Ryzen 5 1500X + AB350M Pro4
Printed From: ASRock.com
Category: OverClocking(OC) Zone
Forum Name: Achivement&Record
Forum Description: Share your HWBOT or Futuremark score here!
URL: https://forum.asrock.com/forum_posts.asp?TID=6971
Printed Date: 23 Nov 2024 at 4:09pm
Software Version: Web Wiz Forums 12.04 - http://www.webwizforums.com
Topic: Ryzen 5 1500X + AB350M Pro4
Posted By: PetrolHead
Subject: Ryzen 5 1500X + AB350M Pro4
Date Posted: 17 Dec 2017 at 3:20pm
|
A new rig, a new thread. Here are the specs: CPU: Ryzen 5 1500X GPU: EVGA GTX 1060 SC 6GB GDDR5 MB: ASRock AB350M Pro4 RAM: 2 x 8 GB G.Skill TridentZ (F4-3466C16D-16GTZSK) PSU: Corsair RM550x HDD: Samsung 960 EVO SSD (NVMe) 250GB + Samsung 850 EVO SSD 500GB Case: Phanteks Enthoo Evolv mATX Tempered Glass CPU Cooler: AMD Wraith Spire Fans: 2 x Phanteks 140 mm case coolers OS: Windows 10 64-bit At the moment I haven't pushed the system very far, since I'm interested in doing comparisons to my old rig at the same CPU clockspeed (even though it's a bit of an arbitrary limitation with CPU architecture, RAM speeds etc. being different). I've also left BIOS settings mostly to their default settings due to the fact that the manual doesn't explain any of them. So, here are the few things I've done: -Set CPU clockspeed to 3.7 GHz -Set core voltage to 1.25 V (seems to stay around 1.28 V when running prime95) -Set the RAM speed to 3200 MHz, since the XMP speed 3466 MHz didn't work (no surprises there) -Set RAM CL to 14 I haven't really stress tested the system - apart from a minute or two in prime95 just to test the temperature behaviour - since I only need it to be "bench stable" at the moment, but so far there have been no issues with stability. Compared to stock settings the CPU seems to run cooler(!) in prime95 and the Vcore has stopped bouncing around like a lunatic and reaching numbers above 1.4 V. The stock cooler does seem to be working very hard even with this four core CPU, and I'm hitting at least 75 C with these settings in prime95. I'm not sure if I want to push the CPU any further than that without upgrading the cooling... P.S. For some reason the CPU cooler seems to be at full blast all the time, even though the BIOS setting should allow it to slow down when the CPU is idle. I'll probably need to do a few checks before blaming the motherboard, but I'm a bit worried. I _really_ do not want to RMA the board. ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |
Replies:
Posted By: PetrolHead
Date Posted: 19 Dec 2017 at 2:10am
|
Okay, time to look at some CPU-Z comparisons. I'll post more benchmarks later. But first, these are the systems I'm comparing to each other: Old CPU: Phenom II X6 1090T @ 3.7GHz GPU: EVGA GTX 1060 SC 6GB GDDR5 @ stock RAM: 2 x 8 GB Kinsgston HyperX Fury Black @ 1600MHz CL9 HDD: Samsung 850 EVO SSD 250GB + Samsung 850 EVO SSD 500GB OS: Windows 10 64-bit New CPU: Ryzen 5 1500X @ 3.7GHz GPU: EVGA GTX 1060 SC 6GB GDDR5 @ stock RAM: 2 x 8 GB G.Skill TridentZ @ 3200MHz CL14 HDD: Samsung 960 EVO SSD (NVMe) 250GB + Samsung 850 EVO SSD 500GB OS: Windows 10 64-bit Neither system was maxed out, but I preferred running the Phenom at 3.7GHz and wanted to see what the Ryzen would be like at the same clock speed, even though it will still be an apples to oranges comparison due to all the other factors. Anyways, here are the averages of five benchmark runs for CPU-Z 1.82: Old CPU Single Thread: 179.36 CPU Multi Thread: 1078.20 New CPU Single Thread: 426.42 CPU Multi Thread: 2294.32 So, an increase of 137% and 113%. Not bad. However, as l will show later, the gains vary a lot depending on the benchmark, and games especially can be totally oblivious to the new CPU due to a GPU bottleneck. ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |
Posted By: PetrolHead
Date Posted: 26 Dec 2017 at 7:48pm
|
More benchmark results. Again, all results are the average of five benchmarks runs and the systems are as described in the previous post. Cinebench R15 Old OpenGL: 69.896 fps CPU: 538.4 cb CPU (Single core): 95.4 cb MP Ratio: 5.66x New OpenGL: 115.248 fps CPU: 824.6 cb CPU (Single core): 150.8 cb MP Ratio: 5.47x Thus, a 56%-65% increase and an MP ratio that's pretty close to the six-core Phenom II. Cinebench doesn't scale very well with threads (for an Intel Xeon with 12C/24T the MP ratio is below 14, and a similar trend is seen for other Intel CPUs as well), so the Ryzen results are as expected. There are two noteworthy things about these results: 1. The first multi-core result with the Phenom II was only 502, without which the average (of four results) would have been about 10 cb higher and the MP ratio would have been over 5.7x. The single-core result for that run was in line with the other results, so I can only assume Windows was doing something in the background when I started the benchmark. 