It had been years since I built a gaming machine and over the past three months I have cobbled a new one together as i forgot how much i enjoyed the building process. My build is as follows:
Asus ROG STRIX Z370-E
16 GB Corsair Vengence RGB DDR4
2 x Corsair Force LS SSD
2 x Corsair Force GT SSD
2 x Lite On SSD
Thermaltake Core P5
EKWB Custom Loop Cooling
XFX RX 580 8GB (Primary) (Purchased when RMA took three months)
XFX RX 570 4GB (Linked) (From RMA'd RX 560)
I have a couple of questions. One, is that I was under the impression that when running in crossfire mode all cards follow the lowest common denominator, so in my case they should both act like the RX 570 correct? That did not seem to be the case at all. I did some testing in Far Cry 5 and while it absolutely shows that it was running in Crossfire, it was doing so at the RX 580 speed and using all 8GB of VRAM. Did I do something wrong?
My second question is that when I go into Radeon settings I see that the core clock and memory speed of the RX 570 are wonky, it says 1100 core clock and 1850 memory speed when it should say 1286 core clock and 1750 memory speed. What could cause this, is it a function of the Crossfire feature?
I am thrilled with the results but want to make sure I am not doing something wrong.
Far Cry 5 (3440x1440, Ultra, RX 580 Only)
Minimum FPS = 36
Maximum FPS = 52
Average FPS = 41
Far Cry 5 (3440x1440, Ultra, RX 580 + RX 570 CF)
Minimum FPS = 43 (+19%)
Maximum FPS = 79 (+52%)
Average FPS = 63 (+50%)
As I said the gains were awesome to see from the entire situation.
So to answer your questions:
Remember that the crossfire behave differently depending the game, engine used.
Also it mean that the crossfire will not behave the same, using different versions of direct X.
So the crossfire scaling depend from a lot of factors, game, engine, directX, hardware, etc.
If you take FC5 as example, it run an engine under DX12 with enhanced features for crossfire and multi-gpu in general.
So on optimized games, the engine, DX/Os recognize both card separately loading them both equally with great results!
On the other hand, if the environment is not well optimized, the scaling is lower, the cards are not evenly used and ran on the slowest one.
That's why the crossfire is a great feature but not always easy to max the fps scaling, maybe it will evolve into multi-gpu for wide ranges purposes in future.
What really count and matter is your clocks under load. Download GPU-z it display T°, voltages monitoring.
Also you could DDU to clean the driver and reinstall the latest ones, maybe the card do not boost to it's max clocks when in crossfire.
I would advice you to have aslo some fun benching 3Dmark, Heaven, Superposition checking your crossfire behaviour.
If you feel something unusual, you can also report it using the AMD Report Form, easy to google it.
Thanks for the response. I figured given the amount of time that Far Cry 5 has spent being advertised in the Radeon settings it was set up to be optimized for Crossfire. I looked it up the two games I primarily play Far Cry 5 and Project Cars 2 which both have decent Crossfire profiles.
I am starting fresh with a fresh installation of Windows 10 64-bit and everything else. I will do as you suggested and play with the benchmarks and see what happens with and without crossfire enabled.
That was actually a lot of fun running the benchmarks last night. I turned on the Crossfire logo so I could ensure when it was actually being used in the benchmarks. Interestingly enough the 3DMark one, the logo did not appear on with Crossfire enabled so I am assuming that it was not on. If it was then there was a negligible difference between the two.
As for Heaven and Valley the difference was amazing. I was getting on average +70% performance gains from Crossfire! I am not sure how to do a screen capture of the video but it was very evident.
Heaven and Valley limited themselves to 4GB of memory so that was following the lowest common denominator however in Project Cars 2 and Far Cry 5 the GPU-Z Log showed topping out in the 7GB range, so that is awesome.
Also the clean install was the key to everything. The 570 is now clocking @ 1386 Mhz on the core and 1750 Mhz on the memory in the Radeon settings information. Something unexpected did happen however during both benchmarks and game play in Far Cry 5 and Project Cars 2. The logs show the RX 570 running at 1425 Mhz on the core and 2025 Mhz on the memory. This is the OC+ numbers that are in Wattman for the RX 580. The 3Dmark system information also showed something similar and instead of two different cards showed 2 x RX 580.
