TR Forums

Edit 1: Gaming performance results in post #15.

Edit 2: Frame latency, Overclocking with AIO Coolers, and a quick look at Bottlenecking in post #54.

Edit 3: Hardware failure in post #61.

Edit 4: Hardware change to Triple R9 290X's and cooling problems in post #104.

For those that saw and contributed to my thread pre-build over the past 2 days, here's the finished article. Below are various pics during assembly, and some notes and whatnot. The full spec is in the thread above, and if you want some high res pics of the smaller samples below, I have zipped them all here (approx. 50MB).

It was a PITA to assemble, especially the fiddly cooler fitment I couldn't get a screwdriver easily to. The graphics cards also had a huge metal back plate attached to nothing on the card itself, but mean's you couldn't fit them next to each other. I had to remove the plate off cards 2 and 3, and the 7 screws for them were only accessible by removing the entire heat sink and fan off the card first. Sigh.

So, yeah - Here's some build pics. The empty case and all panels removed:


Removable HDD/SSD caddy:


The 3 HDD's, the SSD and the cooling fan attached to the caddy:


HDD caddy secured into the case:


Here comes the Motherboard:


Seated the PSU, Power, SATA and all header cables for USB, Sound and FP stuff:


Closer up - I gave up keeping the cables neat in the end:


Added the RAM's:


Seated a trio of Radeons:


The CPU cooler before install:

Dropping in the Piledriver 8350:


The CPU cooler installed in the system:


All finished (Interior view):


Eternal view (Front)


External View (Back)

Win 98 SE DOS BOX??

Great pics, I love this stuff.

I don't have money for that kind of build, i'm drooling over the video cards.

Looks good! You mentioned all that work done to cards 2 and 3. Any closeup on that for others as a reference?

I love the loyalty to AMD. I truly hope your nice pile driver build is not too much CPU limited running Tri Fire. That is the one thing I am worried about for you.

Besides that you have a nice build...I wish you the best of luck with it.
Note: If your GPU temps are high " not much room between them" and your side panel fan is set to intake after testing, retest with the side panel fan set to exhaust. I had to do that with my classified cards in a HAF 922 case that has 2 120mm side panel fans and it dropped my top card temps by 15c and I have a extra slot between my 2 cards. Just food for thought since those 3 cards are going to throw out a ton of heat.

Also use GPUz to monitor your GPU usage on all 3 of your cards if your 3rd card is not doing much of anything you just might get better performance and temps if you remove it and run crossfire and separate your 2 cards so they are not mating with each other...on the other hand maybe they will have a kid or 2. Maybe more if you have 1 male and 2 female cards to propagate the species. Put your male card in the middle for best results.

But can it run word 2013???

And very nice pics.

vargis14 wrote:

I love the loyalty to AMD. I truly hope your nice pile driver build is not too much CPU limited running Tri Fire. That is the one thing I am worried about for you.

Besides that you have a nice build...I wish you the best of luck with it.

Ty. It was built to a tight budget with clear focus on GPU performance above all else. Maybe Intel next time, but I held onto my last PC for 7 years (It has a C2Q in it). I'm going to benchmark the CPU/GPU fully loaded and work on overclocking that FX chip someway. I've seen guides that generally say it needs to be at 4.5Ghz+ to keep within distance of the i7's on threaded workloads. That's what I'm aiming for. Then I also need to consider the CPU's Hyper-Transport link to the PCI-E lanes, as they may be starved of bandwidth ad it's default 5.2GB/Sec speed. Tweaking that is a little trickier though, but at least on this board that chip has a giant heat sink on it...

Flying Fox wrote:

Looks good! You mentioned all that work done to cards 2 and 3. Any closeup on that for others as a reference?

Sure, I'll take some . In essence it's easy enough with small tools, look at the 9th pic in my OP - You can see a black metal plate down the entire backside of the 1st Radeon. The other cards had this too, and it was in the way (it adds an additional 10-12mm off the board) with Crossfire.

Spin the card over onto the topside and you'll see 4 large screws flanking where the GPU socket is thru the 2 fans (yeah, I know - who designs these things?). Undo those, flip the board over. Unscrew the back 6 on that black plate that says R9 280X on it. Flip it over again, and gently wiggle - the HSF comes off, revealing one last screw near the video RAM's that you remove. The black plate then falls off.

Carefully align the 4 larger screw holes above with the board holes and re-seat the HSF. Job done.

vargis14 wrote:

Note: If your GPU temps are high " not much room between them" and your side panel fan is set to intake after testing, retest with the side panel fan set to exhaust.

