Dual R9 390x 8GB or Dual RX 480 8GB for FCPX?

[AndreeOnline]

Could you both also share details about which drive you use, including its speed?

I've edited my post above. rhys216 is on a Hackintosh and can't be compared directly to actual Mac Pros. I mean, it's great that he's got good performance, but it doesn't translate to those of us modding actual Mac Pros.

BruceX generates a small 70MB or so file. A spinning platter, external 2.5" USB3 disk would write that file in 1s. The disk is not a factor.

But I'm using a fast PCIe SSD with around 1400-1500 MB/s.

 

[thephased]

thats the wrong adapter. The search term should be "dual 6 pin to 8 pin PCIe" here's what you'll need

http://a.co/iSmjSQY

View attachment 674581 View attachment 674582

the one you linked on your post is a single 6 pin to dual 8 pin. Thanks for the greetings, happy holidays and I hope this helps!

Thank you so much! I ordered it and it's on its way. I read earlier in the thread about the XFX cable. I even looked on the XFX site for it. I'm worried it may be a too short for the card, but you have the same card as I so it worked for you. The the PCI-E headers on the Mac Pro 4,1/5,1 look close to the slots unlike my Late 2005 G5 Quad Core.

On another note, I'd probably have to have an external PSU if I want to put my R9 Fury in there since they are 2x8 pin and a lot more power hungry. I wasn't really thinking I'd need to worry about faster cards when I bought the 2012 Mac Pro (Server) 6 Core 3.3ghz system. I don't want to burn out the motherboard although I've read no reports of them failing.

And happy holidays to you as well and yes it definitely did help

All your testing is really helpful for comparison and I'll run those tests (I'm curious to see how luxmark performs relative to my Late 2014 5k Retina) and post results here after everything is up and running the next week.

 

[chriz_r]

So I just finished testing out 3 different cards at 6 different configurations. I have a pair of RX 480 Reference card 6pin, RX 480 Red Devil 8pin, and R9 390X OC 6+8pin. They were all powered by the internal PSU, using AndreeOnline's pixlas style mod but with only 4 wires. (Thanks Andree!)

​

I am using 12AWG solid copper wire for the black and the white wires, black being the ground and white carrying the +12V power. One white and one black wire provides the power and the ground to one 8pin.

​

All connections other than the vampire clips were soldered. I checked the temperature of the vampire connections on each wire during full load at 980 watts using a laser temp gun and the wires were always a few degrees below ambient temp so the connection is rock solid.

​

Each card configuration was tested 3 times, rebooting each time. FCPX was as also loaded with a new BruceX 5K xml each time into the library to avoid letting the computer use any cached renders or anything that's in system memory. Background rendering is turned off. Each cards were also stress tested and benchmarked with Luxmark 3.1 using luxball. The tower's power was plugged in to a Kill-a-Watt power meter to record the average power draw during heavy loads and rendering. All the cards were also plugged in to slots 1 and 2 to avoid using the 4x slot, with the backplates and shrouds removed to fit dual configuration since the reference cards with no backplate fits just fine into those slots.

I'm curious about some BruceX times I see. You guys are posting times around 20s for single card setups right?

My 280X time was 40s and 480X around 33-35s. My 5.1 is pretty maxed out and performs normally in other benchmarks. It's a bit of a head scratcher.

To be honest, I actually was able to get low teens before, but thats after running the brucex test multiple times and my guess is that the machine was keeping a cache in memory or the disk and synthetically speeding up rendering, so I switched it up.

The results are pretty weird to me, with the reference cards actually beating all the other aftermarket cards. Even after switching configurations multiple times randomly and resetting PRAM and SMC each time to clear internal memory, the results were still the same.

​

Higher is better in this entire graph, and although dual cards are faster in rendering, the performance ratio is what I tried to determine. All bruce-x times are on the other chart, so if you need to just have raw power, just look at that.

From what i've seen, one reference card was faster than any single aftermarket cards in the list. The 390X price/performance ratio was comparable to the reference card, but as you can see on the red bar, it has the worst power usage with one card practically consuming the same amount of power as two reference 480 cards under fcpx.

