Radeon Pro AI R9700 fan noise? Which model to buy?

A Level1Techs forum user reported that running two AMD Radeon Pro AI R9700 GPUs at 210W with fans at roughly 33% and reduced clock speeds makes the cards silent at the cost of some prompt-processing performance. The user's hashcat load tests at 300W showed the ASRock R9700 peaked at 94°C junction temperature and 3725 RPM after 270 seconds, while the ASUS R9700 reached 103°C and 3967 RPM, with the ASRock card moving significantly more air at lower RPM. The user also reported an ASRock R9700 fan that vibrates the entire desk due to an apparent balance defect, and criticized reviewers for not comparing the cards' cooling designs.

I run both my R9700 at 210W and the fan at around 33%, I also had to lower the clock speed. You will lose some performance in prompt processing, but the GPUs will be silent. 2 Likes Hopefully this helps some folks. I have both the Asus and the Asrock R9700’s. Here is a full load output comparing the two cards. Load was done via hashcat. Asrock card moves significantly more air than the Asus card while keeping a lower RPM. I have not done DB comparisons of these two cards yet but would expect the Asus card to actually be louder. I did have the gigabyte card briefly and felt that card was louder than the Asus card running hashcat. By the looks of the Gigabyte card you would think the back side is used to pull air into the card but its actually closed off like the casting was done before they realized how the fan was going to fit into it. One other thing to note is the ASUS card does not actually recess the fan the same way the Asrock card does and may choke for air if it doesn’t have a slot gap between the cards. Hashcat puts a heavy load on the cards and they don’t get hit nearly as hard for LLM usage. Depending on the model usually one card is at 95% and the other is hovering around 50%. This is 300w on both cards. ASROCK R9700 Time Junc PWM RPM 15s 88°C 122 2354 30s 91°C 170 3321 60s 93°C 181 3533 120s 94°C 188 3688 180s 94°C 191 3725 270s 94°C 191 3725 ASUS R9700 Time Junc PWM RPM 15s 84°C 107 2769 30s 90°C 127 3177 60s 97°C 145 3508 120s 102°C 165 3838 180s 102°C 173 3962 270s 103°C 175 3967 3 Likes I have two of the Asrock ones. Don’t discount the fact that the entire card is solid metal. The fan is basically the only plastic thing on it. I don’t know if that’s standard or not, but I do like it very much. 1 Like rbrtb https://forum.level1techs.com/u/rbrtb 24 Thanks for the data Have you ever tried running the cards at lower power as others have suggested? If so, did you feel that made a significant difference? Started with an XFX, then got a Gigabyte v1.1 to pair with it. The GB affects a soft whoosh under load, while the XFX catches my ear under same. I was going to get an Asrock. Holding the GB in my left and the ASR in my right in the store suggested to me that the GB’s heft promised more cooling power: the ASR was like 1/4 the weight shrugs Because of my slots H13SAE-MF , they are about 4mm apart. I was a little concerned about that, but they hover in the low 30’s at-rest, and can slip just past 70C under load. Case has excellent airflow, but I keep a 120mm fan blowing directly at them for emotional insurance Really bothers me that not one reviewer actually compares the cards in reviews. Afraid to upset the vendors? There are some significant differences between cards that need to be called out. Asus card top image has open back intake but the actual squirrel fan intake is on the inside of the blades and Asus only exposes the inside to the front part of the fan. The ASRock on the other hand has the entire inside of the fan exposed to pull air from the back. When I covered the back of the Asus card with tape and ran tests the temps were exactly the same. I’m not sure it even pulls a vacuum from the back with the way it’s designed. When I first compared the cards, it was such a large difference in airflow that I thought the card was broken. 2 Likes So my ASRock R9700 went into my 9950X3D2 gaming rig for a few weeks yeah, it can game, lol . The fan noise wasn’t the problem it is louder than the whole system . I seem to have a defect in the fan being out of balance as it vibrates the entire desk. I haven’t decided to contact ASRock yet or try to see if there’s something stuck in it. Given, the fan doesn’t spin anywhere this high under LLM workloads. Just gaming keeps it pegged at 100% for a hour or more. So it heatsoaks. Temps stay around 65 C with very high outside airflow dedicated and directly to the GPU better than most rigs , with ambient around 76 F. A note about power draw at gaming… At 100% path tracing GPU load, the entire system pulled around 540W from the wall… with a CPU that idles around 80W and was used at about 12%. On CPU intense missions, the entire system pulled about 580W from the wall. As a comparison, a 4070 Super rated at 220W and at 99% load was showing 340W from