Page 1 of 1

From dual Xeon's to Ryzen

PostPosted: 03/23/2017, 2:34 am
by Farmfield
I documented my dual Xeon build in this thread (link) but 6 months later the dual Xeon setup is sold off and I'm on a Ryzen 1800x. Here's some info, thoughts and comparisons between the two.

So the dual Xeon build served me well, for sure, dual Xeon 2670v2 running at 2.7-3.1Ghz, that's one hell of a performance beast, but subpar single-thread performance was always an issue and as there's just not been any faster (like the E5-2673v2 or 2667v2) Xeon's available on Ebay, when Ryzen was released and I had the option to get a kit cheap through my hardware contacts, I decided to at least try it out. After playing around with it over a weekend, I felt it was stable enough to run. I put my Xeon kit up for sale, had it sold in 2 days, basically getting back what I payed for it.

So here's the reasons in numbers starting with a performance comparison, Ryzen 1800x @ 4.1 Ghz (Dual Xeon setup)


* CPU Multithreaded:

17700 (21500)

* CPU Single-threaded

2180 (1550)

* 3D Mark

13300 (9100)

* Total score

5500 (4500)


* OpenGL

115 (101)

* Multithreaded

1780 (2310)

* Single-threaded

166 (105)

General pro's with each system

* Ryzen

35-50% faster single-threaded performance, 10-15% better RAM performance, 15-50% better 3D performance, more free PCI-e slots, bootable, full speed M.2 slot on the mainboard compared to non-bootable, slower M.2 performance through PCI-e on C600 dual Xeon mainboards, and just overall a more modern platform than the 5 year old Intel C600 based one. :)

* Dual Xeon E5-2670v2

25% higher multithreaded performance, supports [some insane amount] of ECC registered RAM compared to 64Gb max RAM on the Ryzen, no overclocking equals lower CPU temps, I would guess, way better system stability under full load - though not saying Ryzen is unstable, I had no issues to point in that direction, but yeah, Xeon's, ECC RAM, OS/app support, etc, it's a reasonable assumption. ;)


Looking to the Adobe suite, you'll likely never notice the difference between these two systems, though in huge comps, I would guess the dual Xeon box would perform better, especially looking to RAM use. The dual Xeon's has a huge advantage in Nuke as well, it's heavily multithreaded. And as Nuke loves RAM, even not using it, you'll get a performance boost just having 128Gb - no clue why, but it's noticeable, for sure. I didn't compare Fusion, but I'd guess it would be similar to AE and Nuke. So yeah, if I were still working as a comper, I would have stayed on the dual Xeon's.

But in Houdini, the Ryzen system is as like night and day compared to the dual Xeon's. Houdini really suffers from poor single-threaded performance as it jumps a lot between single- and multithreaded computations, so bad single-threaded performance instantly drags down the performance in everything except rendering where it's all about multiple threads. And seeing how 3D applications work, I would guess this would be similar no matter what 3D application you use, there's just a lot that can't be multithreaded in that area, so bad single-thread performance will affect all 3D apps.

So if single threaded performance isn't an issue, for maximum performance rendering, or alike, go with the dual Xeon build. If you want higher single-threaded and better 3D performance, and you have no issues running a heavily overclocked system, the Ryzen is a great alternative to spending a lot more cash on an Intel 6850k/6900k build.

Oh, and don't stress out over the whining about this and that in regard to Ryzen, oh, the RAM isn't quad channel, something something moving data between caches, all that tech stuff, it just doesn't matter because it's not something you'll notice - accept there will be OS and app optimizations making Ryzen 5-15% faster over the next couple of months, you might notice that. But as it stands now, Ryzen will give you a lot of performance and at the price for the CPU's and the mainboards, it's an absolute bargain.

Over and out! :D

Re: From dual Xeon's to Ryzen

PostPosted: 03/23/2017, 7:44 am
by mooviemakers
Thanks for this, very useful info that follows on from the post I was talking to you on the other day. The Ryzen system looks like a great all rounder for production, at a bargain price compared to the nearest Intel alternatives, and it can only get better as the ecosystem matures around it.