2. The OpenGL results were worse on each run with both CPUs. With the Phenom II the results were (first/last) 73.42 fps/ 67.68 fps and with the Ryzen they were 126.98 fps/109.98 fps. It's not because of thermal throttling, so I'm not quite sure what is going on. Maybe quitting Cinebench and launching it again between every bencmark would have solved this issue, but I'll have to test that later. Edit: Found a typo in the Cinebench results for Ryzen. ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |
Posted By: PetrolHead
Date Posted: 27 Dec 2017 at 4:38am
|
Next up with the averages of five runs: Passmark 9.0 Old Total: 3094* (*this is the average of four values, since I forgot to write down the total points for the first run) CPU Mark: 6149.98 2D Graphics Mark: 551.58 3D Graphics Mark: 7641.66 Memory Mark: 1295.94 Disk Mark: 4043.28 For the CPU Mark, I also took the more detailed values: Integer Math: 10374.2 Floating Point Math: 7508 Prime Numbers: 21.4 Extended Instructions (SSE): 29 Compression: 8093.4 Encryption: 1451.6 Physics: 335 Sorting: 5734.8 CPU Single Threaded: 1300.6 New Total: 5287.7 CPU Mark: 10674.8 2D Graphics Mark: 872.58 3D Graphics Mark: 11031.1 Memory Mark: 2178.12 Disk Mark: 11806.84 Integer Math: 21671.8 Floating Point Math: 7992 Prime Numbers: 36.4* (*this contains an outlier value of 30, with the rest being 36-39 and the Phenom also exhibiting only a ~10% difference between runs) Extended Instructions (SSE): 383 Compression: 13391 Encryption: 2117.8 Physics: 638.4 Sorting: 8056.6 CPU Single Threaded: 1970.2 A lot of values to compare, but for the CPU single threaded we're looking at a slightly over 50% increase and for the CPU mark a slightly over 70% increase. There's a nice increase in the overall system score as well, but the 3D graphics mark is strangely CPU bound (in other words it's too easy for the GPU) and the disk mark varied roughly 30% between tests on my new system, so not all the numbers are very reliable or indicative of real world performance gains. ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |
Posted By: PetrolHead
Date Posted: 28 Dec 2017 at 2:13am
|
The series continues. All results again the average of five benchmark runs. Geekbench 4.2 Old OpenCL Score: 131094 (Single-Core Score / Multi-Core Score) Total Score: 2425.8 / 9565 Crypto Score: 153.8 / 908.2 Integer Score: 2805.6 / 13045.4 Floating Point Score: 1935.6 / 9826.4 Memory Score: 2874.4 / 3506.4 New OpenCL Score: 133389.2 (Single-Core Score / Multi-Core Score) Total Score: 4356.4 / 15474.6 Crypto Score: 5695.4 / 14259.6 Integer Score: 4059.4 / 18739.6 Floating Point Score: 3959.2 / 16889.4 Memory Score: 5285.6 / 6309.6 Here we see a ~80% increase in single-core score and ~62% increase in multi-core score for the total result. The crypto results for the Phenom II are abysmal in comparison, possibly due to a more limited collection of instruction sets. Single-threaded floating point performance is in turn doubled when moving to Ryzen, whereas (single-threaded) integer performance sees a comparatively mild 45% increase. The OpenCL test is a separate test and didn't affect the CPU results. The increase is roughly 2%, which is not surprising. After all, the benchmark should measure GPU performance. ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |
Posted By: PetrolHead
Date Posted: 28 Dec 2017 at 6:06am
|
Let's move on to a few gaming related benchmarks. First up Firestrike (averages of five runs): Old 3DMark Score: 9628.2 Graphics Score: 13367 Physics Score: 7807 Combined Score: 3504.6 Graphics Test 1: 63.514 fps Graphics Test 2: 53.57 fps Physics Test: 24.788 fps Combined Test: 16.304 fps New 3DMark Score: 11312.8 Graphics Score: 13504 Physics Score: 12454.4 Combined Score: 4805.8 Graphics Test 1: 64.292 fps Graphics Test 2: 54.028 fps Physics Test: 39.542 fps Combined Test: 22.354 fps Nothing that surprising here. In the graphics tests the results are in practice the same, in the physics test there's a big difference and in the combined test there's a smaller, yet still notable difference. There's another way of seeing these results: In graphically demanding games - note that this was just Firestrike - my GPU was already performing at its limit with the old Phenom II. The