A 8320e Build in 2016 or X5xxx Dual Xeon?

[jfp555]

Hi all,

I'm involved with entry level video production (vast majority of footage based on DSLRs and gopros), and although I prefer to prepare the rough cut on a laptop (lenovo w520 workstation with 2820qm, quadro 2000m, 24gb ram), I would like to have a desktop so I can use it for rendering and post production as well as revisions to the cut.

I already had the laptops fan replaced once due it being worn out due to the rendering. I love the portability and power that the laptop gives me and I like to keep it ready for new work and not bogged down with tons of cached data and raw footage.

We are looking at vimeo/youtube HD/4k level quality final renders. No fancy film camera footage yet.

I've been looking at the fire sale dual xeon builds as well but I just feel that it would be better to be covered by somewhat more recent platform with native USB3 and better SSD support than going with those ancient towers, but, the whole point of this thread is to build the best build with consensus form the forum.

The one thing I might have fun playing with on the dual xeon is hackintoshing it (if thats even possible).

I use Adobe products. (premiere pro mostly)

So, TL;dr: Is an AMD 8320e build (purchased from microcenter) with 32 to 64gb ram and overclocked to 4.5ghz (not all the time!) a great little render machine? x99 Intel just doesnt make sense to me at this time because I cannot justify the cost as I would like to wait for Zen.

 

[Burpo]

http://www.ebay.com/itm/Lenovo-Thin...674917?hash=item210f5d0425:g:vy4AAOSwcL5XNeVu

http://www.ebay.com/itm/Intel-XEON-...c5ac211&pid=100005&rk=1&rkt=6&sd=191868730909

 

[TeknoBug]

What??? $62 for an E5 2670? How much does the 2011 motherboards take you back though?

PS- you don't need dual Xeons for that use.

 

[Burpo]

Well, for less than $400 (add video & hard drive), he can have a pretty serious machine for rendering..

http://www.ebay.com/itm/Dell-Precis...7cb3618&pid=100005&rk=2&rkt=6&sd=371671888413

 

[MarkLuvsCS]

Just for reference. Here's a 9590 @ 4.7ghz with a cinebench around ~750 points. A stock dual 2670 system will score around ~2000 points. If rendering time is critical you can probably cut it in half with a dual 2670 system.

I'm using probably the best case for AMD and the base case for the 2670s. Also Zen is not going to be faster than the dual 2670s, so buying the used enterprise gear will give you better performance now with the only caveat of possibly needing to buy a sound and or USB3 pci-e card.

 

[Burpo]

The machines I linked have onboard sound, sata3 & usb3. One has video & HDD, one doesn't. There's plenty such PC's on Ebay, cheap..

 

[jhu]

Between the two, dual Xeon is the way to go, even if Adobe applications don't make goos use of dual processors. My dual Xeon E5 2670 is at least 2.5x faster than my FX8350.

It's likely to be faster than a single socket Zen. But if you really are hell-bent on waiting foe Zen, then just wait for Zen.

 

[aigomorla]

sigh 8C xeon chip for less then 70 dollars.

Burpo if i could ban you i would for tempting me with crack.... (sarcasm)
I mean it, that price point on that cpu is like crack... must resist...

 

[jhu]

sigh 8C xeon chip for less then 70 dollars.

Burpo if i could ban you i would for tempting me with crack.... (sarcasm)
I mean it, that price point on that cpu is like crack... must resist...

You pay for it with the motherboard.

 

[jfp555]

Thanks for the great replies so far. It seems that dual xeon or even single xeon as superior is the consensus here. A couple of caveats though:

1. I will have the parts brought over by family/friends from the US. So I unfortunately can't have towering server casings carried by them. I will have to buy parts separately though the casings are available in my country. For reference, the extortionists here are charging around $225 for just a single E52670 processor and $500 for the Dell T3600 with a single e5-1660.

