True. But 20w difference over the course of a year is <$20 with what I pay for electricity. When a Xeon-D board is $1k and a used "good enough" server is $300, there's a cost-benefit calculation that's hard to ignore.
Which is what I've been trying to say all along.
Fair point, as 20w isn't at all extreme.
But one thing to point out is only the 8-core Xeon D boards are around $1k. I can get a 6-core model for ~$600ish. And in both single and multithreaded applications, it appears the D-1528 often outperforms the E3-1230 v3 and sometimes even the E3-1275 v5.
This page is for the D-1587, but it is conveniently open in another tab and has the processors I mentioned, and others.
And as for 20w difference, here's the thing, what is that getting me? For a $300ish used server, it's likely going to have a E5xxx or X5xxx CPU or two in there.
https://www.cpubenchmark.net/compare.php?cmp[]=2507&cmp[]=1251&cmp[]=2693
That's comparing a D-1540, E3-1230 v5, and E5640.
I wanted to grab a different Xeon D, but unfortunately, no one has ran anything else through Passmark. Single-thread should be similar for the other models, they are all clocked roughly the same and share the same ratio of cache/core count.
Now, the E5640 is impressively not all that far behind the Xeon D when it comes to single-thread, but passmark shows the E3-1230 v5 relatively smoking the D, which is a little surprising as the other benchmarks aren't showing that extreme of a difference between the two (per servethehome.com), but anandtech's comparison of the D-1540 and the E3-1200s shows roughly the same difference between the E3-1240 v3 and the D-1540. It is close to the Low Power variants of the E3-1200 series but of course those much more similar.
But it is also wise to focus upon total Passmark score, and at Anandtech and STH and other sources, the multi-threaded benchmarks. The Westmere-EP gets destroyed in comparison by both the D-1540 and E3-1230 v5.
The way I figure it, if there's going to be a 20W or higher difference, I want the CPU to have a better perf/watt performance as well, or at least in the same league.
The D-1528 should be at right about the same level as the E3-1230 v5 when comparing multi-threaded performance.
Now I've been looking, and frankly the two would end up at right about the same cost for the package. An E3-1230 v5 and comparable C236 motherboard from Supermicro would come out to be just a little bit cheaper than a D-1528 package.
There are differences in the two packages I'd be comparing here:
X10SDV-6C-TLN4F [D-1528]:
up to 128GB ECC RDIMM or 64GB ECC UDIMM
Dual 1000base-T (via Intel i350-AM2)
Dual 10Gbase-T (via SoC)
2x PCIe 3.0 x8 (one in x16), and 1x PCIe x4 (in x8)
m.2 PCIe 3.0 x4 slot (2242/2280)
X11SSH-LN4F [+ E3-1230 v5]:
up to 64GB ECC UDIMM
Quad 1000base-T (via Intel i210-AT)
same PCIe layout
m.2 PCIe 3.0 x2 (2280)
The X11SSi-LN4F is very similar, except ATX as opposed to the above's microATX form factor. It's drops from 8x SATA3 to 6x SATA3, and uses that to make the m.2 slot PCIe 3.0 x4 as opposed to x2 and offers 2242, 2260, 2280, and 22110 lengths. Roughly the same price.
With the D-1528, I wouldn't actually utilize 10GbE, not worried about that anymore, but it would offer the same 4x gigabit LAN connections which is what I am after.
The E3-1230 v5 might offer more processing headroom in single threaded processes, which could be helpful, but I guess outside of VC-1 transcoding, almost everything else in Plex is multithreaded, which is where I expect the most processing use to come from.
And when comparing the E3 to the Westmere chips, the way I see it, I'd be getting a heck of a lot more performance out of the same wattage. So I could go cheap and get something that might function well enough for everything I need, but I'd expect it to need to remain at higher clocks to compute for far longer than the other two in comparison. So a 20W difference at idle and at full load might be what it is, but how long would the one stay at full load when the other chips to get it back down to idle faster, or might not even need to be a full load to begin with? So that 20W difference becomes a wider gulf when you get down to it.
And I expect the wattage difference to be larger when comparing the Westmere to the E3 CPUs that Anandtech benchmarked.
The one pain in the rear that would come with the E3-1230 v5 is xHCI. Not sure how I would address that.