivers, even on the Xeons. Given that it does actually have a working EDAC it is most likely just a matter of full software support for it landing. I would expect this to happen with the Naples roll out.

And there's the fact that you need rarer, more expensive, unbuffered DIMMS. I don't want single use RAM. It's expensive enough that I want to be able to use it across all systems when I move to DDR4. I believe this might not be software fixable, I think you need hardware support for buffered RAM.

It is the same ECC memory used by Xeon E3.

 

[Aluminum]

E3 can have vpro if implemented correctly, so they win on the datacenter/remote column.

But for workstation/local server? Ryzen is literally exactly like E3 xeons (unbuffered, bios dependent etc) but more bang for your buck.
Depending on your build, it bitchslaps 1P E5 systems pretty hard. As for support, buy smart, buy asrock.

Meanwhile intel just got done with yet another "pray we don't segment any further" by locking down E3 and desktop chipsets. Getting real tired of their bullcrap.

 

[Patrick]

I had this discussion today. The E3 line is marketed as a workstation chip as well.

From a server chip perspective, E3 hands down right now.

From a workstation perspective, the Ryzen is starting to get better.

And as a bit of background, I just made charts for every E3-1200 V6 minus the 1275 and 1245 (since CPU wise they are the same as the 12x0 parts), the Core i7-7700K, i7-7700 and Core i5-7400. In those charts, I also have every Ryzen SKU released to date.

 

[T_Minus]

@Patrick when is that going to be published? looking forward to seeing that before I drop any $$ on a V6 or new desktop i7 for myself

 

[zir_blazer]

I think that Ryzen is a bit lacking in features to be a true Xeon E3 platform replacement. You have the CPU horsepower, but it is missing ECC and remote management (Either vPro or IPMI). I don't know if a Motherboard maker could actually put a BMC on the current Ryzen platform, and make sure that ECC is tested and works. At that point, it would be functionally equivalent.
BTW, didn't AMD had DASH which was supposed to be the AMD version of vPro? What happened to it?

Until Naples prelaunch data comes out, including the different Processor models price and features, plus the Motherboard cost, it will be hard to know if buying a Ryzen now makes sense. I still believe that AMD will release Workstation parts similar to the Ryzen 7 1700 (8C/16T) that are not much more expensive, since they will want something to compete with the Xeon E3 and E5 1600. Besides, in the past AMD did the same thing as Intel and used a superset of the consumer platform for their Opteron 1xx line (Remember the famous Socket 939 Opterons? Well, they were like the Xeon E3, intended for Unbuffered+ECC RAM).
Intel got a bit of overlapping with the Xeon E3-1230v6 and the Xeon E5-1620v4, so similar overlaps could happen in AMD lineup (During the brief time that Haswell-E and Haswell were the current generation of their respective lines, the 1620v3 made much more sense than the 1230v3, but total cost was also much more expensive due to DDR4).

 

[Patrick]

@Patrick when is that going to be published? looking forward to seeing that before I drop any $$ on a V6 or new desktop i7 for myself

Oh sorry, I forgot. It is live.

@zir_blazer is is possible, but the question is whether it is economically viable. Remember, E3 is a low volume SKU.

 

[i386]

I don't know if a Motherboard maker could actually put a BMC on the current Ryzen platform

Technically it's possible. But for gaming systems and "workstations" (thinking of windows equivalents to apples mac pro) it's not necessary and would just add costs for the manufacturer.

 

[lbjm]

@Patrick when is that going to be published? looking forward to seeing that before I drop any $$ on a V6 or new desktop i7 for myself

Even if the numbers worked in Ryzen's favor I'm still hard pressed to let 1st gen Ryzen touch my data.

 

[cheezehead]

If I needed an editing/cad machine (ie any heavily threaded workloads) where normally I would get a high end i7, E3, or some E5 setups... Ryzen makes perfect sense. Outside of that I wouldn't touch it for that end of the spectrum.

 

[MiniKnight]

Based off what exactly?

I'm waiting for June 20. AMD's gotta have OS compatibility by then to sell Zen to server guys. That'll make Ryzen a better server platform.

 

[lbjm]

Based off what exactly?

Prior experiences with Opterons (1st gen mind you) not playing well with a database I was using at the time. Complete data corruption teaches one the hard way that even switches arches is cause for a backup. Remember the Ryzen thread scheduler patch?
Microsoft acknowledges AMD Ryzen performance issues on Windows 10, fix incoming

bleeding edge is NOT a joke.

You can google "ryzen windows thread scheduler"

That's just a symptom that they didn't learn from their last go around of a new arch.

I'm waiting for June 20. AMD's gotta have OS compatibility by then to sell Zen to server guys. That'll make Ryzen a better server platform.

I don't know one would think they'd spot the thread scheduler problem before it reached the end user. I'm sure the Linux guys have things well in hand, but the bigger question: Are the database and VM makers ready?

 

[William]

Plus they need to really work on the BIOS's on these boards, I had so many issues it wasn't funny. Maybe they have fixed that with current versions, I don't know.

 

[lbjm]

Every new platform or architecture has issues. With the latest updates I can say that it's quite a stable platform. I think all those focusing on AMD's issues need to remember that whenever a new architecture was launched there was issues (regardless of Intel or AMD). Take Nehalem for example even with simple things such as memory compatibility. Ryzen is AMD's Nehalem moment in my opinion. Even X99 and X79 had issues upon launch so it's not uncommon no matter who the company is.

Well Sandy-bridge-E gave me issues with an Asus Z9PE-d8 motherboard. That was an Asus issue and their lesser brand ASrock seems to produce better motherboards bioses.. This why I let others become the earlier adopter now ;D

My problem with AMD stems from the fact that I know they are slow to fix problems whether CPU or GPU because they purposely don't put enough funding into the development side of R&D. They choose to burn all their money on Execs. You can look up their R&D vs exec pay in their public FEC filings. I'm not willing to dump money into AMD anymore until they return to the level where they were at back in the Athlon days.

 

[alucasa]

I dislike how AMD lies about their TDP.

AMD CPU TDP is based on average load while Intel's TDP is max load.

So, when AMD CPU says 65w TDP, it is effectively 95w TDP.

 

[Evan]

I dislike how AMD lies about their TDP.

AMD CPU TDP is based on average load while Intel's TDP is max load.

So, when AMD CPU says 65w TDP, it is effectively 95w TDP.

So AMD TDP is more like Intel SDP

On the topic of TDP I am surprised how high the low core count parts are these days, certainly ghz plays a huge role in TDP.