Is the diffeence between the 570 and the 580 only speed or is there something else that I am missing?
I'm happy you took some good time benching, it's always fun check how the hardware behave in different scenarios.
To answer to your question, the only difference between the RX580 and the RX570 is the number of CU within the gpu.
The RX580 care 2304 Steam Processors so 36CU, the RX570 2048SP so 32CU, but the chip itself is the same.
Then the RX570 care maybe some other kind of memory chip and have a lower TDP, power consumption.
About the clocks, it's quite unexpected for the RX570 to boost so high, especially because the best silicon would go toward the RX580 cards.
That's also why the RX580 have higher boost clocks than the RX570, this not mean that the RX570 can't achieve these clocks.
It would maybe required more voltage to do so and suit better to a RX570 card class.
So not sure the software read correctly both card, maybe it mistake with the RX580? What GPU-z said about, you have a tab for each card on the bottom?
Last thing, dunno if you can but try also other games in DX11/DX12 and superposition benchmark, an updated version of heaven/valley.
Also not sure why you would not have any scaling in 3Dmark, it should scale quite nicely.
I decided to start over again at the suggestion of XFX, the card manufacturer, with the installation of everything. What they wanted was me to install one card at a time and then restart each time. After doing so I ran benchmarks again and am getting similar results. The only change is that the audio driver for the 570 is now having issues. I cannot believe it is doing so well in general though.
In general I am seeing a 50-70% increase in performance using Crossfire. All the games I play seem to scale well. What I do find interesting is that the system needs to start using crossfire takes about 2 seconds at the beginning of games or benchmarks.
Thank you for all your help wimpzilla. I really do appreciate it.
Well XFX guy was peaking sense it seems, what he advices you to do is the general good practice to troubleshoot crossfire issues.
Usually you plug a card 1st, then thge second at reboot, usually it help solving crossfire detection issue.
But this procedure is meant to solve issues, not boosting the performances.
So i would say your crossfire is working fine, 50/70% of gain is what usually one get, depending drivers and games.
If you got and issue with the HDMI audio, just DDU the drivers with the safe mode option, then reinstall the latest ones with both card plugged, see if it change something.
I'm not aware of any delays, my last crossfire was HD4870/HD4890, but i can inquire about and let you know!
Remember that the crossfire is really games and driver dependant, so when you have time and want to have fun, compare old vs new drivers, there is always something to discover about, i suppose. ^^
You are welcome!
Isn't FC5 a DX11 game?
You are right, it run under boosted DX11 API.
My bad i thought it was under DX12 API.
Still even if DX11, the game engine is very well made.
So unfortunately I really screwed the system up. I guess my 850 watt PSU was not enough for my system and under heavy gaming load, the whole system shutdown and would not power on again. I think I may be overdoing it with RGB LEDs, currently 22 Corsair LED strips installed on three Lighting Node Pros and 3 Corsair LL120 RGB fans. When i run the PSU calculator it says I am pulling 768 watts under full load and well, I think it fried things. So the RX 580 now does not allow the PC to boot up. If I disconnect that card from the PSU and plug the 570 into the same cable it works. Even a different power supply test showed that the card would not allow the system to power up. I am hoping that I can get the card RMA'd by XFX at this point I was really enjoying the performance.
On a separate note this issue actually affected the 580 and the Motherboard only, now the USB 3.0 ports and Wi-Fi don't work. Did I overload the rail on the PSU and cause this? I know it is my fault just trying to figure out how not to do it again.
I'm really sad to learn about, i was happy to check your answer but was not expecting that, i'm truly sorry.
First, i don't think it is your fault, your PSU is a branded Corsair 850Watt, you can pull far more than 768Watt on that psu.
Counting that the 8700K even OC full load gaming is maximum 300Watt, both the gpu another 350Watt maximum.
You are not even reaching 800Watt with the core pieces of the rig, including mb, ram and SSD drives.
So you are far from reaching the full capacity of your psu, that could even deliver 900Watt peak for a short time.
