D-1541 vs D-1557 - Windows Server 2016 Hyper-V

Rudde93

Limp Gawd
Joined
Nov 19, 2010
Messages
137
Hello!

I'm planning to build a small little NAS running Windows Server 2016 and trying to use the Hyper-V hypervisor. Planning to run ReFS and Storage Spaces for storage and was considering running maybe one Windows VM in addition to 1-2 linux VMs.

And I'm currently down to either D-1541 or D-1557 CPU's.

D-1541 is 8 core 16 thread 2.1 GHz (2.7 GHz turbo) and the D-1557 is a 12 core 24 thread 1.5 GHz (2.1 GHz Turbo) CPU.

So I'm wondering how much of this turbo performance I will actually be able to utilize? And will the 1.5 GHz actually be so low that it's not worth the extra cores since my tasks will be slower? Will my system be any use at 1.5 GHz or should I go with the high clock speed D-1541?
 

evt

Limp Gawd
Joined
Apr 6, 2006
Messages
253
Turbo is only for full load single-core usage. Once you load up the other cores, then the CPU speed will drop down to its normal rated frequency.
Since you are going to use it for VM purposes, you will probably never see the turbo speeds.
 

Rudde93

Limp Gawd
Joined
Nov 19, 2010
Messages
137
So what should I take into consideration using it for my wanted setup? Should 1.5 GHz enough core clock, or is that risky slow?
 

evt

Limp Gawd
Joined
Apr 6, 2006
Messages
253
3l5kOU7.png


I do like the computer power density in the size (roughly) of a NUC.
 

PigLover

[H]ard|Gawd
Joined
Jul 11, 2009
Messages
1,183
The question cannot be answered generically. It depends on your workload, i.e., depends entirely on what the "1 or 2 VMs" will be running and how effectively their workload is threaded.

Either processor will handle the baseline NAS workload just fine - you don't need a Xeon-D at all to handle that part. Lacking any other information, and based on the assumption that most things you might run are not threaded very well, I'd go with the "fewer/faster" core option D-1541. But YMMV and without a lot more detail on what you are planning you can't expect a better answer.
 

DrLobotomy

Supreme [H]ardness
Joined
May 19, 2016
Messages
6,736
Buy both. Test them out. Send back the loser or use it as a router.
 

Phantum

[H]ard|Gawd
Joined
Jul 25, 2001
Messages
1,716
Turbo is only for full load single-core usage. Once you load up the other cores, then the CPU speed will drop down to its normal rated frequency.
Since you are going to use it for VM purposes, you will probably never see the turbo speeds.

I was under the impression that Intel did some re-working of the turbo feature in Brodewell-DE to allow for the turbo'ing of multiple cores while under load...? It was my understanding that it wouldn't be all cores but certainly more than one as it has been in the past.
 

Shadohh

Gawd
Joined
Mar 13, 2012
Messages
818
I was under the impression that Intel did some re-working of the turbo feature in Brodewell-DE to allow for the turbo'ing of multiple cores while under load...? It was my understanding that it wouldn't be all cores but certainly more than one as it has been in the past.

My understanding is that if base is 3ghz on a quad core and turbo is 3.6ghz on a 1 core, then 2 cores might turbo to 3.4ghz and 3 cores might turbo to 3.2ghz. I do not know what models if any support this.
 

PigLover

[H]ard|Gawd
Joined
Jul 11, 2009
Messages
1,183
Broadwell-DE (D-15xx) supports Turbo Boost 2.0, which manages turbo speeds base on thermal and power characteristics measured by sensors in the chip. This allows, potentially, all cores to reach their turbo max simultaneously. YMMV, depending on actual operating conditions, but with lower power CPUs (like the D-series Xeons) and superior cooling solutions you actually can achieve full turbo of all cores at least for brief periods of time.

Of course, most of the boards in the wild (like SM's X10SDV series) have pretty marginal cooling solutions attached. Oh well.
 