Might try that. Under load, the 1st card hit 88C, middle one 87C, bottom 92C. I knew they would run, um, warm. None throttled clocks back though, so how high did AMD set the dam thing before it kicks in???

vargis14 wrote:

On the other hand maybe they will have a kid or 2. Maybe more if you have 1 male and 2 female cards to propagate the species. Put your male card in the middle for best results.

Excellent!

Cut a panel out the bottom of the case if you have a dremmel to assist cooling the bottom card - no need for a fan really it'll be drawn up anyway. Make sure you put a filter on

Excellent pics, i was just awestruck by the pic of the graphic cards.

Congratulations on your new build. Do post some numbers from popular games, I'm curious to see what performance you're getting.

Speaking from my own experience with multiple AMD GPUs, I would suggest trying to undervolt them, if they have multiple BIOSes (most of them do). Another, simpler, solution would be to use MSI Afterburner. Undervolting can often make a very big difference in heat and temperature. My cards would run stable at 1.137 then 1.087 and finally 1.081 V. Many people have had similar results.

A good reminder to clean out my PC.

3 280x in a 2H 2014 build.

Interesting choice....

Prestige Worldwide wrote:

3 280x in a 2H 2014 build.

Interesting choice....

Means I don't need heating in my room over winter.

But thanks for the comments all - Not had much time to test gaming performance as much as I would like, but anyone want to hazard a guess at what's going on with these?

To set a few things first - No components are overclocked for these. The AMD drivers are the 14.4 release. Operating system is Windows 8.1 with all applicable updates. Other than system services, the only programs running were Steam and Origin. AMD Raptr was used to track FPS. Afterburner was tracking temperatures, but I felt including them was pointless. All cards under load hovered between 87-92C solidly and didn't budge. That's hot folks, but no throttling in sight.

3DMark 2013 - Firestrike Test 1 - Extreme Preset 8xAA/16x Filtering - 1920 x 1080
2 x R9 280X - FPS Min (17) FPS Max (83) FPS Average (77)
3 x R9 280X - FPS Min (6) FPS Max (115) FPS Average (98)

3DMark 2013 - Firestrike Test 2 - Extreme Preset 8xAA/16x Filtering - 1920 x 1080
2 x R9 280X - FPS Min (3) FPS Max (28) FPS Average (22)
3 x R9 280X - FPS Min (3) FPS Max (17) FPS Average (15)

3DMark 2013 - Firestrike Test 1 - Extreme Preset 8xAA/16x Filtering - 5760 x 1080
2 x R9 280X - FPS Min (4) FPS Max (52) FPS Average (77)
3 x R9 280X - FPS Min (13) FPS Max (66) FPS Average (57)

3DMark 2013 - Firestrike Test 2 - Extreme Preset 8xAA/16x Filtering - 5760 x 1080
2 x R9 280X - FPS Min (1) FPS Max (7) FPS Average (6)
3 x R9 280X - FPS Min (1) FPS Max (6) FPS Average (6)

Valley Benchmark - Ultra Preset - Max Tessalation 8xAA/16x Filtering - 1920 x 1080
2 x R9 280X - FPS Min (17) FPS Max (112) FPS Average (101)
3 x R9 280X - FPS Min (18) FPS Max (137) FPS Average (122)

Valley Benchmark - Ultra Preset - Max Tessalation 8xAA/16x Filtering - 5760 x 1080
2 x R9 280X - FPS Min (12) FPS Max (78) FPS Average (67)
3 x R9 280X - FPS Min (14) FPS Max (83) FPS Average (69)

GRID Autosport - Ultra Preset (Benchmark Mode) - Max Tessalation 8xAA/16x Filtering - 1920 x 1080
2 x R9 280X - FPS Min (77) FPS Max (122) FPS Average (104)
3 x R9 280X - FPS Min (94) FPS Max (153) FPS Average (138)

GRID Autosport - Ultra Preset (Benchmark Mode) - Max Tessalation 8xAA/16x Filtering - 5760 x 1080
2 x R9 280X - FPS Min (51) FPS Max (71) FPS Average (64)
3 x R9 280X - FPS Min (68) FPS Max (95) FPS Average (88)

In a real world game, it's plenty fast with 2 cards. The Firestrike 2 test seems troublesome for whatever reason across all resolutions, and it runs even slower using 3 cars! I'll get some more tests done on BF4 when I get a moment. This lot took me almost 2 hours!

