Overclock.net banner
41 - 60 of 68 Posts
Discussion starter · #41 ·
Ah i see what you mean. i think Nvidia does not report target clock but actual clk. That is just the PLL trying to lock on. It will never hit the target clock.
I found a bug with my Vega 64 when flashing bios, it would lock the highest p-state after the first reboot. I was able to benchmark at 1827Mhz, while 1800Mhz was 100% stable. It just frustrates me their cards always leave a bit of performance on the table. As soon as I restart I was back at only 1700Mhz.
 
Its similar to Nvidia. You are power limited first, then voltage then temperature.
...yeah, I picked up the cheapest (albeit 3-x Pin) 6900 XT I could find, and right out of the box, it benched several 4K at between 2700 MHz and 2800 Mhz on air...that said, there are some benches that like higher clocks, while others seem to prefer a much larger power target (and undervolting). Once I customized things with the MorePowerTool (MPT), I realized how much more the card still has in the tank, but then temps stopped me, for now - until I get around to mount that new Byksi block waiting patiently on my desk. The 3090 card (part of the same productivity+entertainment build) behaved very much like it when comparing 'stock air' to full 'water block.'..the lower the temps, the better, what with boost algorithms that have temps as one of the inputs
 
Discussion starter · #43 ·
...yeah, I picked up the cheapest (albeit 3-x Pin) 6900 XT I could find, and right out of the box, it benched several 4K at between 2700 MHz and 2800 Mhz on air...that said, there are some benches that like higher clocks, while others seem to prefer a much larger power target (and undervolting). Once I customized things with the MorePowerTool (MPT), I realized how much more the card still has in the tank, but then temps stopped me, for now - until I get around to mount that new Byksi block waiting patiently on my desk. The 3090 card (part of the same productivity+entertainment build) behaved very much like it when comparing 'stock air' to full 'water block.'..the lower the temps, the better, what with boost algorithms that have temps as one of the inputs
I do prefer AMD's handling of thermal throttling better than Nvidia's which starts at 40c. While AMD cards won't begin to throttle until 105c(might have changed with RDNA) on the junction temperature. When both temperature/power limitations are out of the question I don't like AMD's boost algorithm. Their "max clock" p-state leaves performance on the table.
 
I do prefer AMD's handling of thermal throttling better than Nvidia's which starts at 40c. While AMD cards won't begin to throttle until 105c(might have changed with RDNA) on the junction temperature. When both temperature/power limitations are out of the question I don't like AMD's boost algorithm. Their "max clock" p-state leaves performance on the table.
They are not leaving any performance in the table. You just have to understand how their frequency generation works. Too much info to talk about in a forum post.
 
It's definitely leaving performance on the table. My Vega 64 benchmarks prove it's leaving performance on the table.
Vega 64 is difference from Radeon 7 and again different from 5700XT/6800XT
 
Discussion starter · #47 ·
Vega 64 is difference from Radeon 7 and again different from 5700XT/6800XT
The Vega 64 and Radeon VII behaved the same. RDNA1 algorithm looked different but it was higher than the 50Mhz difference on Vega 64/Radeon VII.

At the end of the day a higher frequency will provide better performance. Unfortunately AMD doesn't let their users set their exact frequency any longer.
 
The different clocks in navi 2 on a fixed max p state cap just reflects the different workloads and their power consumption

navi2 just boosts at the max that speciific workload allow them to stay in the usual power limit/temp/voltage matrix

fixed frequency gpus times are long gone (ironically it started with nvidia)
 
The different clocks in navi 2 on a fixed max p state cap just reflects the different workloads and their power consumption

navi2 just boosts at the max that speciific workload allow them to stay in the usual power limit/temp/voltage matrix

fixed frequency gpus times are long gone (ironically it started with nvidia)

He does not mean variable frequency. He means if you set modern AMD GPU to 2700 it will hit 2600-2650 even with all the limits removed like power/temperature. In the end it does not really matter. If you want 2700MHz just set to 2750MHz. All cards have limits.
 
I read a blurb in Tom's Hardware the other day that the only place you can get the thing is in India, and that AMD is going to hand pick who ever they want to have it. I believe it's got an outlandish price also. Personally, I'll stick with Nvidia.
Which card doesn't have an insanely outlandish price?
 
Only cards that are not actually for sale, everything that is for sale has an insane price. At least as far as I can tell.
Exactly, 3090's for 3000+ (sometimes 3500), 6700's for 1000+. Total madness.
 
I do think it is worth noting that the market in the US is still somewhat different than for example Europe or Asia - and even Canada...while prices are generally 'high', they are still lower and availability better elsewhere, due in part because of left-overs from the tariff wars etc re. the US. I base this observation on weekly checks of bigger European, US and other on all overline retailers. For the record, I got a 3090 Strix OC is early February at a good MSRP for one system, and followed that up last month with a 6900 XT for a another work system, again at MSRP, here in Canada. I wouldn't call either purchase 'cheap', but nowhere near the scalper prices I see quoted.
 
Here in China prices are beyond madness. I can't believe that you were able to buy these cards at MSRP, I'm damned jealous.
 
41 - 60 of 68 Posts