Re: From dual Xeon's to Ryzen

PostPosted: 03/23/2017, 12:09 pm
by Dorin
Congratulation Farmfield for the new system. :ugeek:
Motherboard BIOS is not 100% up to speed just yet for AM4 platform, check frequently for new updates that will increase stability and compatibility cu higher frequency RAM. Like the dual Xeon setup, Ryzen also supports ECC RAM, it's up to motherboard if the feature is enabled or not. An example of motherboard that supports ECC Un-buffered is Gigabyte GA-AX370-Gaming K7.

If you are using Linux the kernel 4.10 bings support for Ryzen multicore / multithreading.
If you are using Windows 10 set the power saving options to "high performance" because W10 has some problems distinguishing between real cores or SMT ones and is not always prioritizing the actual cores like in the case of Intel CPUs. Microsoft surpassingly is working to a patch to fix that but do not hold your breath just yet.

Re: From dual Xeon's to Ryzen

PostPosted: 03/23/2017, 12:27 pm
by Farmfield
Hehe, I'm well aware about all of that, I spent a couple of years designing custom CAD/3D workstation, back when I got my first kid, so I've been into the hardware side of this crap for decades now. :D

And I'm guessing 5-15% speed increase with Ryzen from optimized EFI's, chipset drivers, OS tweaks and application support, in the months to come. My highest Cinebench scores so far (4.1Ghz/1.45v) are 1790 on multithreaded and 268 on single-threaded, I'd love to get that up to ~19000 & 180 on 4.1/1.4v, that would be really sweet. :D

Re: From dual Xeon's to Ryzen

PostPosted: 03/24/2017, 5:11 am
by Farmfield
BIOS update from Asus today and a little bump in Cinebench...


The old numbers were 115 (OpenGL), 1780 (Multithreaded) & 166 (Single-threaded).

I'm sure the multithreaded performance will eventually end up 100-150 points above these, but though I would really like to have a boost in single-thread performance, it doesn't feel very likely that'll get that much higher, low to mid 170's at the most, I'd say.

Re: From dual Xeon's to Ryzen

PostPosted: 03/25/2017, 2:51 am
by Tiner
Oh Farmfield oh Farmfield :P What have you done now :P

Re: From dual Xeon's to Ryzen

PostPosted: 03/25/2017, 5:34 am
by Farmfield
Tiner wrote:Oh Farmfield oh Farmfield :P What have you done now :P

I've just continued to do what I always do, all the time. Have fun. :D

And to be honest, picking the stuff up that friday, I was pretty sure the Ryzen, mainboard & RAM would go back to the store after the weekend, but I was amazed how well it performed - and from there I tuned it and with the latest BIOS and system updates, it's pretty friggin amazing, to be honest.

I have 80% of the multithreaded performance of the dual Xeon build and got 75% better single speed performance as well as 25% better GPU performance (in OpenGL, DX, no change in OpenCL/CUDA). In Passmark I've got 17600 total, 2180 single threaded, and the total Passmark score jumped from ~4 500 to ~5 500 points.

And this thing is rock solid, I've had it running OCCT more than once, overnight.

Re: From dual Xeon's to Ryzen

PostPosted: 03/25/2017, 5:46 am
by Farmfield
And here's some details for the hardware nerds... :D

Though I can run this Ryzen 7 1800x on 4 150 MHz @ 1.45 volts, and I can run it on 4 100 @ 1.38125, I settled for running 4 100 MHz @ 1.4v as that seems to give be the best balance between performance and temperatures, running idle at 37-45º C, maxing out at 60-61º C.

I run my Corsair Vengeance LPX 2666 at 2666 MHz, 1.2v, 15-15-15-36, and I seem to be one of the lucky ones (likely my sticks run Samsung "B-die" memory) because many can't even run those at 2400 MHz (likely using some earlier revision of the LPX sticks with Hynix memory).

Now, the latest BIOS updates might fix that for some, but it seems like the best option is making sure to buy memory, whatever the brand, that run Samsung "B-die" memory.

I might try to go above 2666 MHz down the line, but chasing memory performance is likely the most useless of all OC pursuits, because it really doesn't do much for final performance - BIOS, OS and applications updates will, for sure. So not messing about with memory settings, it's just not worth it.