Was about to switch from Intel>Ryzen, but now I'm not so sure after all!

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

MrCalvin

IT consultant, Denmark
Aug 22, 2016
87
15
8
51
Denmark
www.wit.dk
It was my belief that AMD's newer CPUs was ahead Intel now, and I finally decided to get my first test-server, Asrock board + Ryzen 5 2600 (no GPU) to test it out.
My use-case is std. business servers (Linux KVM/QEMU: VPN firewall, Windows AD, Windows RDP VMs, Storage, Web, Mail, FTP and the like).
Today I use Supermicro X11 boards + i3-7100 (yes, this CPU rocks and are up for the taske without any problems whatsoever!). Beside the low-price I also get some crazy power-consumptions (full-system, idle): With one NVMe as low as 14watt, with 2xNVMe+4xHDD in RAID between 19-25Watt.

But then I ran into this review comparing power-consumption and it seems I will never see these watt numbers with the Ryzen setup. I get this out of the figures:
In idle: Ryzen use 12w more (but the idle TPD of this CPU might be more than my whole i3 system)
Peak, single-threaded: Ryzen use 24w more
Peak, Multi-thread: Ryzen use 73W more (well, Ryzen also have 6(12) cores compared to i3's 2(4))
Smell like old days (AMD had hopeless power-consumptions).
Well, statistic can be turned in any way you want, and the last figure in the review show the efficiency, power compared to the actually work you get done by the CPU (lower is better):
Single-thread: i3-7100: 12.2kj, Ryzen 22.6kj
Multi-thread: i3-7100: 7.1kj, Ryzen 5.1kj (I assume this must be per core or thread, but now we start to see something)

Is this really true? If so it guess I continue to buy Intel systems after all.
Other considerations: AMD is more secure, especially in virtualization. But Ryzen don't support ECC (officially)...don't know how it is IRL. I know people say it runs fine, but does it actually support the ECC feature or just run the memory like any non-ECC RAM!?
I know I could aim for a Ryzen 3 xxxx with 4(4) cores and no GPU but I couldn't find it in retail channel.
 

alex_stief

Well-Known Member
May 31, 2016
884
312
63
38
Yes, idle power consumption is not the strong suit of Ryzen CPUs. If you don't need many threads, and idle power consumption is so important to you, there is no need to switch CPU manufacturers.

but does it actually support the ECC feature or just run the memory like any non-ECC RAM!?
That would not really be a selling point, since most systems without ECC support can run ECC memory just fine.

I know I could aim for a Ryzen 3 xxxx with 4(4) cores and no GPU but I couldn't find it in retail channel.
There are none. AMD serves the low-end market with last-gen chips. Even the Ryzen 3 3200G and 3400G are not Zen 2.
 

T_Minus

Build. Break. Fix. Repeat
Feb 15, 2015
7,625
2,043
113
"With one NVMe as low as 14watt, with 2xNVMe+4xHDD in RAID between 19-25Watt."
Nope. No way. There's no way 4x spinning HDDS aren't using ~11-14w at idle alone, and your entire other system with 2x nvme is not running on 5-10w. The margin for error in power readings that low with a cheap device is pretty high. I'd say you're more like 30-45w, assuming when you say RAID you don't mean HW raid.


With that said, I went from intel to amd and my idle is WAY UP. So much so that I've now enabled sleep or I power off at night to even out the power usage. I just wanted to try AMD, I can't tell the difference (which is nice), but I can see the power increase so if I could send them back and swap back I would do it for the lower idle, and piece of mind of saving $ over the years I'll keep this system .
 

MrCalvin

IT consultant, Denmark
Aug 22, 2016
87
15
8
51
Denmark
www.wit.dk
I believe those power number are correct. First of all NVMe use mW in idle.
Secondly, I must admit I use 2.5" HDDs: Seagate BarraCuda Pro or Seagate Enterprise Capacity (although the later is getting to expensive compared to SSDs, so I'll not get any more of those).
Seagate BarraCuda Pro in RAID deliver sufficient performance for 1GB LAN storage.

And to @alex_stief:
"....and idle power consumption is so important to you", I sure hope it's important to everyone.
I didn't know ECC modules will run fine on any system. I guess that confirm my suspicion, that using ECC with Ryzen (and e.g. Asrock server-boards with chipset x470) you'll not getting ECC. I fear many people don't know this.
 

RageBone

Active Member
Jul 11, 2017
617
159
43
I guess that confirm my suspicion, that using ECC with Ryzen (and e.g. Asrock server-boards with chipset x470) you'll not getting ECC. I fear many people don't know this.
I believe that That suspicion is untrue!
Since there is a thread on here and Level1Techs about that exact board and setup, that would have been noticed.
I have touched a X470D4U which had problems with various things including throwing ECC Errors.
Did memtest a defective stick with ecc errors on a 2700x and a asus crosshair 6 hero that does not officially support ecc errors.