the wall at the highest CPU intense missions. So same game, same Path ray tracing to peg the GPU, same missions… the 4070 Super pulls 240 Watts less than the R9700 ? It’s noted that the 4070 Super in Cyberpunk one of my tests only pulls 190W, which reviewers say is very efficient. Could you provide the Linux details? I tried I think three different vbios versions. Stock, Asus site Default,Update . The one labelled Default was probably the worst one as it idled the fan at 2k RPM. Was hoping they would have had a more aggressive fan speed as hotspot approaches 100 degrees with the updates. Still runs over 100 with the card at 300w in hashcat. I have a sapphire r9700 arriving tomorrow. Let me know what benchmarks you want. It will be paired with a asrock r9700. My concern is that I set the asrock up 6-8 weeks ago, and today I was preparing my plans for when the card arrives wanna squeeze every minute from the weekend, so I like to be ready . The environment already seems drastically different since I set up my first. Thats a good thing, but I feel simply using it, not configuring it for 2 months has led me to become “stale”, “moldy”, and “old”. Is there a modern spec sheet on vllm vs llama.cpp, Vulcan? Amd 3700x on an x570 board only 32gb ram yes, I know, I need more . Bare metal ubuntu 24.04 due to ram. If i sell some livers and hearts, ill get enough ram to run proxmox and virtualize the thing. For noise, I keep it in a 4u server, on the floor next to rack, until im done tinkering seems like a trademark of mine . Sfp+ 10gb connection to my storage array zfs rz.2 18tbx8 with 2 special vdevs 4x1gb, and a second pool of mirrored/stripe 2tb sata drives We all have become stale, moldy, and old for not taking care of our stuff. I think only the MTP stuff is newer than our original setups. Oh, and be sure to grab the Qwen chat template fix if you do agents override the GGUF one . 1 Like This is how I change the GPU settings. 2 Likes Hondo https://forum.level1techs.com/u/Hondo 33 I’m teetering on the edge of buying one of these cards with the expectation I’ll get a second later on. I’m buying one at a time because noise is a definite concern with me and I want to hear one in person first. It sounds like ASRock is the winner in terms of noise then? Someone did mention a Sapphire one but they hadn’t got it yet. My use is LLMs and I’m buying it for the extra memory more than tok/s so I don’t have a problem undervolting it a little. My current card is a Sapphire 7900XT why or why didn’t I spring for the XTX and that precious extra 4GB - who knew back then? which I can hear but isn’t bothersome. I presume even with undervolting it’s optimistic to think I can get it to the same noise levels as that? I have a reference edition 7900XT, and can’t hear the fans at all. It helps that I have two 120mm fans blowing directly into the GPU fans bottom case fans , so the GPU under 100% stress test loads never exceeds 62C. With that said, I did test an ASRock R9700 in the same chassis with the 120mm fans directly into it. The R9700 didn’t care and was noticeably very audible at 80% fan speed. I was not undervolting then, but I will say that the ASRock R9700 is audible. Not bad actually, but it’s not silent. For idle, I couldn’t really hear it at all. For playing games and LLM uaage, I had to use headphones. My office is pindrop silent, so I hear every raindrop on the roof. It’s no where near as bad as many other blower GPUs I’ve had in years past. But for someone running a silent office, it’s noticeable. i can’t say as the R9700s are now in the garage. But I can still hear them a little on long multi-agent usages. So I’d say you still can hear. 1 Like Hondo https://forum.level1techs.com/u/Hondo 35 Thanks. I don’t mind if they ramp up a little sometimes but I’d hate continuous noise. I’ve pretty much made up my mind to get an R9700 and the ASRock if available. I guess I’ll find out. 1 Like peteru https://forum.level1techs.com/u/peteru 36 Any experience with the noise levels on Powercolor R9700 cards? Linux only here, so I’m not sure if fan curve and power limiting is even an option. I’m still unclear on which card is the least annoying, while still having adequate cooling. Silence is golden, but pointless if the card gets to 100+℃ under load. Here in Australia, the Gigabyte R9700 card is retailing for $200 more than AsRock, XFX and Powercolor. Asus is not available at all. What kind of magic sauce do Gigabyte have to get away with that kind of pricing premium? peteru https://forum.level1techs.com/u/peteru 37 Looking at the difference in spec, the Gigabyte card has one HDMI 2.1b port, whereas all the other cards are DisplayPort 2.1a only. Are they really charging a $200 premium to replace one DisplayPort 2.1a with a HDMI 2.1b port? Seems kind of pointless for an “AI” card that is most likely to run headless anyway. Poleron https://forum.level1techs.com/u/Poleron 38 How is Coil Whine on the ASrock?