Topic Title: AMD cpu choices.
Topic Summary: Want as much cpu as possible, dual socket maybe?
Created On: 05/07/2013 10:03 PM
Status: Post and Reply
Linear : Threading : Single : Branch
Search Topic Search Topic
Topic Tools Topic Tools
View similar topics View similar topics
View topic in raw text format. Print this topic.
 05/07/2013 10:03 PM
User is offline View Users Profile Print this message

Author Icon
lezzmeister
Peon

Posts: 5
Joined: 04/08/2013

So here I am, trying to do some work in Blender and Maya and trying to tune a system and (attempt to) write some code(FreeBSD mostly). And the rendering and compiling takes ages. I need more cpu power.

I get issues with running a system more than a few days. People on other forums(tweakers.net & hardware.info) told me to use Opteron or Xeon as this will solve some of the issues I am having, and I know these come in dual socket. Not needed I guess, but if I could, I'd be really happy...

Now the problem is this is not making me any money for now since I suck too much at this to give up my dayjob, so I'd like to not spend a boatload of cash. So I browsed a bit and came to the conclusion Intel is costly, and then I saw this: http://tweakers.net/pricewatch/295821/amd-opteron-4226-boxed.html 12 cores in a single system, now that is looking quite tempting.

Then I turned my attention to the motherboards. G34 is through the damn roof, that is a no-go. So C32 it is. This would be an excellent platform for the more enthusiast homeuser, but it has a rather outdated chipset. As in its 2008/2009 parts outdated, and this is not too tempting on a desktop. Which is too bad or it would be on order already. The alternative would be socket 2011's(yes the competition) Intel Xeon E5-1650. But that one cpu is more than 2x Opteron 4226 and a C32 board combined, at least where I live. The AMD SR5670 still is the latest chipset for G34 and C32 right? Correct me if I'm wrong.

Now I know most news reported Steamroller is still on for 2013, so I thought what if I wait it out a bit and see.  Depending upon who you believe, there could be a new socket in it. And a new socket may mean a new modern chipset.

But is Steamroller in 2013 only coming for the Opteron as some reports said, or is it only coming for APU's as other sites reported? Or both? And is there really a new socket in it for Steamroller? Or is that only for Excavator? Is there a new Opteronsocket or chipset on its way? Will all Opterons switch to a single socket if that is the case and ditch C32 and G34? I know that a single socket for APU and FX is the goal this next generation or the one after that, at least for the desktop.

Is there ever anyone from AMD on this forum that could at least drop a hint on when the new socket/chipset is here for Opteron?

If anyone here has any suggestions on how to cram more horsepower into a case, I am all ears.

Props to AMD on keeping the sockets so you do not need to change entire systems, but it seems there can be downsides afterall.

 05/11/2013 07:30 AM
User is offline View Users Profile Print this message

Author Icon
zipsi
Elite

Posts: 1050
Joined: 11/22/2010

You should look into OpenCL, gpu accelerated rendering beats any cpu out there.



-------------------------
Intel I7 3770K @ 4.5ghz, Cooler Master Hyper 212 Evo, Gigabyte Z77X-UD3H, Gigabyte HD7970 ghz edition, 4x4gb Kingston HyperX Beast 2133mhz, Seasonic Platinum 860, Samsung 840 Pro 256gb, OCZ Vertex 4 128gb, WD Black 1TB, WD Green 3TB+1.5TB, ASUS Xonar Essence ST, ASUS VE278Q, Windows 3.11
 05/12/2013 10:46 AM
User is offline View Users Profile Print this message

Author Icon
Mime
Forum Moderator

Posts: 578
Joined: 08/28/2012

If its a data-parallel application... yeah, GPU computing can help a lot with performance.  Whether or not that's feasible and will make the extra complexity which goes with it worthwhile depends on the application.

Opterons and Xenons are basically the same chips as we see on the desktop, so most of what makes workstations and servers different is found in the motherboards and elsewhere in the system.  In terms of stability that extra stuff can make a real difference.... in terms of performance... it depends.  If performance was the only concern, I'd be tempted to stick with desktop hardware.  If you've already got a chip capable of running 6 or 8 threads, then you've got to be doing pretty gnarly stuff to make more crunch power necessary.

Another approach is to use something like distcc to distribute compilation to multiple systems.  That way you could use any old boat anchor as part of the compile farm, and maybe not spend a bundle on hardware.



-------------------------

Do not meddle in the affairs of archers, for they are subtle and quick to anger.
Post Count: +8510
Troll Hunter

The opinions expressed above do not represent those of Advanced Micro Devices or any of their affiliates.

 05/12/2013 11:08 AM
User is offline View Users Profile Print this message

Author Icon
lezzmeister
Peon

Posts: 5
Joined: 04/08/2013

Originally posted by: zipsi You should look into OpenCL, gpu accelerated rendering beats any cpu out there.

I know of OpenCL, and that will work with about 25% of the work, but for the rest I will need raw cpu horsepower.

Originally posted by: Mime If its a data-parallel application... yeah, GPU computing can help a lot with performance.  Whether or not that's feasible and will make the extra complexity which goes with it worthwhile depends on the application.

 

Opterons and Xenons are basically the same chips as we see on the desktop, so most of what makes workstations and servers different is found in the motherboards and elsewhere in the system.  In terms of stability that extra stuff can make a real difference.... in terms of performance... it depends.  If performance was the only concern, I'd be tempted to stick with desktop hardware.  If you've already got a chip capable of running 6 or 8 threads, then you've got to be doing pretty gnarly stuff to make more crunch power necessary.

 

Another approach is to use something like distcc to distribute compilation to multiple systems.  That way you could use any old boat anchor as part of the compile farm, and maybe not spend a bundle on hardware.

 

Xeons have the ECC that the lowerend Pentiums have, but that has been ripped out of the mid-end cpu's, go figure. Tight market segregration.

Stability is the primary concern here, since desktop tend to lockup or reboot if they have to push 100% for 3 days(i7-2600 on Intel mobo) or just fail and turn themselves off(A10-5800K, 4,2Ghz)like a power outage. I figured if I could get dual socket that would be a nice bonus, but the platform itself is pretty outdated when it comes to Opterons. I could buy Xeon but those require me to sell a kidney.

Distcc might be an option, but then I may be better off buying just 2 of the fastest cpu's I can find, put them in eenie meenie mini-itx housings and connect them to the rig I do the compiling on. If I could get my hands on one more infiniband card they would be connected with lightspeed as well. The reason I have 2 infiniband cards and the extra A10-5800 is I tried this already. I figured the gpu part could serve as a gpu backend/renderfarm for OpenL as well. But it is not all that easy to do as it would seem.

The lowerend Opterons(C32 socket) could be the way for someone that is getting into 3D and development more seriously, but apart from some new cpu's it looks like the forgotten baby a bit. As I am typing this, it dawns on me this may seem a little harshas you just read it, but no insult intended.

 05/12/2013 10:49 PM
User is offline View Users Profile Print this message

Author Icon
Mime
Forum Moderator

Posts: 578
Joined: 08/28/2012

For AMDs stuff, it's usually in the motherboard where the support for ECC gets cut.  The memory controller in the CPU its self still supports it.  For Xeon's I'm not sure, but I wouldn't be surprised if the story was the same.

I believe most AM3 boards from Asus still carry ECC support, so that might be a thing to look at if you want to try sticking with cheaper desktop hardware.  Server boards are often more stable in general since their PCB is often built using more layers, among other things... but they're also more expensive.

And yeah... I can imagine it being a mess trying to setup distcc for GPU computing.  If you're going to go the distcc route I'd leave the GPU out of it.



-------------------------

Do not meddle in the affairs of archers, for they are subtle and quick to anger.
Post Count: +8510
Troll Hunter

The opinions expressed above do not represent those of Advanced Micro Devices or any of their affiliates.

Statistics
85357 users are registered to the AMD Support and Game forum.
There are currently 4 users logged in.

FuseTalk Hosting Executive Plan v3.2 - © 1999-2014 FuseTalk Inc. All rights reserved.