Sapphire R9 390 Nitro overheating

[gupsterg]

As you bought card Nov 15 you will have warranty til Nov 17, so yeah RMA if you like. Yeah card should hold factory OC, I would expect that as well.

They do test GPU in factory but not sometimes to the extent we may think. As it's an OC edition card, Sapphire have set a +19mV GPU core voltage offset, it maybe over time now the card needs a bit more. In the RX 480 they have setup voltage recalibration, as the silicon ages it may require more juice than when new (marketing slide link), previous gen GPU had no such feature.

[8bit]

I tried the game last night with the card set to standard, reference clock speeds and voltages for a regular R9 390 and it crashed pretty quickly there too. I've opened a ticket with the developer as I'm thinking this might be an issue with the game now but I'll do some further testing with Elite tonight since I saw a similar crash there too.

[gupsterg]

Surprised the card crashed at reference clocks.

When you say set it to standard voltage how would you know for the LeakageID,etc what is the VID of the GPU for a DPM state? Below is a table for 290X (Hawaii XT), this is the AMD GoldenDB data that can be found in a ROM, if we view DPM 7 (highest state) we can see a Low LeakageID GPU will be 1.281V VID and a high LeakageID 1.206V. There are not only 3 LeakageID bands, so more intermediate VIDs exist.





Like I said before the ROM & the driver assesses what to set the GPU's VID per DPM, using several factors. Another factor is GPU clock in ROM, for example deviation of AMD stock clocks in ROM will create VIDs which may not be "true" VID of the GPU. Take the 290, AMD stock DPM 7 is 947 MHz, when a ROM has 1000MHz on a OC card the DPM 7 VID will differ to 947MHz, usually it reduces as GPU clock is increased in ROM from testing I did, hence OC cards have the + voltage offset.

We can use AiDA64 to gain the EVV VID per DPM, view this video. You must enable status bar in the view menu to be able to right click it to bring up registers menu. Even trial version will get you the info, download and use latest version. The registers dump excludes global voltage offset set via ROM/IR3567B (the voltage control chip), ie the +19mV you see in TriXX.

If you do set the card at reference clocks don't adjust the voltage IMO.

Just wondering can you run say Heaven or Valley looped on your system approx 30min?

[8bit]

Re. standard voltage, I mean setting the GPU Voltage control in TRIXX to 0mV - I presumed the default +19mV was over and above standard voltage for a non-OC'd card, maybe that was wrong.

I just tried the Valley benchmark and it ran for about four minutes before the black screen of death. I was out of the room at the exact time of the crash but I've pasted the last 30 seconds of GPU-Z output which was running at the time. Nothing looks worrying to me, only thing is the 12V value is slightly under 12 volts but I don't know if that's an issue. This was run with the card at default settings, 1040MHz GPU, 1500MHz VRAM and +19mV.

        Date        	 GPU Core Clock [MHz] 	 GPU Memory Clock [MHz] 	 GPU Temperature [°C] 	 Fan Speed (%) [%] 	 Fan Speed (RPM) [RPM] 	 GPU Load [%] 	 Memory Controller Load [%] 	 Memory Usage (Dedicated) [MB] 	 Memory Usage (Dynamic) [MB] 	 12V [V] 	 VDDC [V] 	 VDDCI [V] 	 VDDC Current In [A] 	 VDDC Current Out [A] 	 VDDC Power In [W] 	 VDDC Power Out [W] 	 VRM Temperature 1 [°C] 	 VRM Temperature 2 [°C] 
2016-09-14 22:28:26 	985	1500	77	56	2280	100	48	1129	58	11.63	1.094	1.047	19.1	170	228.3	178.5	83	77
2016-09-14 22:28:27 	984.6	1500	77	56	2285	99	36	1120	58	11.63	1.148	1.047	23.1	177	271.5	221.3	83	77
2016-09-14 22:28:28 	1030.8	1500	77	56	2280	99	48	1120	58	11.5	1.141	1.047	22.6	181.5	297	224.8	83	77
2016-09-14 22:28:29 	985.2	1500	77	56	2280	100	55	1120	58	11.63	1.094	1.047	20.9	181	263	209.8	83	77
2016-09-14 22:28:30 	1031.7	1500	77	56	2275	100	14	1120	58	11.5	1.133	1.047	24.6	189.5	287	204.5	83	78
2016-09-14 22:28:31 	1034.8	1500	77	56	2275	100	45	1120	58	11.63	1.156	1.047	24.3	201.5	290.5	215	84	78
2016-09-14 22:28:32 	1014.8	1500	77	56	2280	98	39	1120	58	11.5	1.141	1.047	22.7	186.5	302	220.5	84	78
2016-09-14 22:28:33 	994.9	1500	77	56	2275	99	37	1120	58	11.63	1.086	1.047	19.7	150.5	216.8	180	84	78
2016-09-14 22:28:34 	995.8	1500	77	56	2275	99	38	1120	58	11.5	1.141	1.047	24.4	174.5	265.5	223	84	78
2016-09-14 22:28:35 	1029.1	1500	77	56	2280	99	26	1121	58	11.5	1.141	1.047	22.6	177	279	228.5	84	78
2016-09-14 22:28:36 	1027.4	1500	77	56	2280	99	53	1122	58	11.5	1.141	1.047	24.8	185	263.5	206.3	84	78
2016-09-14 22:28:37 	992.6	1500	77	56	2270	99	27	1122	58	11.63	1.078	1.047	19.7	170	238	171.5	84	78
2016-09-14 22:28:38 	997.8	1500	77	56	2275	99	24	1122	58	11.63	1.086	1.047	19.9	164.5	218.5	168.3	84	78
2016-09-14 22:28:39 	1034.5	1500	77	56	2280	100	26	1122	58	11.5	1.063	1.047	20.3	158	233.8	189	84	78
2016-09-14 22:28:40 	1015.7	1500	77	56	2275	100	50	1122	58	11.5	1.141	1.047	22.8	180.5	283.5	229.8	84	78
2016-09-14 22:28:41 	1023.7	1500	77	56	2275	100	34	1122	58	11.63	1.148	1.047	26.6	195.5	293	222.3	84	78
2016-09-14 22:28:42 	1015	1500	77	56	2269	96	39	1122	58	11.5	1.141	1.047	25.6	191	266.5	208	84	78
2016-09-14 22:28:43 	1014.9	1500	77	56	2280	99	22	1122	58	11.63	1.078	1.047	19.5	153	221.3	180.5	85	78
2016-09-14 22:28:44 	991.4	1500	77	56	2280	100	50	1122	58	11.5	1.148	1.047	23.2	195	294	222	85	78
2016-09-14 22:28:45 	1028.3	1500	77	56	2274	99	23	1122	58	11.5	1.141	1.047	23.6	180.5	284	223	85	78
2016-09-14 22:28:46 	826.4	1500	75	56	2290	100	18	1145	58	11.63	1.023	1.047	17.8	148	193.8	160.8	85	78
2016-09-14 22:28:47 	986	1500	77	56	2280	100	45	1132	58	11.5	1.141	1.047	23.2	180	284	225.5	84	78
2016-09-14 22:28:48 	1009	1500	77	56	2280	100	34	1125	58	11.63	1.078	1.047	20.6	168	219.3	169	84	78
2016-09-14 22:28:49 	981.4	1500	77	56	2280	100	44	1118	58	11.63	1.078	1.047	23.7	186	270	206.3	84	78
2016-09-14 22:28:50 	1008.8	1500	77	56	2274	100	83	1118	58	11.63	1.148	1.047	25.5	204	308.5	221.3	85	78
2016-09-14 22:28:51 	984.8	1500	77	56	2274	100	66	1118	58	11.63	1.094	1.047	19.7	173.5	261.5	188.3	85	79
2016-09-14 22:28:52 	985.4	1500	77	56	2275	100	25	1118	58	11.63	1.086	1.047	18.9	157.5	242	217.3	85	79
2016-09-14 22:28:53 	992	1500	77	56	2274	100	35	1118	58	11.63	1.094	1.047	18.9	163	245.8	194	85	79
2016-09-14 22:28:54 	1021.6	1500	77	56	2274	100	67	1118	58	11.5	1.148	1.047	25.6	203.5	319	224.3	85	79
2016-09-14 22:28:55 	984.7	1500	77	56	2274	100	45	1118	58	11.63	1.094	1.047	18.9	164	243.8	190.5	85	79

[8bit]

I just clocked the RAM back to 1200MHz (all other settings default) and tested No Man's Sky. Played for 35 mins with no crash. Not sure what that tells us though, possible bad RAM chip on the card?

[gupsterg]

AMD R9 390 stock is supposed to be 1000MHz GPU clock, so I would assume as ROM has 1040MHz the VID for DPM 7 will be lower than if it was 1000MHz, I conducted such tests on my Hawaii cards. There is information like this in a Hawaii bios mod thread I did on Overclock.net.

Don't worry about the 12V rail showing as 11.xxV , I had 4 differing cards and none showed correct V in monitoring , I confirmed all was well via using a digital multimeter on the PCI-E plugs .

390/X has newer RAM IC revisions than 290/X, they are 1500MHz spec, so only your GPU is OC'd from factory. Currently I don't think you have bad RAM IC on graphics card. I reckon your GPU voltage is too low. If I was in your shoes I would carry on testing upping RAM clock in 50MHz steps until you have the issue and not at all meddle with GPU clock/voltage. I have seen a lot of 390/X owners on OCN achieve easily 1650MHz+ RAM clock as a 24/7 usage OC plus increased GPU clock.

Just wondering have you ever OC'd card? I only ask to see if you've tested what headroom GPU/RAM has plus how it reacts to OC.

[8bit]

No, never overclocked the card at all.

Didn't get a chance to play with this last night, I'll try underclocking the RAM again and see if I get any better result with that Valley benchmark.

[8bit]

So last night I tried this again, underclocked the GPU memory to 1200MHz and ran the Valley benchmark again. It ran for about 25 minutes this time but then black screened as before. Started filling out a support ticket for Sapphire, while checking the motherboard BIOS version I noticed there was a new one available which said it "improved PCIe compatibility" so I decided to try that.

That installed fine so I tried the card at factory clocks etc. and ran the valley benchmark. After about 5 minutes there was a loud POP and all the sockets in the house tripped at the fuse box. I haven't tried firing the machine up again yet but I suspect the power supply has blown.

So I guess this is just a theory for now but maybe the "graphics issues" were just a symptom of a fault in the PSU which causing output to dip when under load before eventually giving up altogether. I think I've got a couple of older power supplies stashed in a cupboard at home, I'll try one of them if it does look like the current one is terminal.

[gupsterg]

Sorry to read this, hope the rest of the system is OK.

[8bit]

Cheers Thankfully the rest of the PC survived intact and it definitely was the PSU that went bang, I swapped in an old Coolermaster 550w one and it's working again. Better yet, the graphics crashes haven't come back, I played No Man's Sky for over an hour last night and ran the Unigine Valley benchmark for about 25 minutes tonight.

Thanks everyone for your help, especially gupsterg. Looks like it wasn't the graphics card after all but I still learned some useful stuff