Ryzen upgrade only helps in situations where the Phenom II would be stressed to its limit due to complex physics/AI or where my GPU is overkill and the CPU sets the pace. Unfortunately I didn't benchmark any such games. I was, however curious to see whether the upgrade would have an effect on the lowest FPS figures in a game I was about to start playing, so I did one gaming benchmark: Metro Last Light Redux (3.00 x64), max settings @1080p (Averages of five benchmark runs) Old Total Frames: 8573.6 Total Time: 171.0154 Average Framerate: 50.168 fps Maximum Framerate: 112.44 fps Minimum Framerate: 15.144 New Total Frames: 8814 Total Time: 171.15328 Average Framerate: 51.534 fps Maximum Framerate: 162.356 fps Minimum Framerate: 15.004 Not much of a difference, except for the maximum framerate. The minimum framerate is practically the same, although it has to be said that with the Phenom II it varied between 13.72 fps and 16.27 fps, whereas with the Ryzen 5 it varied between 9.73 fps and 18.25 fps. Maybe more benchmark runs would have given a more representative average, maybe the test is just sensitive to whatever Windows is doing in the background or maybe frametimes just are a bit all over the place with my Ryzen 5 at its current settings. To sum up the benchmarks, the Ryzen 5 1500X is a good replacement for my old Phenom II, which was still totally usable despite its age. In gaming I am limited by my GPU at the moment, so that's what I'll upgrade next, but I could also try to overclock my GPU a bit further. As for the 1500X, I know there's still some performance to be squeezed out of it. I'll start with memory timings and see where I end up. I'll post the timings and some benchmark results when I'm ready to move on to bumping the CPU multiplier a bit higher. ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |
Posted By: PetrolHead
Date Posted: 30 Dec 2017 at 7:19pm
|
Now, onto the baseline memory benchmarks. The CPU is running @ 3.7 GHz and the RAM is configured to run @ 3200 MHz (14-18-18-38 CR1). Results are averages of five benchmark runs. Passmark 9.0 Memory Mark: 2187.2 Database Operations: 91.4 KOps./s Memory Read Cached: 27160.6 MB/s Memory Read Uncached: 18245.6 MB/s Memory Write: 8579.6 MB/s Available RAM: 14135 MB Memory Latency: 59.2 ns Memory Threaded: 43223.8 MB/s AIDA64 Memory Read: 46849.6 MB/s Memory Write: 45354.4 MB/s Memory Copy: 40837.8 MB/s Memory Latency: 77.48 ns L1 Cache Read: 459.156 GB/s L1 Cache Write: 230.184 GB/s L1 Cache Copy: 459.12 GB/s L1 Cache Latency: 1.1 ns L2 Cache Read: 433.8 GB/s L2 Cache Write: 225.972 GB/s L2 Cache Copy: 388.276 GB/s L2 Cache Latency: 4.7 ns L3 Cache Read: 306.972 GB/s L3 Cache Write: 218.552 GB/s L3 Cache Copy: 292.34 GB/s L3 Cache Latency: 12.2 ns Out of these benchmarks, the L3 Cache Read and L3 Cache Copy had the biggest variation with about 26% and 18% differences between the worst and best results, respectively. Otherwise all results seemed to be reasonably stable. ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |
Posted By: PetrolHead
Date Posted: 30 Dec 2017 at 7:44pm
|
As I stated in the first post, the XMP speed of 3466MHz didn't work, so I've set the speed to 3200MHz and lowered the CL value to 14 from 16. Other than that all RAM timings have been XMP values. The next step was to feed my XMP profile info to Ryzen DRAM calculator by 1usmus to get timings for the speed 3200MHz. I'll first use the "Safe" preset and if everything works fine, I might go for the "Fast" preset or 3466MHz and "Safe". We'll see what happens. These are the timings that the calculator gives me: tCL: 14 tRCDWR: 16 tRCDRD: 16 tRP: 16 tRAS: 35 tRC: 51 tRRDS: 6 tRRDL: 9 tFAW: 39 tFAWDLR: 0 tFAWSLR: 0 tWTRS: 4 tWTRL: 12 tWR: 24 tRCPage: 0 tRDRD SCL: 3 tWRWR SCL: 3 tRFC: 560 tRFC 2: 416.1 tRFC 4: 256 tRFC (alt): 416 tRFC 2 (alt): 309.1 tRFC 4 (alt): 190.2 tCWL: 14 tRTP: 12 tRDWR: 7 tWRRD: 3 tWRWR SC: 1 tWRWR SD: 7 tWRWR DD: 7 tRDRD SC: 1 tRDRD SD: 5 tRDRD DD: 5 tCKE: 8 I'll write down the old values when I edit them in BIOS. ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |
Posted By: PetrolHead
Date Posted: 31 Dec 2017 at 12:00am
|