TL;DR : reference 480 is better than 390x, consuming less power for relatively similar price.

Now I need those dual 280X for comparison.

 

[AndreeOnline]

That's for the write-up! It takes work, so much appreciated!

 

[rhys216]

From what i've seen, one reference card was faster than any single aftermarket cards in the list. The 390X price/performance ratio was comparable to the reference card, but as you can see on the red bar, it has the worst power usage with one card practically consuming the same amount of power as two reference 480 cards under fcpx.

It could be that the reference 480 has a really good asic quality so maybe doesn't go over tdp/temp limits where it would otherwise downclock?

People have also been getting better performance from the 480 by manually undervolting it via vbios. This enables the boost clocks to be much more stable and the card uses less power.

I downloaded Powercolors unlocked bios, which lifts the TDP cap, so the card doesn't downclock due to power consumption.
I recently edited ithe bios further by undervolting it and raising the boost clocks to 1370mhz.
Also changed the fan profile so the fans don't spin up until the card hits 70c.
It runs silent most of the time, and still pretty quiet under load.

 

[chriz_r]

yeah i was thinking that asic had something to do with it, but thats only on the extremes if im doing full on benchmarking and stress testing. If you look at my table with the luxmark and geekbench test results, you could probably tell just by the score which one was the reference, reddevil, and 390x. But when it came to FCPX realworld/bruce-x use, the results manifest quite the opposite way. I wasnt hitting temp limits during fcpx usage so I kinda ruled out asic, but started thinking that maybe it does have something to do with software/firmware interactions that somehow the more "standard" card just works and the aftermarket ones have to hesitate. Like high horsepower cars just spinning the tires off the line while the mediocre civic is already moving ahead of the drag race.

I also tried to keep the test as stock as possible on the cpu side and closer to what most people might have access to (2.4GHz quad) which could be the bottleneck, but i'm confident that anyone who has higher spec than my tests should be able to get better results even if its just marginal.

I didn't want to do over/under clocking procedures because not everyone is comfortable with doing it, so it didn't make sense for me to introduce that variation. To many people, the power supply mod is already daunting and 'risky' so most look for and opt for the "pop it and leave it" approach.

 

[theitsage]

I agree with @rhys216. Same model GPUs are not always built of the same quality. I have a triple RX 480 Crossfire setup (all reference) in my Alienware Area-51 R2. The Alienware BIOS prefers the bottom PCIe slot as the main one to output video signal at boot. I would get random lockups until I moved an RX 480 into the bottom slot.

 

[chriz_r]

Yup! the dreaded silicon lottery. Same goes for my PC cpu's. some would let me go 4.6Ghz on one chip but when I pop in another identical one on the same system, can't even hit 4.3Ghz.

dang thats a pretty alienware!

yeah i'm wondering how dual 470's would work for my use...

 

[theitsage]

Yup! the dreaded silicon lottery. Same goes for my PC cpu's. some would let me go 4.6Ghz on one chip but when I pop in another identical one on the same system, can't even hit 4.3Ghz.

dang thats a pretty alienware!

yeah i'm wondering how dual 470's would work for my use...

You can run 3 RX 470s with the internal PSU. I tried in my 5,1 and it ran fine in both macOS and Windows 10. MicroCenter typically has open box deals so you can get the RX 470 closer to $150 a pop.

 

[chriz_r]

yeah i got a lot of my cards from microcenter, though even their new cards sometimes are priced lower than amazon or other retailers so i always keep an eye on that. I'll probably get 3 470's to test out for finalcut and see if its better than dual 480 performance

 

[mavericks7913]

RX 470 and 480 are not the good choice for now because Mac OS does not support these GPU for 100%. Right now, I feel that it just perform like HD5870 so dont buy it.

 

[chriz_r]

its not completely supported but the 8GB vram enabled me to get a bit more done now for my usage, thats why i went with them. Sure, older cards are able to hit good target scores, but with power efficiency/usage, stuttering on multiple 4K streams for fcpx, and noise, the RX series held up just a bit better (for my case at least) thats why I chose to utilize them.

