ZFS Athlon X2 vs. X3

hotzen

Limp Gawd
Joined
Jan 29, 2011
Messages
349
Hello

what will ZFS prefer?
AMD Athlon II X2 250e, 2x 3.00GHz vs. AMD Athlon II X3 415e, 3x 2.50GHz
I really want to encrypt my data using Solaris11 ZFS-encryption or BSD GELI.

Does BSD/Solaris like the AMD 785G/SB71 chipset? I assume an Atheros AR8121 Gigabit NIC is just fine as well.
There is a very nice industrial motherboard available: http://www.sapphiretech.com/presentation/product/?psn=0001&pid=319

Any opinions?
 
Can't speak for Solaris, but the triple core would be better for BSD GELI encryption.

Just note; if you're looking for low idle power than a 95W CPU may be more power efficient than 45W or 35W CPU. Faster CPU would be finished quicker and return to low idle power state, so overall it may use less power.

Core i3 with AESNI also should deserve your attention. Not sure i particularly like that board, but yes at least the chipset SATA shouldn't be a problem. But SB700 family is old; the new SB850 would be better since it has native 6Gbps SATA and uses newest chip.

Whether the wireless works is always a 'wait and see'. I can imagine support under Solaris being much poorer compared to BSD; generally BSD has better hardware compatibility (still worse than Linux).
 
Also to add to sub.mesa's post, Intel and AMD chips are not able to be compared on a TDP basis since the 32nm Intel Clarkdale and Sandy Bridge parts have integrated GPUs, and all of those transistors that are not stressed in servers still need to fit in the same TDP.

Also, AES-NI uses less power than the standard x86 path... but if you are set on the motherboard then I would go with the x3.
 
Besides TDP is an extremely rough measure of power consumption; it's a thermal specification not a power consumption one.

Take the new Sandy Bridge CPUs which are 95W TDP, but only use 65W when SMP-stressed. In idle these CPUs are extremely efficient, but of course expensive and the chipset has a bug which complicates this choice.

Since AMD is quite good in AES compression, you may not need AESNI. But of course when using the encryption this will tax the CPU. Though i think you will be fine. I estimate about 100MB/s per core that your CPU can do. That should be enough but know that ZFS uses transaction groups, so when flushing the txg you will internally write at max speed. This will cause the encryption to 'suck up' any available CPU cycles, meaning 100% CPU usage on all cores for a second or so. This may cause a small 'lockup' for a small moment. This may be slightly inconvenient, so of course AESNI is better. Still given your budget constraints i think the AMD approach is a reasonable alternative that should work decently well. If you want you can force GELI encryption to use a maximum of 2 cores on your 3core CPU. A low-end quadcore is the best choice generally. Bulldozer with AESNI-clone would be a very interesting choice too, but may be available in Q3 so think October or something.
 
I have an Athlon 2 with two cores @2.9ghz running ZFS encrypted with GELI(AES-128) and gzip-9 compression.

This is my stats with dd
dd if=/dev/random of=/tank/file bs=1M count=1024
1024+0 records in
1024+0 records out
1073741824 bytes transferred in 37.623134 secs (28539404 bytes/sec)

Without compression = off I get
1073741824 bytes transferred in 18.083226 secs (59377780 bytes/sec)

(This is on a 9 disk(500gb 7SATA and 2 IDE) raidz2 setup)

Geli is very fast, so any modern cpu shouldn't be a bottleneck. But if you're gonna use compression and need speed I highly recommend more cpu cores.
 
Whether the wireless works is always a 'wait and see'. I can imagine support under Solaris being much poorer compared to BSD; generally BSD has better hardware compatibility (still worse than Linux).
Just for the record, the Atheros is a Gigabit Ethernet, not a Wireless NIC.

But SB700 family is old; the new SB850 would be better since it has native 6Gbps SATA and uses newest chip.
There is just not much alternative on mini-ITX boards. Plus, the board is sold as industrial/durable and has pretty good sata-connections and plain ddr3 (vs. so-dimm).

Since the board only has one Atheros NIC (and only one expansion-slot which I wont use yet for just another NIC), I only have to saturate one gigabit-connection. I hope one of the cpus will achieve this with encryption enabled (still prefering ZFS-encryption under solaris due to its ease-of-use)

So I'm leaning to the x3 ... Thanks!
 
TDP is a really bad measure for efficiency. I tested several processors and the 65W 45nm Core2Duos came out way ahead of 45W Athlon X2 5050e if totally idle.
The most efficient system I have is an i3-530 on an MSI H55M-E33 which goes down to 20W while idling.
 
Last edited:
If you compare different sockets then you compare different motherboards as well. The efficiency of motherboards varies wildly, and the VRM plays a major part here. So comparing CPUs between AMD or Intel cannot be done without DC measuring of what the CPU *really* consumes; regardless of the (in)efficiency of the VRM and other motherboard junk.

For example, a Core i3 system that does 25W in default at the wall socket, can be brought down to under 10 watts without compromise to performance. It's not that the Core i3 uses that much power, it's just that motherboards waste alot of power and are extremely inefficient.

You can build very efficient AMD platforms using the same regime; i don't think either Intel or AMD really has an edge over idle power consumption. Generally for Intel the chipsets consume more power, while AMD has very low idle power chipsets. So comparing is hard. I do agree TDP is no measure of comparing power consumption. Besides you're wanting to compare idle power consumption, not stressed power consumption.
 
I agree here. Motherboard manufacturers include a lot of superfluous stuff these days even on budget boards. I can't even remember when I used one of these Firewire ports or additional SATA controllers.

Core i3 with AESNI also should deserve your attention.
There is no i3 with AES-NI support AFAIK. Intel is to blame here for arbitrarily disabling functions from their processors.
 
Last edited:
Google appears to agree with you. :)

The 32nm Core i5 and newer Sandy Bridge CPUs should support it, and upcoming Bulldozer, though perhaps FreeBSD needs to write a specific driver for AMD supported 'AESNI' so it may not be available straight away. Bulldozer would become an interesting product for home servers as well, and the first AMD product that is both 32nm and High-K Metal Gate, so idle power consumption should be very low.

At this very moment, Intel has an edge if you want to combine both performance with low idle power. AMD has low-power quadcores but they are not as powerful as the Intel ones. If you're on a budget then AMD still is a very interesting choice, especially if you want to gain cheap access to the ECC functionality that is part of Athlon / Phenom CPUs.
 
TDP is a really bad measure for efficiency. I tested several processors and the 65W 45nm Core2Duos came out way ahead of 45W Athlon X2 5050e if totally idle.
That is very interesting since I am using an C2D E6600 with 65W TDP but 65nm...

tough decisions to be made...
 
The 65nm Intels use alot more energy when idling in particular, i remember Q6600 values of 25W idle power for the CPU only; which is alot considering most modern CPUs never use more than 10W when idling. As always, it depends on your budget and priorities what is the best solution available to you. Write down your priorities and review each option based on those criteria.
 
The only thing you would need to look for is idle power consumption; your NAS won't be loaded 24/7. In fact, probably less than 0,05% on average. So the idle power is the dominant factor in NAS power consumption.

One word of caution about Toms Hardware: many reviews i've come to disagree with after careful study were written by Patrick Schmid and Achim Roos. I suggest caution at trusting these reviews, and instead look at a wide range of reviews from multiple sources to inform yourself. Relying on only toms hardware may yield conclusions that are unjustified and inappropriate.
 
Back
Top