Now i'm more interested about the RGB's and RGB's controller. Did the RGB controller are plugged into the USB?
Did it smell somewhere the usual burnt components?
I'm not expert about RGB dunno how much power these strips use, a couple of them is not an issue for sure, but 22 dunno?
I know these strips run on 5/12v, you could have overloaded the 5/12v rails, but the psu would be smart enough to detect any current overload.
Still a voltage/current spike could happen if the psu is reaching his critical limit, causing damages.
Usually the psu ,especially the branded ones, are build for avoid this kind of nasty situations.
In fact you psu did not died, but something definitively happened between the psu, gpu and motherboard.
So to me it seems something shorted out and/or overloaded, i really don't know if the gpu failed because you were pushing, again your psu should hold the load.
Could be also some RGB strips that caused issues, since the controller is connected over USB, that isn't working now.
Tbh when a gpu die by his own, rarely it take with it other part of the motherboard.
Either the mb die with the gpu/psu, either the mb is fast and smart enough to save itself and the rest.
Here you have 2 different components that are all linked to the pci-e root, the USB 3.0 chip hub and the Wifi chip that died.
So either in your case you were unlucky and the gpu died taking with it some pci-e stuff, even if not on the same bus (CPU/Chipset) but share some hardware.
Either something on the USB shorted/overloaded, killing the USB 3.0 hub, the WIfi chip, that are connected in the same way by pci-e, finally killing also the gpu.
I don't think it is you fault, check carefully each part for components damage and sent what is not working back to warranty if apply.
I have requested RMA's on the Motherboard and the Graphics card now, hopefully they will offer to replace or repair at this point. I believe you are onto something with those LEDs. The PSU I had was an HX850i which I thought should be able to handle what I was doing. But maybe this is why I have had problems all along with LEDs and other USB devices. Below is the picture from an online PSU calculator that I used to try and see just how off I am. I have used branded ones, but this one is more generic and is still recommending jumping up to the 1000 watt.
Never mind the amount I am paying each year to run this beast. However, it is perplexing that I am having issues. Basically all in all I have about 150 individually addressable RGB LEDs installed in the system, plus the ones that are already built into the motherboard and the ones in the CPU water block. Crazy enough that the HX850i has the ability to monitor input and output power through the Corsair software and I never saw it jump way about 550 watts. However I was never really looking at it during heavy gaming obviously. Actually there are more LEDs on the RAM too.
The more I think about it, the more I am realizing that I may have actually overloaded the system. Who knows, I did get a new HX1000i just in case and will see when all the components get back if that solves the problem. Thank you for your help with my now not crossfire related issue. It is good to know that there are people in this community willing to jump in and help when they can.
This break down has given me the opportunity to redo the layout of my rig too. I am 3D Printing some parts that will be utilized to relocate the radiator from the area it is supposed to go on the Core P5 Case to being a top mount, pulling air in. I did notice that during use the 580 got way hotter than the 570 so I had added two smaller fans blowing on to the cards to try and keep them cool.
What maybe could represent an issue, is to run a lot of led strips on the same controller.
I mean you could get a 1000Watt i'm happy with that, but the parts you are using are very good, already your 850Watt is damn nice.
So maybe it's more about the led controller that is somehow overloaded because too many strips attached.
Since it is connected to the 5/12v and also through USB, any hardware issue on the led controller would impact the mb and the psu.
I did not spend time to calculate the single led strip power consumption, but as any electric parts, if you put too much of them in series you will stress the controller.
So i would advice you in future, to split the led strips into multiples controllers, to avoid too much current draw and stress on a single RGB controller.
I'm really not sure it's your fault, aside the led question, your psu is already enough for your system even if overclocked.
The psu would maybe not hit his best efficiency, but the power delivery as Ampere/Current is here.
I will be honest, i'm not a fan of RGB stuff, i'm more the old kind of hardware overclocker, still i would try to balance the RGB led load on more controller to be sure to not overload them.
I'm pretty sure that the cause of your issue could be more a mb, led issue rather than the gpu dying on it's own, quite unfortunate and happy to know you sent the parts back.
Again your are welcome, if i can somehow help.