• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

AMD RDNA4 Information...Pre/Post Launch

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.
Those mag fans seem nice, but it totally depends on how long they will be available as replacement parts.

I have no idea about XFX, but Sapphire probably started with these fans, and they keep replacements for the product's life so 3+1 years or something after the manufacture date. Read it as 3 years of warranty and +1 year as a support after the product goes EOL status.
It depends on the warranty and country whether it's a better option. It's much easier to send a single fan, but most countries don't have manufacturer's support in the same country, so users send products to the store for RMA.
 
You can sell the ebay card
good thinking... you can likely cut your losses/hedge against it and sell it for less. I'd rather eat $200 than $400 (but if people are paying $1K for it... maybe others will continue to do so).
 
I was right with these cards. If you lower the voltage, the card uses as much power as at stock voltage - in both cases goes up to the max power limit, but the lower voltage gives higher FPS. If you lower the voltage and unlock maximum power limit (+10% for most cards), then it goes to the max power limit too. If these cards had much higher power limit, then would boost higher and use much more power.
It explains why overclocking the core doesn't improve performance (or not much). The card simply asks for more power, and the max core frequency (as a boost limit) is already higher than the card can make. You can check the max boost limit in GPU-Z or hwinfo64.

I doubt that any official BIOS or software unlocks the power limit much above "stock". If the card can utilize everything it gets, then at +50% it would have a constant ~500W+ in 3D. Expect some hard mods and LN2 results soon ;)
 
Not sure if I should put this here or in a thread proclaiming, again, how trash Userbenchmark is, LOL!


1741905100910.png

Unsubscribe. Delete from bookmarks... big yikes.
 
Like UserBenchmark was a reliable source of results. I'm surprised that Tom's Hardware even posted an article about it.
... at the same time, the 9070XT is not much different from the RTX5080 in most of the tests I have done so far
 
Like UserBenchmark was a reliable source of results.
lol, some people, albeit lunatics, swear by it. But yeah, you're preaching to the choir here. The point was to, perhaps, put a nail in the dumpster fire that is user bench.
 
"Tearing Down Sapphire's RX 9070 XT Pulse: Thermals, Fan Response, & Noise" ⇾ "We tear down the Sapphire RX 9070 XT Pulse video card and inspect it for build quality, assembly and disassembly process, and design. We'll also benchmark and review the card for its thermals (GPU and memory temperature), frequency and clock behaviour, and RPM response to GPU temperature. We've already benchmarked the RX 9070 XT for gaming performance and power consumption, so you can check out our full review for that (using this same card)."

00:00 - Sapphire RX 9070 XT Pulse Review
01:25 - Card Basics and Construction
04:58 - Tear-Down of the Pulse XT
08:32 - Well Done on Fan Access
10:21 - Exposing the GPU
14:40 - Cooling Solution
18:03 - Recap of Tear Down
18:46 - RPM Problem and Thermal Benchmarks
21:47 - Acoustics in Hemi-Anechoic Chamber
22:39 - GPU Frequency
23:00 - Conclusion

 
I noticed that undervolting/negative offset OC is somehow unpredictable, and it may cause instability in some games/tests. For some reason, my card passes most tests at -100mV, which gives a +/- 3200MHz core. However, in 3DMark Steel Nomad, it crashes, and I see the core clock at about 3300MHz with spikes up to 3440MHz. When I set -50mV, then the card has +/- 3100MHz, but Steel Nomad still crashes, and GPU-Z shows 3300MHz+.

I also noticed that all cards mentioned in early undervolting videos had lower power limits, and the optimal negative offset was -140mV. When I set the same on my card, then it boosts above stable clocks, something like 3300MHz. When I add +10% power limit (max for my card) then it goes even higher.

I don't need overclocking, but it's a bit weird how these cards act.

Btw. GB 9070XT Elite has some coil whine, but it's relatively quiet. What is weird, is the clicking sound each time it switches from high to low power mode (like when it finishes higher load 3D and switches to desktop). The cooler is also quiet.
My card's power limit is 340W +10%, but it sometimes goes as high as 385W and spikes up to 600W after OC.
 
Clicking sound? Like a relay sound or less click?

Like the Predator standing in the corner of my room :LOL:

I have to check if it's not related to the magical RGB, as when I hear it, the illumination mode changes from that halo, blinking something to a rainbow.
 
Maybe the LEDs... If something like that happened in my lab I wouldn't rest until I understood. Mostly because thats the sound of something really bad happening...

If its not the LEDs than it could be backfeed from the power build up on the VRM core side. You gotta dump 400W somewhere so it has to be done quickly through ground, could be a FET trying to dump quickly. But it to make noise would mean something bad is going on. What I'm getting at is that you could be hearing the audible dump of a powered load... not good lol.
 
I had some other problems to solve, so I just left it for now. If I'm right, I can register the card for 4 years of warranty. I'm unsure if it's available in Poland, but in the worst case, it's still 3 years standard, so I have time.

Edit:
The noise is from the fans. It happens only when they stop, and all three of them make it. The fans are really quiet, and I can't hear this noise even at ~500 RPM or something near. It's only when they go into 0 RPM mode. It's also not loud, but I have a generally quiet room, and I often test in the evening/night, so it is quiet at home. It's still weird, but I guess they're designed this way as if it was one, then it could be faulty, but all 3 act the same.
 
Last edited:
Back