Easier to read graphs of the above data table.

Edit: Links removed - I lost the files after moving the server, sorry!

Where's your frame time plots? It'd be nice to see if 3 cards can keep consistent per-frame times.

I think your last benchmark answers my question. But what display setup are you using? 3 x 1920x1080?

Captain Ned wrote:

Where's your frame time plots? It'd be nice to see if 3 cards can keep consistent per-frame times.

Would love to include those, but I don't know how to do the frame latency stuff TR include in the articles. Anyone able to help?

From the motion on screen as I perceive it, 2 cards seemed less "stuttery" than 3, but only in the Synthetics. I'm going to look further tomorrow, but suspect CPU overhead, or a bottleneck on something this has called a HT Link may be the culprit. As I understand it, the HT Link is some bandwidth interconnect between the CPU and PCI-E Bus. I assume the faster that goes the better? GRID is a dream to play with enough performance overhead though to use AMD's fancy 16x AA engine. It looks awesome.

liquid_mage wrote:

I think your last benchmark answers my question. But what display setup are you using? 3 x 1920x1080?

What was your question (you want these?). I have 3 Asus monitors in a 3 x 1 config, but GRID and BF4 run so smooth with everything cranked, I may have overshot my GPU requirements somewhat. I'll call it future proofing.

geekl33tgamer wrote:

Prestige Worldwide wrote:

3 280x in a 2H 2014 build.

Interesting choice....

Means I don't need heating in my room over winter.

But thanks for the comments all - Not had much time to test gaming performance as much as I would like, but anyone want to hazard a guess at what's going on with these?

To set a few things first - No components are overclocked for these. The AMD drivers are the 14.4 release. Operating system is Windows 8.1 with all applicable updates. Other than system services, the only programs running were Steam and Origin. AMD Raptr was used to track FPS. Afterburner was tracking temperatures, but I felt including them was pointless. All cards under load hovered between 87-92C solidly and didn't budge. That's hot folks, but no throttling in sight.

*benches*

In a real world game, it's plenty fast with 2 cards. The Firestrike 2 test seems troublesome for whatever reason across all resolutions, and it runs even slower using 3 cars! I'll get some more tests done on BF4 when I get a moment. This lot took me almost 2 hours!

I just think it's a weird choice to run three 2.67 year old GPUs (rebranded / tweaked 7970s) which I imagine would have cost you $600-900, when you could have gotten 2 290 or 290x in the same ballpark and likely have better performance and less power consumption.

Not to say it's not a good setup, but triple GPU is generally less reliable than dual, and it doesn't seem like an efficient way to go about your build. I would also consider Tahiti GPUs to be too old to be considered "future proof" at this point.

I got them for £180 each with tax (R9 290X starts at £370 each with tax) - 3 of these cost less than 2 290X's and when they *work* perform a little faster overall. I appreciate they are older silicone (AMD's Overdrive utility lists them as a 7970 for example). Performance is great in games at my target res, as long as we stay above 60 for the most part anything after that becomes irrelevant. Remember I'm also hitting smooth frame rates with crazy levels of filtering and AA applied. I can always dial back the settings on AA especially if games stress it in the future?

Take the good with the bad I guess. GRID Autosport for example runs smoothly. Assetto Corsa does not - Crossfire enabled with either 2 or 3 cards is a stuttering nightmare. That's just one game tho, and it's beta. Yes, things are patchy if the game isn't designed for more than one card, but that's always been the case.

Comments about the R9 280X being old are kind of silly.
Most things on the market are old, right now. Kepler, Tahiti, Pitcairn. The GK110 and Hawaii chips may be a little more recent but they're not exactly spring chickens and they are obnoxiousy large, hot, noisy and power hungry for a barely-justifiable bump in performance.

I refuse to buy a 780Ti or 290X because they offer such terrible value for money and my pair of 280X cards do fine. I almost never use them in Crossfire and if I'd had to pay for them myself I'd have been perfectly content with just one (for my 1440p gaming resolution).

Well said Chripsy. I'm happy with how it performs, despite the heat it chucks out - My case is venting it ok and nothing is being clock throttled so it's not bothering me.

I begin my overclocking adventure this morning tho, and have not got very far. My 8350 either isn't a very good overclocking sample, or the board isn't keeping the power supply up to what's needed. I also think I hit a thermal limit past about 4.5Ghz and 60C, cores were shutting down in Prime95.