Phantum

[H]ard|Gawd
Joined
Jul 25, 2001
Messages
1,716
Broadwell-DE (D-15xx) supports Turbo Boost 2.0, which manages turbo speeds base on thermal and power characteristics measured by sensors in the chip. This allows, potentially, all cores to reach their turbo max simultaneously. YMMV, depending on actual operating conditions, but with lower power CPUs (like the D-series Xeons) and superior cooling solutions you actually can achieve full turbo of all cores at least for brief periods of time.

Of course, most of the boards in the wild (like SM's X10SDV series) have pretty marginal cooling solutions attached. Oh well.

Yeah, I believe you are correct, it's coming back to me now. Apparently I interpreted the info wildly different from what I actually read.
 

Rudde93

Limp Gawd
Joined
Nov 19, 2010
Messages
137
Hey guys, I had some economical issued and could not go trough with this at the time I tough. I've had some time in-between and really had to pull myself down to a C2750 level, so I'm not looking at the C2750 or considering if the G4500 could be a better option? It scores better on static tests even tho it only have 2 cores and C2750 have 8. Same application, would it be better to pack the performance in less cores?
 

sinisterDei

[H]ard|Gawd
Joined
Dec 1, 2004
Messages
1,521
I use the X10SDV-TLN4F (which has the D-1541 on it) all the time. It's a good little performer, and typically I don't think the higher core count chips/boards are worth the cost investment. I think it breaks the value proposal and building a 'real' single-socket Xeon E5 system starts to be worth it.

That said, you have to be careful buying Supermicro's X10SDV line- they don't all include *fans* on the damn heatsinks. The TLN4F does, and any that have a + in the model number (like the X10SDV-4C+-TLN4F) include fans, but most of the rest do not. I once made the mistake of purchasing a X10SDV-4C-TLN4F and had to strap (using zip ties) a small fan onto the heatsink to prevent horrible thermal throttling.
 

sinisterDei

[H]ard|Gawd
Joined
Dec 1, 2004
Messages
1,521
As to regarding the C2750 vs the G4500, it really depends on your workload.

The G4500 will be significantly faster on single threaded workloads - like *way* faster - and the CPU has competent integrated video. Lots of people multi-purpose their NAS devices as things like PLEX servers and such, and the G4500 will be much more suited to that than the C2750. If you plan on presenting iSCSI storage LUNs to other systems or anything fancy like that, something like the G4500 would likely have higher throughput in an environment with a small number of connected devices.

That said, the C2750 is lower power and 'wider'. I like the analogy of "wide versus tall" when comparing systems with a smaller number of faster cores versus a system with a larger number of slower cores. The C2750 won't do anything as fast as the G4500, but 8 cores is a *lot* compared to the 2 on the G4500, so it's possible it could overall handle more processes concurrently than the G4500 could.

Neither of them is a great VM host, but either CPU would probably be OK for a device primarily used as a NAS.

There might be other considerations - typically the C2570 being 'server' CPUs are placed on boards with AST BMCs that allow remote IPMI KVM access, as well as multiple NICs and things like that. For the most part, no G4500 board will ever offer those features.

The G4500 would also be upgradable to a real i3 or i5 later down the line if you want, whereas most the Avoton CPUs are integrated onto their motherboards.

So it really just depends which of all of those things you place value on. Good luck!
 

lopoetve

Fully [H]
Joined
Oct 11, 2001
Messages
32,577
Hey guys, I had some economical issued and could not go trough with this at the time I tough. I've had some time in-between and really had to pull myself down to a C2750 level, so I'm not looking at the C2750 or considering if the G4500 could be a better option? It scores better on static tests even tho it only have 2 cores and C2750 have 8. Same application, would it be better to pack the performance in less cores?
I've got a set of the 2750/2550s. Great for being low power and low cooling, but IPC is pretty low on them. Any significantly interactive workload seems to suffer a bit, or anything compiling or doing major transcoding, but they're great for general workloads.
 
Top