 

[mavericks7913]

But still, it has potential performance but Mac OS does not support it at all. I talked to Apple and they said until they announce a new iMac or other computers with RX 470 or 480, they WILL NOT going to support ANY GPUs AT ALL. They might skip Polaris 10 series I think unless they announce new iMac with RX 470 and 480 since Vega series are high end GPUs. I guess we should wait for new iMac because RX 470 and RX 480 are the only choice .

 

[chriz_r]

Yep its not 'supported' at all but many thanks to the modding community, it kinda 'works' and luckily for my use. There's just so much speculations and ambiguities when it comes to apple products and it's always been that way, sadly I'm in no position to wait and I have to use and maximize what I can for the budget that I have. That, like you said, entails resorting to rx480/480 as those are on the list of choices.

 

[rhys216]

But still, it has potential performance but Mac OS does not support it at all. I talked to Apple and they said until they announce a new iMac or other computers with RX 470 or 480, they WILL NOT going to support ANY GPUs AT ALL. They might skip Polaris 10 series I think unless they announce new iMac with RX 470 and 480 since Vega series are high end GPUs. I guess we should wait for new iMac because RX 470 and RX 480 are the only choice .

After you update the 4100.kext What's not actually supported with a 480?
So far for me, everything works nicely and runs fast.

 

[chriz_r]

youre running a hackintosh on a newer cpu, on a totally different platform with newer PCIe busses and have overall more "upgradeability" so im not sure if we can really compare apples to oranges here with my 2010 cMP. I have two hackintosh machines but they're just not reliable for me when it comes to work use. I do keep them because I love tinkering and figuring how to make them work if it breaks.

Btw, whats your unigine heaven score on the same settings you did with valley?

 

[rhys216]

It does take some tinkering and I use native apple parts where possible (wifi/bluetooth etc), but I also have a 5k imac, and it runs everything just as stable, so I use my hack as my main work computer (photography/video) as it's faster/quieter.

 

[chriz_r]

...It does take some tinkering...

Yep you're absolutely right. Unfortunately I cant tinker with the system or even have downtime when projects are due with multiple deliverables required.

Looks like you're getting at least 500-700 points more than what i normally get on my mac pro with rx480. But I wonder if the iGPU is actually contributing to the score? Because if it does then we're on the same ballpark with just 480 on the equation.

 

[rhys216]

Yep you're absolutely right. Unfortunately I cant tinker with the system or even have downtime when projects are due with multiple deliverables required.

Looks like you're getting at least 500-700 points more than what i normally get on my mac pro with rx480. But I wonder if the iGPU is actually contributing to the score? Because if it does then we're on the same ballpark with just 480 on the equation.

Yeh I no what you mean. I used my imac, and tinkered in my spare time until I was confident everything worked as it should. I think it's a shame apple moved to the trash can shape for mac pro's, because you can't upgrade it easily and because I loved the look of the apple tower case. Trying to find anything that looks as remotely nice in the PC world is difficult. But below is what I settled on.

I don't think the igpu is contributing. If I run the display off the igpu rather than the 480, I get about 5fps so the 480 doesn't get used. Luxmark etc. can use the igpu as well as the 480 but not Heaven. I don't think fcpx uses igpu either, as I think fcpx only uses multiple GPU's if they are the same chip.

Edit:
500-700 points seems quite a lot. Perhaps real Mac pro's are not utilising the 480 like hacks do?
I set my hack up to think it's a 5k imac, so maybe that has something to do with it also, because I believe hack have differing performance depending on which system definition you use.
It would also explain the slower fcpx times, although I also think slower CPU clocks are a factor as well.

 

[TitouDoc]

What tower is this ? Very nice...

 

[rhys216]

It's a umx4 without a window.
Btw it has a corsair dual water cooler at the top. You need to use 15mm fans (that come with the case) as there isn't much room.