It's stock speed is 4.0Ghz, and the board chose voltage of 1.325 to be appropriate from day 1. I'm working up from there, and started straight in at it's turbo core speed on all 8 (temps are loaded values when Prime95 is running Blend mode):

4.20Ghz | 1.325v | 34C - Pass
4.25Ghz | 1.325v | 35C - Pass
4.30Ghz | 1.325v | 38C - Pass
4.35Ghz | 1.325v | 42C - Pass
4.40Ghz | 1.325v | 47C - Pass
4.45Ghz | 1.325v | 53C - Failed (Random errors on several cores)
4.45Ghz | 1.375v | 57C - Pass
4.50Ghz | 1.375v | 58C - Pass
4.55Ghz | 1.375v | 59C - Failed (Random errors on several cores)
4.55Ghz | 1.405v | 64C - Failed (Cores shut down to idle)
4.55Ghz | 1.455v | 69C - Failed (Wouldn't POST)

Because of temps, I've stuck with 4.4Ghz at stock voltages for now. The additional 100Mhz isn't worth the significant temp increase that come with it.

Edit: The graphics card will NOT be overclocked. They already run too close to thermal limits if anyone was expecting to see that.

I wouldn't worry about temps in the 50s. When I run the Folding@home client on FX series CPUs I tune the fan profile to keep the temps just below 60C, as this minimizes fan noise while keeping core temperatures (barely) within spec. I've never seen any stability issues from doing this.

geekl33tgamer wrote:

Well said Chripsy. I'm happy with how it performs, despite the heat it chucks out - My case is venting it ok and nothing is being clock throttled so it's not bothering me.

I begin my overclocking adventure this morning tho, and have not got very far. My 8350 either isn't a very good overclocking sample, or the board isn't keeping the power supply up to what's needed. I also think I hit a thermal limit past about 4.5Ghz and 60C, cores were shutting down in Prime95.

It's stock speed is 4.0Ghz, and the board chose voltage of 1.325 to be appropriate from day 1. I'm working up from there, and started straight in at it's turbo core speed on all 8 (temps are loaded values when Prime95 is running Blend mode):

4.20Ghz | 1.325v | 34C - Pass
4.25Ghz | 1.325v | 35C - Pass
4.30Ghz | 1.325v | 38C - Pass
4.35Ghz | 1.325v | 42C - Pass
4.40Ghz | 1.325v | 47C - Pass
4.45Ghz | 1.325v | 53C - Failed (Random errors on several cores)
4.45Ghz | 1.375v | 57C - Pass
4.50Ghz | 1.375v | 58C - Pass
4.55Ghz | 1.375v | 59C - Failed (Random errors on several cores)
4.55Ghz | 1.405v | 64C - Failed (Cores shut down to idle)
4.55Ghz | 1.455v | 69C - Failed (Wouldn't POST)

Because of temps, I've stuck with 4.4Ghz at stock voltages for now. The additional 100Mhz isn't worth the significant temp increase that come with it.

Edit: The graphics card will NOT be overclocked. They already run too close to thermal limits if anyone was expecting to see that.

I needed 1.52Volts on my 8120 to hit 4.5GHz stable, although it's a slightly different class of chip. This was under water with a triple-rad and EK Block. Can't remember exact temps but it was warm. Highest I managed was 4.7GHz briefly for benching but voltage was far too high, 4.8GHz was simply a no-go on any combination of HTT/Multi/Voltage and I spent ages fiddling with settings. Was a fun chip though to play with

As for your CPU temps I am surprised you went with a top down cooler instead of a single wide or twin tower cooler that gives you so much more fin area with the same amont of heatpipes to pull heat off your CPU and aim it directly out the rear exhaust fan.

But your temps look fine..now test it with a full GPU load and CPU load. Furmark and say prime95 just make sure your PSU+ heat sink can handle a full overload since using furmark and prime loads the system to the max. A good bit more then just gaming a heavy CPU game. It really is not a realistic load but it will tell you if your wattage is good enough on your PSU.

Sorry I forgot how many watts and make you PSU is and if is has a single 12v rail or multiple 12v rails. With the GPUs using around 200+ watts each and a CPU that uses around 150 loaded.

A good way to lower your GPU wattage use is to enable v-sync since you have 60htz monitors there is no reason to make the GPU's produce frame that never even get used and just get flushed down out of the buffer.