I'm now running on the "Safe" timings that are listed in the previous post. I haven't done extensive stability testing yet, but six MemTest instances each with 2 GB allocated just ran for over an hour without errors. I did also try the "Fast" timings given by the program but these resulted in a MemTest failure within a minute or two. So, I'll keep testing with the timings I have. I doubt I'll be able to go to 3466MHz, since apparently my board is limiting me to 1.35V on the RAM. I haven't found a way to adjust the SoC voltage either. Here are the timings I had before: tCL: 14 tRCDWR: 18 tRCDRD: 18 tRP: 18 tRAS: 38 tRC: 56 tRRDS: 7 tRRDL: 9 tFAW: 42 tFAWDLR: - tFAWSLR: - tWTRS: 4 tWTRL: 12 tWR: 24 tRCPage: 0 tRDRD SCL: 6 tWRWR SCL: 6 tRFC: 607 tRFC 2: 451 tRFC 4: 278 tRFC (alt): - tRFC 2 (alt): - tRFC 4 (alt): - tCWL: 14 tRTP: 12 tRDWR: 6 tWRRD: 3 tWRWR SC: 1 tWRWR SD: 7 tWRWR DD: 7 tRDRD SC: 1 tRDRD SD: 5 tRDRD DD: 5 tCKE: 8 In other words no radical changes and one of the values (tRDWR) actually gets bumped upwards. The fast timings were considerably tighter, so I was half expecting to be unable to boot, but everything actually seemed fine before running MemTest. I even managed a few tentative benchmarks. The gains were not huge, so I'm not expecting very notable gains with these timings, either. I'll be back with the numbers soon... ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |
Posted By: PetrolHead
Date Posted: 31 Dec 2017 at 12:45am
|
Here are the benchmark results with the new timings. Passmark 9.0 Memory Mark: 2216.6 Database Operations: 92.8 KOps./s Memory Read Cached: 27128.4 MB/s Memory Read Uncached: 17578.8 MB/s Memory Write: 8987.2 MB/s Available RAM: 14487.6 MB Memory Latency: 58 ns Memory Threaded: 44914 MB/s AIDA64 Memory Read: 48462.2 MB/s Memory Write: 47562 MB/s Memory Copy: 44226.2 MB/s Memory Latency: 76.48 ns L1 Cache Read: 459.12 GB/s L1 Cache Write: 230.188 GB/s L1 Cache Copy: 459.116 GB/s L1 Cache Latency: 1.1 ns L2 Cache Read: 423.664 GB/s L2 Cache Write: 226.626 GB/s L2 Cache Copy: 384.774 GB/s L2 Cache Latency: 4.7 ns L3 Cache Read: 315.002 GB/s L3 Cache Write: 217.934 GB/s L3 Cache Copy: 248.62 GB/s L3 Cache Latency: 12.2 ns The cache results should not really be affected by the timing changes and differences between these five benchmark sets are mostly indicative of how much the results vary from run to run. Aida64 shows small gains for the memory accross the board, but weirdly the Passmark memory test shows worse results for the read tests. I wonder if that small bump in read-to-write latency affects those numbers at all. I might test setting tRDWR back to six, but otherwise I think I won't tweak the timings more - at least not before we have a working version of AGESA 1.0.0.7. I doubt the gains (and losses) seen here affect other benchmark results in any noteworthy way, but I'll probably run some tests just out of curiosity. ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |
Posted By: PetrolHead
Date Posted: 31 Dec 2017 at 2:37am
|
" rel="nofollow - I noticed that I had had gear down mode enabled the whole time, which means the CR1 I stated earlier wasn't exactly accurate. I've set tRDWR to 6 instead of seven, disabled gear down mode and set command rate manually to 1T and the next step is to do stability testing. The L3 Cache test on Aida64 has me a bit worried. The results are all over the place, with the worst runs scoring less than half of what the best runs score. Even the latency, which most of the time is 12.2 ns, can sometimes hop to over 20 ns. Maybe the RAM timings are somehow indirectly affecting L3 Cache? Or maybe the benchmark is just very sensitive to background processes. ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |
Posted By: PetrolHead
Date Posted: 31 Dec 2017 at 6:17am
|
Hmph. I did some benchmarking and was a little worried because I was seeing lower numbers than in my original benchmark runs, so I reverted back to my old timings and now I'm seeing even lower results. Weird. I think I'm going to stick with these looser timings for now and maybe wait for AGESA 1.0.0.7 before trying again. ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |
Posted By: PetrolHead
Date Posted: 31 Dec 2017 at 11:36pm
|
The CPU settings had at some point switched to auto, which meant the CPU was running at 3.6 GHz. This explains large part of the lower scores in benchmarks and once I corrected the clock speed to 3.7 GHz I started seeing familiar figures. For some reason I still seem to get the best scores in CPU-Z, if I keep the timings at stock XMP timings with CL set to 14 (I haven't tested with CL16, though). As soon as I switch from 14-18-18-18-38-56 to, say, 14-16-16-16-35-51, I seem to lose some points. Also, Aida64's L3 Cache results start to show very low figures, although I'm not quite yet sure if this is just bad luck. In any case, I ran Passmark's MemTest86 from a bootable USB stick at 14-18-18-18-38-56 and