 

[thephased]

Okay, so I did not read the seller's explicit description that his Mac Pro is actually a 4,1 flashed to 5,1 listed as a 2012 Mac Pro Server 5,1 6-Core Xenon Westmere. It is still a good deal considering the CPU in the system is actually a single X5680 instead of the Early 2009 6-Core W3580 or even the 2010/2012 6-Core W3680. They didn't even know it had an X series chip in it. They were a personal seller and it came with 32gb ram, new 3TB HD, and a AMD HD 7950 which works well and shows the EFI boot screen with DVI at least. No problems with MDP (did not test HDMI). CPU temps and overall temps look really good so whoever swapped the processor looks like they used good techniques. Everything is quiet, stable, and works well. I just hope these cMP flashed 4,1 systems work on future mac OS.

Sorry for the long tangent. This post is about testing and comparison. Hopefully it will help.

System: Mac Pro 4,1 flashed to 5,1
OS: Sierra 10.12.1
CPU: 1x Xeon 6 Core X5680 3.33GHz
Memory: 4x8gb DDR3 1333 (running at 1333)
Graphics:
MSI Radeon HD 7950 Twin Frozr III - slot 1 / PowerColor RED DEVIL Radeon RX 480 - slot 1 (2x6 to 8pin)
Storage: 2x 2.5" SATA Samsung 850 EVO 1TB
PCIe devices:
Apricorn Velocity X2 Duo w/ SSDs in RAID 0 using Disk Utility (boot drive) - slot 3
CalDigit FASTA-6GU3 Plus - slot 4

Benchmarks:
Geekbench 4: CPU 64bit and GPU Compute
Luxmark 3.1 (RX480 results only)
Valley benchmark
Xbench 1.3
QuickBench 4.0
BruceX

Results:

Geekbench 4: CPU 64bit and GPU Compute

(I have since put the 8gb stick back into slot 4 for 32gb ram, but I'm not sure if it would affect the score much. I wanted to make sure triple channel was being used and I'm not sure if using all four slots will break it)

RX480

HD 7950

Luxmark 3.1 (RX480 results only)

Valley Benchmark

RX480

HD 7950

Xbench 1.3

QuickBench 4.0

BruceX - ~24 second average (3 tests)

_____________

I'm seeing noticeable improvements in more demanding Steam games with the RX480 even though it's only using 16CU so real-world performance looks to be quite an upgrade still while we wait for full Polaris support. I would gather Apple will use a full CU mobile Polaris GPU in their next 5k Retina iMac much like they've used the full Tonga shader cores/CUs in the M295X and M395X with lower clock speeds. Looks like they keep sticking with AMD. And I don't want to take the plunge yet into an external PSU setup. I want to see how stable bootcamp and mac OS are with heavy GPU loads and keep watching the wattage. I don't plan on heavy gaming. I have a PC for that.

I wish the Velocity X2 Duo in RAID 0 was faster with these drives, but I've seen other reports on here with similar speeds on the 850 EVO in particular to what I'm getting and it's a PCIe 2x card not 4x. I may consider an M.2 drive, or do a multiple SSDs RAID 0 setup through the internal SATA II ports. I can always boot off of the eSATA ports as well. Or just leave as is. I'm just spoiled with an 950 Pro NVMe in a PC and can definitely notice the difference from AHCI interface SATA SSD drives in certain situations.

I have not tried bootcamp yet.

Sorry for the really long, detailed post with large images, but I wanted to provide people a reference for comparison.

Hope this helps and feel free to ask any questions.

 

[h9826790]

I have since put the 8gb stick back into slot 4 for 32gb ram, but I'm not sure if it would affect the score much. I wanted to make sure triple channel was being used and I'm not sure if using all four slots will break it.

That's easy to know the answer, pull out one of the RAM stick and run the test again. Anyway, the overall score difference may be around 500. There is almost zero difference in real world (you should not able to feel it), and running out of RAM always have more penalty then not optimise the triple channel configuration. Even though you didn't run out of RAM, more spare RAM still helps to let the OS use them as cache to speed up the whole system.

 

[thephased]

That's easy to know the answer, pull out one of the RAM stick and run the test again. Anyway, the overall score difference may be around 500. There is almost zero difference in real world (you should not able to feel it), and running out of RAM always have more penalty then not optimise the triple channel configuration. Even though you didn't run out of RAM, more spare RAM still helps to let the OS use them as cache to speed up the whole system.

I see, thanks for the knowledge! I'll keep it at 32gb.

 

[knowhate]

.