On my SLI 770 rig enabling vsync on most everything reduces power draw and GPU temps dramatically.

vargis14 wrote:

As for your CPU temps I am surprised you went with a top down cooler instead of a single wide or twin tower cooler that gives you so much more fin area with the same amont of heatpipes to pull heat off your CPU and aim it directly out the rear exhaust fan.

But your temps look fine..now test it with a full GPU load and CPU load. Furmark and say prime95 just make sure your PSU+ heat sink can handle a full overload since using furmark and prime loads the system to the max. A good bit more then just gaming a heavy CPU game. It really is not a realistic load but it will tell you if your wattage is good enough on your PSU.

Sorry I forgot how many watts and make you PSU is and if is has a single 12v rail or multiple 12v rails. With the GPUs using around 200+ watts each and a CPU that uses around 150 loaded.

A good way to lower your GPU wattage use is to enable v-sync since you have 60htz monitors there is no reason to make the GPU's produce frame that never even get used and just get flushed down out of the buffer.

On my SLI 770 rig enabling vsync on most everything reduces power draw and GPU temps dramatically.

It'll be over 150W @ 4.55GHz 1.45V more like 200W. They're 125W to begin with.

JBI, I assume 60C is the chips cut-off point? It was fine up to that then went off a cliff, lol.

vargis14 wrote:

As for your CPU temps I am surprised you went with a top down cooler instead of a single wide or twin tower cooler that gives you so much more fin area with the same amont of heatpipes to pull heat off your CPU and aim it directly out the rear exhaust fan.

My case has an unusual dual-fan vent on the side panel. The depth of the fans when the panel is in position internally protrudes downwards into the CPU cooler area by 18mm. Not a huge problem by itself, but there's a small circuit board up there too that controls the power to them, and connects it to a mobo PWM header. Not a lot I know, but enough to cause a clearance issue with most tower coolers (My old Zalman CNPS tower wont fit for example - it's 145mm tall). The highest one can be about 90-100mm as an absolute maximum, or I forgo some of the venting benefits it gives the graphics instead (both fans exhaust air, and are PWM controlled)?

vargis14 wrote:

Sorry I forgot how many watts and make you PSU is and if is has a single 12v rail or multiple 12v rails. With the GPUs using around 200+ watts each and a CPU that uses around 150 loaded.

I don't haven any specs on what it can handle, but it's a SuperFlower Leadex Gold 1300W Fully Modular "80 Plus Gold" Power Supply on the store page I bought it from. I don't know much about PSU's, other than peeps on here saying SuperFlower is a decent brand, you need modular, and get as much wattage as you can afford. So I did.

vargis14 wrote:

A good way to lower your GPU wattage use is to enable v-sync since you have 60htz monitors there is no reason to make the GPU's produce frame that never even get used and just get flushed down out of the buffer.

I usually have it on to remove the tearing when to many frames are rendered. It appears jittery otherwise (kinda hard to explain). Never knew it helps temps though (unless of course the 3 cards cant hit 60 - I assume they go all our then?)...

geekl33tgamer wrote:

JBI, I assume 60C is the chips cut-off point? It was fine up to that then went off a cliff, lol.

I believe the specified upper temperature is 62C, but I am not 100% sure on that. The fact that yours didn't like it is probably a combination of luck of the draw, and the particulars of your OC. I've had mine above 60C (and never saw any instability), but that was at stock clocks and voltage while I was playing around with the fan speed profiles to find the best tradeoff between heat and noise.

As an aside, a few years ago I (accidentally) ran my Phenom 9550 all the way up to 100C. The CPU fan died and I didn't even notice, since the system never became unstable. I have no idea how long it ran like that, and don't understand why it didn't shut down. But it does go to show that AMD's CPUs (at least as of the Phenom generation) were pretty darned tough. Ever since then, I've paid more attention to monitoring CPU temperatures and fan RPMs!

geekl33tgamer wrote:

4.40Ghz | 1.325v | 47C - Pass
4.45Ghz | 1.325v | 53C - Failed (Random errors on several cores)
4.45Ghz | 1.375v | 57C - Pass
4.50Ghz | 1.375v | 58C - Pass
4.55Ghz | 1.375v | 59C - Failed (Random errors on several cores)

I would expect a slightly higher overclock. Maybe the M/B or PSU cannot keep voltages sufficiently stable with 3x GPUs. Anyway, 4.4GHz is a good compromise of performance/heat and stability. As I said before, try undervolting the GPUs. That will make a huge difference in noise and heat.