no errors were encountered (4 passes, roughly 3.5 hours). At least for now, I'll consider these memory settings stable. At the moment I'm stress testing 3.8GHz @ 1.3V with Aida64. The stress test crashed within a minute @ 1.25V and within two minutes @ 1.275V, but has now been running without issues for about half an hour. The measured Vcore seems to be constantly higher than what I've set in the BIOS. Aida64 is now showing a steady Vcore of 1.318V, and the offset was more or less the same with the other settings. Now, assuming 3.7GHz was stable at 1.25V (I'm not 100% it was, although I didn't encounter any issues), and assuming 3.8GHz is stable at 1.3V, getting 3.9 GHz stable would likely require unhealhy amounts of Vcore. We'll see how stable this current setting is, but it may be that I'm close to the limit already. The positive thing about the current settings is that temperatures do not seem to have skyrocketed. Instead, I'm looking at an increase of about 5C compared to my previous settings, which leaves me hovering at 77-78C, apart from a few very short spikes to slightly over 80C. I'll have to see where Prime95 takes those numbers, but so far the numbers feel acceptable. Slightly off topic, but while doing all this, I've encountered a few issues with my motherboard: -The CPU fan header doesn't seem to know any other settings than "full blast". -The BIOS profiles do not work. If I save a set of settings and try to load it, the BIOS crashes slightly after the settings have been loaded or reboots the computer even before I have the chance to click "ok". (This is on BIOS 3.3, the only BIOS version I've tried.) -Booting from an USB stick is a pain, since for some reason the motherboard is picky about which port I use. I had this issue already when trying to install Windows and had to try a few different ports before the BIOS realized there was an USB option for booting. Same thing happened today with the Passmark MemTest86 USB-stick (USB 3.0 just like the Windows installation media). I don't remember which port accepted the Windows USB-stick, but with MemTest86 I tested one of the front panel USB 3.0 ports and one of the rear panel USB 3.0 ports until a rear panel USB 2.0 port finally found the USB stick (edit: This seems to only occur at boot. When Windows has started, it hasn't so far mattered which port is used). Needless to say, I'm not very impressed with the quality control of ASRock. This stuff is basic and any manufacturer should be able to get these features rock solid, but for some reason it doesn't seem to be happening at ASRock (I don't know about the others). I also had USB port issues with my previous ASRock board, the 970M Pro3, so it feels like ASRock keeps repeating at least some of its mistakes. ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |
Posted By: PetrolHead
Date Posted: 01 Jan 2018 at 5:03pm
|
3.8 GHz was stable @ 1.3V (BIOS, 1.328V according to HWMonitor) for one hour in Aida64. However, it wasn't stable in Prime95 for five minutes at that voltage. Instead, I got a blue screen and the machine rebooted. I increased the voltage to 1.3125V (or 1.344V according to HWMonitor) and I was stable for a few minutes longer before the machine crashed again. The dump file indicated a memory related issue with CUE.exe (Corsair Utility Engine) and I noticed there was an update for it, so I tried if that would help. It didn't, so I dropped the clock speed to 3.7 GHz while keeping the voltage @ 1.3125V (1.344V). Now Prime95 has been stable for two hours. Temperatures have maxed out at 84C and socket temperature has maxed out at 61C. Considering that the real Vcore seems to be constantly higher than what I set in BIOS, I don't think I'll be able to hit 3.8GHz on this system without going over the safe limit of 1.35V. Better cooling could possibly help, but I was not hitting over 80C/60C before the crashes so I should have been "safe". Maybe the CPU just can't handle 3.8GHz, or maybe the 3 phase VRM of my motherboard isn't able to keep the voltage steady enough to achieve stability at those settings (even though the time resolution of HWMonitor is such that it seems to be rock solid). Or maybe I'm on the edge of my CPUs IMCs capabilities with running my RAM at 3200MHz, which is why pushing the cores makes it become unstable. I'm not yet sure what I'll do next. I will either try to see how low in voltage I can go before 3.7GHz becomes unstable, or then I'll see how much over the safe limit I have to go to reach 3.8GHz. With the latter option heat might become an issue with the stock cooler. I'm planning on getting new case fans today, though, so I'll see if those affect temperatures at all. ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |
Posted By: PetrolHead
Date Posted: 01 Jan 2018 at 10:08pm
|
3.7GHz @ 1.3125V (1.344V) was stable for 4 hours in the standard Prime95 blend test. Max temps were 84C CPU and 61C socket. 3.7GHz @ 1.275V (1.296V) was stable for 1 hour in the standard Prime95 blend test. Max temps were 80C and 59C socket. I'll do a longer test later, but so far so good. ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |
Posted By: PetrolHead
Date Posted: 02 Jan 2018 at 4:37am
I'm happy to report I was wrong about this. I just chose my custom fan profile settings so that there didn't seem to be much difference. The fan tends to spin around 2000rpm, but it actually has roughly 700rpm in reserve. An it's really loud when it's actually spinning at full speed. I'll probably play around with the custom fan setting a bit more in the near future. Today I switched the stock Phanteks case fans to Fractal Design Venturi HF-14 units. I'm using the low speed adapter in order to keep them more silent, but frankly they weren't a huge improvement over the Phanteks units. The frequency of the noise is a bit better, the volume of the noise is roughly the same and after another hour of Prime95 the temps had maxed out at 79C / 58C, which is probably within the margin of error. Well, at least nothing seems to have changed for the worse. ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |
PetrolHead wrote:Posted By: PetrolHead
Date Posted: 06 Jan 2018 at 3:53pm
|
3.7GHz @ 1.25V (1.280V) was stable for 8 hours in the standard Prime95 blend test. Max temps were 77C CPU and 57C socket. ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |
Posted By: PetrolHead
Date Posted: 06 Jan 2018 at 4:41pm
|
3.7GHz @ 1.2375V (1.264V) was stable for an hour in the standard Prime95 blend test, but BSOD struck and the computer rebooted. I think temps maxed out at 74C CPU and 55C socket. In any case, the instabilities seem to be memory related, so I'm guessing if I ran the RAM at 2933MHZ or maybe with looser timings, I could push the CPU further. I'm not sure it would be worth it from a performance perspective, though, unless I could push the CPU to 3.9GHz or 4.0GHz. With a custom fan profile I could also try and keep the temperatures lower under stress testing without suffering from increased noise in normal use. That might be next on the list of things to try. ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |
Posted By: PetrolHead
Date Posted: 11 Jan 2018 at 5:48am
|
It seems I was partially right about the CPU fan header after all, but the issue seems to be related to software, not hardware. The cooler spins at different rpms when I use silent, standard or performance mode, but the custom setting doesn't work. The fan spins at roughly the same rpm regardless of whether the CPU is under the 65% limit or over the 90% limit. I guess I'll have to wait and see if the next BIOS update will fix it. ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |
Posted By: PetrolHead
Date Posted: 02 Apr 2018 at 4:15am
|
This weekend I tried the newest version of Ryzen DRAM Calculator by 1usmus (1.1.0 Beta 1). The timings it suggested were a lot tighter than before (using the Safe-preset), but I gave them a shot. Benchmarks did improvements on average, but again some benchmark results (for example Memory Read Unchached in Passmark's Performance Test 9.0 build 1024) gave worse results than with looser timings. Also, even though I had no trouble running benchmarks after benchmarks, Aida64 stress test reported hardware errors within a few minutes of starting the test, so I went back to my old timings after a few attempts to loosen the timings a bit more to find stability. Since fiddling with timings didn't seem to give me any results, I decided to give 3333MHz a try. The computer wouldn't boot to Windows with CL14, but with CL16 (and thus normal XMP timings for 3466MHz) it did and managed to finish a few benchmarks as well. However, two hours into Prime95 blend test the machine booted itself, so I went back to 3200MHz, since I can't increase RAM voltage beyond its current 1.35V setting. Furthermore, I set CL to 16 because my computer had a BSOD for the second time during this year. It could just be a driver issue, but let's see if this helps. I also did a few feeble attempts to OC the CPU a bit, but decided anything less than 100MHz was not worth the effort at this point (and I already know that 3.8GHz requires more voltage than what I'm willing to give the CPU). If I decide to upgrade the CPU cooler, I may give OC'ing the CPU another go. ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |
Posted By: PetrolHead
Date Posted: 18 Apr 2018 at 5:48am
|
I decided to upgrade my CPU cooler, and after some thought I decided to go with an AIO liquid cooler, namely Fractacl Design Celcius S24. Installation was relatively easy, but would have been even easier if the manual hadn't instructed to twist the Intel mount in the wrong direction to get it off. The end result is good regardless. There was no increase in noise levels*, yet max temps** in P95 blend test (after roughly 20 minutes) fell from 74.3 C to 61.8 C, and max temps in AIDA64 stress test (after the temperature plateaued) fell from 70.9 C to 59 C. Socket max temps fell from 54.5 C to 46.0 C and from 52.5 C to 44 C, respectively. There were some temperature spikes - up to 66.8C - during AIDA64 after the installation, but I'm not sure if these are just sensor glitches or if it's something related to running in the cooler. I saw no such spikes during Prime95 and the cooler seems quick to react to increases in load. Overall, the drop in max temperature is maybe not the biggest, considering I was using the stock Wraith Spire cooler. On the other hand, the Wraith Spire is very capable for a stock cooler. Running the CPU fan at higher RPM could have resulted in even lower temperatures, but it would have also increased the noise. Another nice feature of the liquid cooler is that now hot air is blown directly out of the case (from the top). The airflow over the VRMs' heatsinks may not be as good as before, but what air flows through them should now be cooler. All that's left to do now is to see if I can run higher clocks and voltages without the system becoming unstable. *The cooler has two settings: "Auto" and "PWD". Since reviews saw no significant difference between the two, I'm running the cooler on "Auto". **Temperatures from HWiNFO64. ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |
Posted By: PetrolHead
Date Posted: 18 Apr 2018 at 6:23am
|
Hmm. In January I posted that 3.8 GHz @ 1.3125 V was only stable for a few minutes in Prime95. No temperatures were written down for that combination. On the other hand 3.7 GHz was stable at that voltage and temperatures had maxed out at 84 C (CPU) and 61 C (socket) after two hours. Well, now I've been running the 3.8 GHz @ 1.3125 V settings for half an hour and temperatures have maxed out at 66.9 C (CPU) and 48 C (socket). Looks like higher clocks may be achievable after all with reasonable voltages. ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |
Posted By: PetrolHead
Date Posted: 21 Apr 2018 at 12:13am
|
Testing is still ongoing, but looks like 3.9 GHz requires more voltage than I'd like to use. However, 3.8 GHz was stable in Prime95 blend test for 8 hours @ 1.325 V - and with RAM speed increased to 3333 MHz! It seems that the lower temperatures and/or higher Vcore also help the IMC handle higher clockspeeds. I haven't touched the timings yet, but I might give that another try as well. The maximum temperatures during the 8 hour Prime95 stress test were 68.5 C for the CPU and 49.5 C for the socket. Vcore (SVI2 TFN) sat mostly at 1.337 V under load, so a bit more than what was set in BIOS and still under the "safe" value. ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |
Posted By: PetrolHead
Date Posted: 26 Jul 2018 at 6:53am
|
Tightening the timings was a no-go with RAM sitting at 3333 MHz. The system is stable at stock timings but I think (I did the testing a few months ago) just dropping from CL 16 to CL 14 made BIOS crash within a few seconds. It's a shame the board doesn't allow me to bump up RAM voltage above 1.35V, since this severely limits any further testing. Anyways, here are finally some benchmark results for my current 3.8GHz / 3333MHz (16-18-18-38 CR1) settings: Passmark 9.0 Memory Mark: 2218.52 Database Operations: 90.6 KOps./s Memory Read Cached: 27825.8 MB/s Memory Read Uncached: 18488.8 MB/s Memory Write: 8626.6 MB/s Available RAM: 14269.9 MB Memory Latency: 58 ns Memory Threaded: 44931.8 MB/s AIDA64 Memory Read: 48677.2 MB/s Memory Write: 46550 MB/s Memory Copy: 42012.6 MB/s Memory Latency: 76 ns L1 Cache Read: 471.588 GB/s L1 Cache Write: 236.436 GB/s L1 Cache Copy: 471.686 GB/s L1 Cache Latency: 1.1 ns L2 Cache Read: 448.282 GB/s L2 Cache Write: 231.6 GB/s L2 Cache Copy: 398.214 GB/s L2 Cache Latency: 