2. If I do end up buy the CPU's and RAM and mobo, what kind of cost am I looking at? I do need to RAID atleast a couple of drives in the system, have 1 SSD for boot, and another drive for rendered output.

So 1 x SSD (boot)
2 x HDD (RAID0) or RAID1
1 x HDD for final renders destination
and 1 x SSD (if possible, for cache purposes if the need arises)

I would prefer atleast a couple of USB3 ports and the option to add a modern gpu (or 2) I could use hardware passthrough to have multiple OS's running (but would need to add a couple more drives as well.

Also, i believe I need 2670's v2 version as they allow vt-d passthrough. Any idea if the ones linked are capable?

From another post regarding dual xeon on this subforum:

http://www.natex.us/product-p/intels2600cp2j-custom.htm

I can configure this with dual 2670's but they only have V1.
And the motherboard is also quite crappy looking with minimalist features.

Could you guys help point me to some decent motherboards? I could even buy a brandname barebones tower locally (Dell, or HP z800 series seems available easily) and have the V2 2670's along with ECC RAM brought from the US?

I'm also open to buying a motherboard from the US as well provided it isnt brutally expensive and has the features I an interested in.

Thanks again for the excellent advice and patience to understand the situation I'm in.

 

[Burpo]

Motherboards are brutally expensive. Your best bet is to buy a tower locally. The Dell T3600 or HP800 is the way to go for your needs, and inexpensive. The E5 2670V2 is needed for vt-d, but not sure if you can upgrade to Ivybridge tho.

http://www.ebay.com/itm/Intel-Xeon-...451860?hash=item236bae7994:g:100AAOSwjXRXYWGM

IMO you don't need a dual cpu setup.

 

[MiddleOfTheRoad]

You guys are once again totally missing the point.

He's talking about Premiere Pro -- and that expensive Xeon setup (when factoring in the motherboard) is an overpriced and largely incremental improvement over an FX-8320e under the best of circumstances in Adobe Suite.

All his money should go into a dedicated Nvidia video card with as many CUDA cores as possible. GPU acceleration is way faster than the CPU rendering under Premiere Pro.

An FX-6300 with a Geforce 650 Ti (GPU acceleration cuts rendering to 84 seconds versus CPU rendering of 527 seconds). That FX-8320e with a GeForce will render around 4-5 times faster than the Xeon you are talking about (assuming the Xeon is equipped with a non-Cuda video card, but still.... BTW, OpenCL still doesn't work particularly well in Adobe).

Bottom line: The CPU is largely irrelevant to this discussion (as long as you have a quad core or better). For optimal Adobe Premiere rendering, it's all about having a strong Nvidia video card for GPU acceleration using the Mercury Playback Engine.

 

[Burpo]

Still rendering with X58 @ 4+Ghz w/GTX-570 here.. Fast enough..

 

[DrMrLordX]

Still rendering with X58 @ 4+Ghz w/GTX-570 here.. Fast enough..

Yeah, but he's got a good point. GPU rendering is going to be the best way to go, bang per buck. He doesn't have to get an AMD rig to do that, and he definitely should get an Nvidia card so long as the OpenCL support remains in a shoddy state.

 

[jhu]

Yeah, but he's got a good point. GPU rendering is going to be the best way to go, bang per buck. He doesn't have to get an AMD rig to do that, and he definitely should get an Nvidia card so long as the OpenCL support remains in a shoddy state.

There might be some "gotchas" with regard to that though. Coming from Blender, where CUDA rendering is well supported, the issue with GPU rendering is that the scene must fit within the GPU's memory. This works fine for some scenes, but highly complex scenes won't work with this because they're just too big. Personally, I can't rely on GPU rendering until 16GiB is common for GPUs.

Not sure Adobe Premier has similar issues.

 

[richaron]

There might be some "gotchas" with regard to that though. Coming from Blender, where CUDA rendering is well supported, the issue with GPU rendering is that the scene must fit within the GPU's memory. This works fine for some scenes, but highly complex scenes won't work with this because they're just too big. Personally, I can't rely on GPU rendering until 16GiB is common for GPUs.

Not sure Adobe Premier has similar issues.

I'm assuming this could be a niche case for an APU? I was thinking along the same lines for a potential science-y build.

Still faster than CPU based, but much slower than a proper GPU. I'm thinking there's a middle ground where a bunch of system RAM might open up options when the GPU can not access the normal virtual memory subsystem, and/or when the APU has a bandwidth/latency advantage vs GPU swapping in virtual mem over PCIe. Am I missing something?

 

[jfp555]

OK guys. This is turning into a very interesting conversation. I'd like to share a couple of things here. Since there are conflicting opinions here which are technically accurate to some extant but have other issues.

This info is divided into 2 sections. One is related to setting up hardware around workflow rather than universal scenario.

The second pertains to hardware alone.

1. WORKFLOW:

The way I use Adobe (and mine is a fairly universal method):

The GPU (both AMD and Nvidia, AMD FuryX is at this point even faster than nvidia due to HBM2 in the latest updates) are used for smooth real time playback and some basic effects DURING the edit.

Most people render out the final, edited file using solely CPU for most pro work. This has to do with GPU renders not being entirely accurate or the quality being lacking (due to being based on floating point i believe) and not universally supported for most formats. So while most gameplay style casual videos can be rendered out through GPU, but most pro softwares default to using CPU for rendering out.

The GPU is invaluable in ensuring that I can take just most footage, chuck it onto the timeline (WITHOUT the old approach of transcoding the entire footage into lower res for smooth cpu based footage handling in the timeline).

Were I working in a scenario with film quality footage (very high bitrate RAW, uncompressed footage obtained from cameras running in the tens of thousands of dollars) I would need 2p or even 4p xeon system with very fast raided storage to edit with my current workflow as there isn't much software development on that end and most consumer hardware not being fast/cheap enough.

Previously, transcoding everything was standard practice (hence its presence in benchmarks) but now in most usage scenarios, workload has shifted to GPU with only editing out final footage being CPU based.

So as @MiddleOfTheRoad mentioned, the GPU is crucial (though AMD is equally as good if not better in Adobe currently).


2: Hardware Utilization:

According to detailed testing done by a youtuber

https://www.youtube.com/watch?v=L2Kr_yjQiDU

They found out that Adobe (being prosumer rather than pro oriented) CANNOT utilize more than 8 cores.

That essentially means that the only place where the xeon will help me is either extensive transcoding or final rendering (which is not ultra time sensitive in my short term workflow needs).

Even right now, I use throttlestop to limit my laptops 2820QM by turning off turboboost to keep down temps and it works perfectly fine for rudimentary editing.

So, with all that out of the way, the current takeways are:

Tl:dr:

a) Abobe, for me is GPU dependent and brand agnostic. CPU is secondary (fast quad core or decent octa core being good enough)

b) A single or dual xeon is still amazing value for money if I were to be using multiple software simultaneously. And compared to the somewhat woeful IPC of AMD, very competitive even at relatively lower clocks (2.6ghz vs 4.5on AMD)

In the end it all boils down to platform. The reason for considering the AMD was that people were complaining about the limited connectivity/workarounds needed to get the xeon working smoothly. With a newer board with the AMD procs, I can get a dirt cheap proc with a good board that has all the connectivity options such as m.2, pcie ssd, pci3 gpu etc etc.

So, with all that out of the way, where do we stand now?

 

[superstition]

You guys are once again totally missing the point.

He's talking about Premiere Pro -- and that expensive Xeon setup (when factoring in the motherboard) is an overpriced and largely incremental improvement over an FX-8320e under the best of circumstances in Adobe Suite.

Isn't Cinebench basically an FPU test with a rendering skin? Given that the FX has just 4 floating point units (and 8 integer units) it's not going to do well in a heavily FP test.

 

[MiddleOfTheRoad]

Isn't Cinebench basically an FPU test with a rendering skin? Given that the FX has just 4 floating point units (and 8 integer units) it's not going to do well in a heavily FP test.

It is, exactly that. But Cinebench is generally an academic discussion -- because 95% of the video editors that I've worked with never touch Cinema 4D. It's largely a niche program for a small group of experts.

 

[MiddleOfTheRoad]

1. WORKFLOW:[/B]

The way I use Adobe (and mine is a fairly universal method):

The GPU (both AMD and Nvidia, AMD FuryX is at this point even faster than nvidia due to HBM2 in the latest updates) are used for smooth real time playback and some basic effects DURING the edit.

Most people render out the final, edited file using solely CPU for most pro work. This has to do with GPU renders not being entirely accurate or the quality being lacking (due to being based on floating point i believe) and not universally supported for most formats. So while most gameplay style casual videos can be rendered out through GPU, but most pro softwares default to using CPU for rendering out.

The GPU is invaluable in ensuring that I can take just most footage, chuck it onto the timeline (WITHOUT the old approach of transcoding the entire footage into lower res for smooth cpu based footage handling in the timeline).

Were I working in a scenario with film quality footage (very high bitrate RAW, uncompressed footage obtained from cameras running in the tens of thousands of dollars) I would need 2p or even 4p xeon system with very fast raided storage to edit with my current workflow as there isn't much software development on that end and most consumer hardware not being fast/cheap enough.

Previously, transcoding everything was standard practice (hence its presence in benchmarks) but now in most usage scenarios, workload has shifted to GPU with only editing out final footage being CPU based.

So as @MiddleOfTheRoad mentioned, the GPU is crucial (though AMD is equally as good if not better in Adobe currently).


2: Hardware Utilization:

They found out that Adobe (being prosumer rather than pro oriented) CANNOT utilize more than 8 cores.

a) Abobe, for me is GPU dependent and brand agnostic. CPU is secondary (fast quad core or decent octa core being good enough)

In the end it all boils down to platform. The reason for considering the AMD was that people were complaining about the limited connectivity/workarounds needed to get the xeon working smoothly. With a newer board with the AMD procs, I can get a dirt cheap proc with a good board that has all the connectivity options such as m.2, pcie ssd, pci3 gpu etc etc.

So, with all that out of the way, where do we stand now?

Very good points. I haven't tested Radeon cards under Adobe in about a year -- so it is good to hear that they have improved. I will say this, when I last tested the Mercury Playback Engine using Catalyst drivers -- OpenCL acceleration was no faster than CPU only rendering in Premiere CC. In some instances, OpenCL actually was even slower than CPU only rendering. I tested using both an R7 360 and an R9 380X. So it sounds like the Crimson drivers have improved things a great deal under Adobe Suite.

It sounds like your instincts are correct. In this situation, I'd probably get the modern motherboard (with the somewhat slower CPU). Pump as much money into the GPU and storage (SSD's) as your can.

 

[MiddleOfTheRoad]

There might be some "gotchas" with regard to that though. Coming from Blender, where CUDA rendering is well supported, the issue with GPU rendering is that the scene must fit within the GPU's memory. This works fine for some scenes, but highly complex scenes won't work with this because they're just too big. Personally, I can't rely on GPU rendering until 16GiB is common for GPUs.

Not sure Adobe Premier has similar issues.

Well, 16GB video cards are starting to show up a lot now in industry workstations. But FirePro's aren't particularly cheap. The video cards generally sell for around $3,000 by itself.

 

[DrMrLordX]

I'm assuming this could be a niche case for an APU? I was thinking along the same lines for a potential science-y build.

Downside: most UEFIs/BIOSes limit the amount of memory that can be assigned to framebuffer. FM2+ generally limits framebuffer allocation to 2Gb. So, no real advantage there unless you have some crazy HSA/OpenCL 2.0-style stuff going on . . .

Okay, so it's interesting to hear that you can't rely on GPUs for your pro-level render output. I'm a bit put-off that people are having accuracy/quality issues using GPUs for the final render! Is that with consumer-level cards? Maybe the pro cards (that cost $$$$$) and their drivers do a better job of addressing the problem?

In the event that you need CPU-level rendering and modern or semi-modern features, I can see why you might look at the FX since several board OEMs released refreshes of their AM3+ lineup with new features tacked onto the aging 970 and 990fx chipsets. I'm not really sure how FX stacks up in Premiere Pro in 2016 vs. a similarly-priced Intel CPU. Most FX chips are in the price range of a locked i5 Haswell or Skylake (think 4460 or 6400). Something like the 8320e can be had for less, though you will have to mess with clockspeeds (basically set it to max turbo, or close to it) to get the most out of that chip in any long-term render workload.

 

[superstition]

Something like the 8320e can be had for less, though you will have to mess with clockspeeds (basically set it to max turbo, or close to it) to get the most out of that chip in any long-term render workload.

If you have a Microcenter near, best bang-for-the-buck is to get a UD3P board with the bundle price and run the chip at 4.4 GHz. Disable APM (which turns off turbo) or you'll get throttling. Don't skimp on a cooler, though. I'd get a Noctua D15 or a 360mm AIO if not going for a custom loop. You'll also need to mount a fan to blow through the VRM sink.

 

[DrMrLordX]

4.4 GHz shouldn't require anything that extreme. The stock HSF probably won't do, but something like the Wraith would be fine (or an Evo 212+ or whatever). However, since he is doing some semi-professional or professional work, I would think he should be careful engaging in any actual overclocking - that is, pushing the CPU past its maximum turbo limits. 4.0 GHz should be much easier to attain, and he might even be able to sustain that clockspeed with the stock HSF.

 

[superstition]

4.4 GHz shouldn't require anything that extreme. The stock HSF probably won't do, but something like the Wraith would be fine (or an Evo 212+ or whatever).

A single fan 120mm single tower cooler will probably not be adequate unless you put a Delta fan on it.

I have built and tested 8320E and 8370E systems with air and water cooling.

The only way to get a reasonable level of quiet is to have a lot of airflow from a lot of big fans or a lot of radiator space.

People see the GHz number but underestimate how much heat is generated by all those 32nm cores at higher clocks (if you're not throttling) when running Prime.

I would never go with less than a 140mm dual tower dual fan cooler if overclocking an FX to 4.4. But, I also don't like loud systems and am conservative with temps.

Part of the issue could be heat from the VRMs traveling back to the CPU due to the lackluster heatsink on the UD3P. But, I have the strong impression that people typically overestimate the abilities of that EVO cooler.

Of course, one can always back off on the clockspeed and reduce the voltage to generate significantly less heat.

Here's Ian Cutress' 8320E overclocking chart that gives a rough idea of power consumption increases:

I don't know if that voltage he used for 4.5 would have been stable under Prime in a closed case. I have a feeling it wouldn't be unless you were on a custom loop. Even then I would expect that it wouldn't pass Prime large FFT 768–900K. As I recall, 1.385 was the bare minimum at that clock with both the 8320E and the 8370E I tested — with less droop (via Gigabyte's medium setting).

Going by that chart, though, one can get to roughly 4.1 GHz while being under 125W — so the EVO can certainly be a choice if you go for a lower clock.

I also suspect, based on how little voltage he needed at 4.5, that the reason his overclock failed at 4.9 is not due to the chip being unable to get there. Rather, I would suspect something like the VRMs of the board, the high level of droop the board appears to have, VRM cooling being inadequate, and the CPU cooler not being powerful enough (only a dual fan AIO I think). Also, I lapped both of my CPUs. The 8320E in particular was quite concave. The voltages are so high in that chart past a certain point that something seems amiss.

The best way to see how far an FX will go (for everyday use) is to use a Sabertooth or Crosshair board with a custom loop, of course.