Asrack could not state actual udimm ecc support if the setup wouldn't do ecc.
They additionally state that ECC on Picasso and Raven Ridge is only supported on PRO CPUs.
Those are the ones with the integrated Vega chips. Those are also indeed one generation behind in technology.

Yes Intel is pretty good in "low" power consumption, and IPC is also a good point for Kabbylake vs Zen+

I assume that things do get better in both terms if you were to go Zen2 with Ryzen 3k.

But lets just be aware that it's kinda hard to benchmark and compare hardware and that there are many many fricken factors here at play, including Bios settings. Those can destroy any CPU in any Benchmark depending on how good or bad the settings are.
Not to ignore features on the Board like an ipmi, that will take about 6W i assume, so depending on your current setup which might not have one, that makes things harder to compare.

For me, Techpowerup isn't providing nearly enough information on the test setup and such things, for me to not doubt the accuracy, especially with the idle power consumption.
Fans and coolers used, Bios settings, the actual specific 1080ti model used, the Powerplan in windows, the god damn details!
 

alex_stief

Well-Known Member
May 31, 2016
884
312
63
38
"....and idle power consumption is so important to you", I sure hope it's important to everyone.
Why would it be? I can't speak for everyone, but I personally don't care too much about idle.
It depends on the use-case. Sure, if you build a server that runs 24/7 and is mostly idle, it is important. But this is not the only use-case for PCs, and certainly not for most Ryzen-based systems.
Don't assume that your use-case is representative for everyone else.
 

T_Minus

Build. Break. Fix. Repeat
Feb 15, 2015
7,625
2,043
113
What's your definition of "Idle" I guess is the real question then?
To me idle is sitting in windows moving the mouse around, the drives are not technically at 100% idle when this is occurring.

The 2.5" HDD seem to use much less power so that's nice for sure.
 

EffrafaxOfWug

Radioactive Member
Feb 12, 2015
1,394
511
113
One thing to bear in mind from those benches is that they're all using the X570 chipset which is, to put it mildly, a power hog (rated for 15W TDP and seemingly little in the way of power saving); the X470/B450 chipsets have significantly lower idle draw. Sadly TPU didn't do power readings when they tested Zen2 performance on these older chipsets. My own X470D4U with a 3800X, 64GB of RAM and two P4101 NVME drives idled at 23-24W.

I've not quantified the power measurements since the system went live (it's plugged in to a UPS now and I don't have a power meter that works with C13s) but the load on the UPS dropped considerably, from about 30% to 20% (there's other stuff plugged in to the UPS too so this is very much a finger-in-the-air figure but indicates the idle load is considerably lower than the haswell system that preceded it).

Ryzen efficiency (perf per watt) under load is, quite simply, the best available right now for most workloads. Idle load is in the same ballpark as Intel as long as you avoid the X570 chipsets. Hopefully this year there'll be some less rushed and more power-efficient chipsets landing whilst hopefully still providing PCIe 4.0.

(As an aside, I've even been inspired in to trying underclocking on my workstation; reducing the voltage to my workstation chip has resulted in not only lower temperatures and power draw but also, paradoxically enough, increased performance - I suspect the chip itself is a fairly good one, and the undervolt means less heat allowing higher/longer boost clocks)
 

T_Minus

Build. Break. Fix. Repeat
Feb 15, 2015
7,625
2,043
113
One thing to bear in mind from those benches is that they're all using the X570 chipset which is, to put it mildly, a power hog (rated for 15W TDP and seemingly little in the way of power saving); the X470/B450 chipsets have significantly lower idle draw. Sadly TPU didn't do power readings when they tested Zen2 performance on these older chipsets. My own X470D4U with a 3800X, 64GB of RAM and two P4101 NVME drives idled at 23-24W.

I've not quantified the power measurements since the system went live (it's plugged in to a UPS now and I don't have a power meter that works with C13s) but the load on the UPS dropped considerably, from about 30% to 20% (there's other stuff plugged in to the UPS too so this is very much a finger-in-the-air figure but indicates the idle load is considerably lower than the haswell system that preceded it).

Ryzen efficiency (perf per watt) under load is, quite simply, the best available right now for most workloads. Idle load is in the same ballpark as Intel as long as you avoid the X570 chipsets. Hopefully this year there'll be some less rushed and more power-efficient chipsets landing whilst hopefully still providing PCIe 4.0.

(As an aside, I've even been inspired in to trying underclocking on my workstation; reducing the voltage to my workstation chip has resulted in not only lower temperatures and power draw but also, paradoxically enough, increased performance - I suspect the chip itself is a fairly good one, and the undervolt means less heat allowing higher/longer boost clocks)
That's good to know, and I should note I'm using X570 :D