4.5 ns L3 Cache Read: 315.69 GB/s L3 Cache Write: 221.878 GB/s L3 Cache Copy: 286.204 GB/s L3 Cache Latency: 11.9 ns So, nothing major in the memory department. Some benchmarks show worse results than 3200 (14-18-18-38 CR1), some better and overall the performance is in practice the same. As can be expected, other benchmarks showed gains you could expect from a 100Mhz bump on the CPU frequency: CPU-Z CPU Single Thread: 437.18 CPU Multi Thread: 2344.96 Passmark Performance Test 9.0 Total: 5511.66 CPU Mark: 11026.5 2D Graphics Mark: 903.48 3D Graphics Mark: 11056.26 Memory Mark: 2218.52 Disk Mark: 15805.46 Integer Math: 22165.6 Floating Point Math: 8166 Prime Numbers: 41.8 Extended Instructions (SSE): 391.6 Compression: 13678.8 Encryption: 2170.8 Physics: 648.6 Sorting: 8185.6 CPU Single Threaded: 2017.6 There's a notable difference in the Disk Mark results and the disk mark results are far more stable than what they were in the first benchmarks with this hardware. This can be due to Windows updates, driver updates, updates to the benchmark engine or all combined. All these should also be taken into account when comparing any other differences with previous numbers, so it's a bit of an apples to oranges comparison, but the overall picture is what I'd expect. A small bump in clock speeds gives a small bump in performance. However, the difference is only really noticeable in benchmarks. ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |
Posted By: PetrolHead
Date Posted: 08 Aug 2018 at 7:31am
|
This is slightly off topic, but since I'm planning to upgrade my GPU at some point, I decided to find out what sort of power headroom I have with my PSU. So, I went out and bought a plug power meter to measure my system's power consumption at the wall. When running in-place large FFTs with Prime95 (supposed to give the maximum CPU power consumption), the meter showed my system was drawing ~140 W. Running Superposition @ 1080p (windowed and maximum settings) alongside Prime95 raised the figure to ~260 W, which fits quite nicely with the 120W TDP of my GPU. Although this may not be the absolute maximum amount of power my system can draw from the wall, I doubt I'll see anything beyond this in normal use. On the other hand I should be prepared to see power consumption similar to this during gaming, since I saw consumption figures of up to ~240 W during Firestrike (the combined benchmark). So, since my PSU can sustain 550W and since my PSU's +12V rail is rated at 45.8 A (~550 W), in principle I should be able to run any contemporary GPU without issues (if we disregard the fact that I would need to use a split cable instead of two separate cables). Even if I wanted to leave 100 W of "breathing room" for the PSU, I could still go for a ~300 W GPU, which includes anything up to a reference Vega 64. ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |
Posted By: PetrolHead
Date Posted: 22 Dec 2018 at 10:26am
|
The GPU upgrade mentioned in the previous post happened recently and the EVGA GTX 1060 SC 6GB gave way to Sapphire Pulse Radeon RX Vega 56 8GB. So, another GPU with a half-sized PCB, but the cooler in this Vega 56 is twice as large as the PCB. I've already tweaked the settings a bit, landing with 10% increase in power limit, +150MHz overclock on the memory and undervolting so that power states 5-7 are set to 1050 mV. With these settings my system's power draw from the wall has increased to ~450W while running Superposition and Prime95 simultaneously with the same settings as previously. Comparing with my previous numbers, that means roughly 310W of the total is due to the GPU - and by GPU I mean the whole graphics card. Based on GPU-Z readings the GPU itself accounts for 200W and the memory, fanse etc. account for 110W. Other than the Superposition + Prime95 test, I don't think I've seen numbers above 400W during normal usage or even Firestrike's combined test. However, I doubt I'll push the GPU any harder as long as I have my current PSU - at least not for 24/7 use. Not that I really need to push the GPU harder, since the upgrade is in any case significant. Compared to my old GPU, graphics scores are not up by ~65% in 3DMark Timespy and ~75% in 3DMark Firestrike. In Firestrike's combined test the score went up only ny ~25%, which means my CPU is now the bottleneck, but I'm going to replace the CPU after a suitable Zen 2 Ryzen CPU is released. Hopefully ASRock can provide a decent BIOS for this motherboard then.. ------------- Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit |