Anyone planning to buy the RX 6900 XT, rather than RTX 3080?

[Besty]

Someone has already run time spy with a 6900xt with an average frequency of 2.7GHz.

Nobody knows the Timespy result for an AIB 6900XT 2753mhz 400W+ card until someone posts a valid score with said card on air and then warm liquid.

I won't really trust the scores until well-known OCUK forum members posts them, i don't really trust the press or youtube anymore.

Won't have to wait long, 12 weeks maximum maybe.

As for the 3090's yes we can all see the top scores, they are most probably shunt-modded with a soldering iron and bios cross-flashed 3090 MSI/Asus Strix OC cards under a liquid chiller and not by EVGA at all. There's nothing wrong with that (this is an overclockers forum) but there is a question of data integrity here and what's realistic. In management consulting its called 'making sense of sense'.

Even a stalwart like myself does not really want or have an interest in buying a £1,800 item to then mutilate it with a soldering iron, I have cross-flashed GPU's before when they cost £350 but the thought of bricking a £1,800 soldered card with a cross-flash on it (as no doubt some have already done) just to get another 1,000 points on timespy does not sit well with me but i am sure EVGA are really happy about that situation as the get all the credit for none of the risk from people that post up the scores on forums like these. If the card gets bricked, its not even an EVGA card.

 

[zx128k]

Nobody knows the Timespy result for an AIB 6900XT 2753mhz 400W+ card until someone posts a valid score with said card on air and then warm liquid.

I won't really trust the scores until well-known OCUK forum members posts them, i don't really trust the press or youtube anymore.

Won't have to wait long, 12 weeks maximum maybe.

As for the 3090's yes we can all see the top scores, they are most probably shunt-modded with a soldering iron and bios cross-flashed 3090 MSI/Asus Strix OC cards under a liquid chiller and not by EVGA at all. There's nothing wrong with that (this is an overclockers forum) but there is a question of data integrity here and what's realistic. In management consulting its called 'making sense of sense'.

Even a stalwart like myself does not really want or have an interest in buying a £1,800 item to then mutilate it with a soldering iron, I have cross-flashed GPU's before when they cost £350 but the thought of bricking a £1,800 soldered card with a cross-flash on it (as no doubt some have already done) just to get another 1,000 points on timespy does not sit well with me but i am sure EVGA are really happy about that situation as the get all the credit for none of the risk from people that post up the scores on forums like these. If the card gets bricked, its not even an EVGA card.

Again 2.7GHz average is 2.7GHz with any power level. You have to have enough power to reach a 2.7Ghz average. Adding more power does nothing for performance, unless the frequency increases further.

If you flash a card you need to backup to bios. If the new bios does not work, you can buy a kit to reprogram the bios chip. Most of the time the kit will clip to the bios chip and you dont have to remove it from the board. Power limits exist to stop you from killing the gpu. Most of the time you need extreme cooling to keep the gpu chip operating at higher power limits. If you run too much power for too long,without the right cooling the card will die on you.

 

[xcx xcxvgyt]

Again 2.7GHz average is 2.7GHz with any power level. You have to have enough power to reach a 2.7Ghz average. Adding more power does nothing for performance, unless the frequency increases further.

If you flash a card you need to backup to bios. If the new bios does not work, you can buy a kit to reprogram the bios chip. Most of the time the kit will clip to the bios chip and you dont have to remove it from the board. Power limits exist to stop you from killing the gpu. Most of the time you need extreme cooling to keep the gpu chip operating at higher power limits. If you run too much power for too long,without the right cooling the card will die on you.

It seems that you underestimate AMD's wattman. It can boost from 2250mhz to 3000mhz without provide an extra 1fps. And it is not a bug it is a feature.

 

[nvidiamd]

It seems that you underestimate AMD's wattman. It can boost from 2250mhz to 3000mhz without provide an extra 1fps. And it is not a bug it is a feature.

You know that the graphics score in timespy is calculated by the FPS and not by the frequency. So it seems like increasing the frequency gives you more FPS.
BTW right now the 6800xt is more than 235 points ahead of 3080 on average.

 

[xcx xcxvgyt]

You know that the graphics score in timespy is calculated by the FPS and not by the frequency. So it seems like increasing the frequency gives you more FPS.
BTW right now the 6800xt is more than 235 points ahead of 3080 on average.

Not necessary, For ex. check this: https://www.youtube.com/watch?v=fkPpZ9pWT4w

2750mhz to 2700mhz with exact same settings and gain %10 fps?

As i told you it's a feature. If you like more frequency Amd will provide you. If you like performance also AMD able to provide it too. What a nice feature

 

[zx128k]

It seems that you underestimate AMD's wattman. It can boost from 2250mhz to 3000mhz without provide an extra 1fps. And it is not a bug it is a feature.

You cant boost to 3GHz on a 6800xt or a 6900xt. With the 6800xt its bios limited to 2.8GHz. To get 3GHz on a 6900xt you need LN2. Both cards like 2.7GHz maximum at the moment in time spy benchmarks and much lower in port royal. This means that the max stable OC frequency is about 2.5GHz air and 2.6GHz water because thats the maximum for RT games. This is why the Time spy scores are so high but the cards are not stable in all games at that frequency. Thus the preformance of the 3080 and 3090 once overclocked is far higher. This is why the average for time spy for the 6800xt and 6900xt is misleading.

RT reduces max frequency on all of the cards. 6800xt/6900xt dont go much above 2.5GHz. This is port royal, control reduces frequency even more on all RT cards.

 

[xcx xcxvgyt]

You cant boost to 3GHz on a 6800xt or a 6900xt. With the 6800xt its bios limited to 2.8GHz. To get 3GHz on a 6900xt you need LN2. Both cards like 2.7GHz maximum at the moment in time spy benchmarks and much lower in port royal. This means that the max stable OC frequency is about 2.5GHz air and 2.6GHz water because thats the maximum for RT games. This is why the Time spy scores are so high but the cards are not stable in all games at that frequency. Thus the preformance of the 3080 and 3090 once overclocked is far higher. This is why the average for time spy for the 6800xt and 6900xt is misleading.

RT reduces max frequency on all of the cards. 6800xt/6900xt dont go much above 2.5GHz. This is port royal, control reduces frequency even more on all RT cards.

I disagree. For ex check this: https://www.3dmark.com/pr/667349

This on air rock stable and "Average clock frequency 2.738 MHz".

This is the practical limit of AMDs. When card goes above 2750mhz bios always kicks in and lower the frequency.

It's all about the max frequency at Radeon software. If you set 2700mhz max. It will boost 2640mhz. If you set 2500mhz it will boost 2440mhz and so on.

6800xt limit is 2750 and 6900xt 2950mhz for now. I do not know why there are no 6900xt's boost more than 2750mhz. But it should be there is no limit for it.

https://hizliresim.com/nc695N

As you can see above it's a costant line at 2740mhz.

 

[nvidiamd]

6800xt limit is 2750 and 6900xt 2950mhz for now. I do not know why there are no 6900xt's boost more than 2750mhz. But it should be there is no limit for it.

Because the reference card is power starved. Let's wait until we see the ones with 3x8 pin connectors.

 

[xcx xcxvgyt]

Because the reference card is power starved. Let's wait until we see the ones with 3x8 pin connectors.

I disagree again. If it's power limited why If you set max 2600mhz at radeon software it boost to 2550mhz?

Please check picture above. It is not a power limit there are artificial limit for card. Otherwise no one can create a constant line like this.

 

[nvidiamd]

But the best 6900xt boost to an average of 2710 on Timespy. We don't see the same thing on the 6900xt the average clock is all over the place. So this could be a power/temperature limitation for every card. We will have to wait until we can see what a Red Devil or Asus with 3x8 pins can do. Let's hope they come with higher TDP because otherwise why would they use 3x8 pins?

 

[xcx xcxvgyt]

But the best 6900xt boost to an average of 2710 on Timespy. We don't see the same thing on the 6900xt the average clock is all over the place. So this could be a power/temperature limitation for every card. We will have to wait until we can see what a Red Devil or Asus with 3x8 pins can do. Let's hope they come with higher TDP because otherwise why would they use 3x8 pins?

Please check this video to understand what tried to say. (Due to hdr colors shifted) and please consider this is only 1440p bencmark. At 4k or port royal ect. boost algorithm never let you go above 2750mhz. (Hd version will be avaible around 5 min)

https://youtu.be/Ozu0SAXV9mM

And i don't know why 6900xt never boost more than 2800mhz. I believe if there is no limit for 6800xt can boost more than 2800mhz.

Note: MTP used due to eliminate power limit.

 

[zx128k]

I disagree. For ex check this: https://www.3dmark.com/pr/667349

This on air rock stable and "Average clock frequency 2.738 MHz".

This is the practical limit of AMDs. When card goes above 2750mhz bios always kicks in and lower the frequency.

It's all about the max frequency at Radeon software. If you set 2700mhz max. It will boost 2640mhz. If you set 2500mhz it will boost 2440mhz and so on.

6800xt limit is 2750 and 6900xt 2950mhz for now. I do not know why there are no 6900xt's boost more than 2750mhz. But it should be there is no limit for it.

https://hizliresim.com/nc695N

As you can see above it's a costant line at 2740mhz.

Yes for time spy, once you run an RT game you wont get 2.7GHz. The max frequency is reduced.

Max in Port Royal, top result for the 6900xt.
Average clock frequency 2,639 MHz
Average temperature 34 °C which implies water.
https://www.3dmark.com/pr/638640

This is 2.6GHz air https://www.3dmark.com/pr/671189

This is a rare maximum overclock. The 6900xt wont reach higher atm.

6900xt cant do 2.7GHz stable. Its only stable enough to pass Time Spy at this frequency. The argument that these cards will get faster and faster. Is non sense. The higher average in Time Spy is not the 6800xt or 6900xt's real performance. They both can't match an overclocked RTX 3080 in DX12. Only the highest overclocked 6900xt's can match a rtx 3080 FE in port royal. Both AMD 6000 series cards are slower. The real stable max OC could be between 2.4GHz-2.6GHz. Getting more rare as you move towards 2.6GHz.

 

[xcx xcxvgyt]

Yes for time spy, once you run an RT game you wont get 2.7GHz. The max frequency is reduced.

Max in Port Royal, top result for the 6900xt.
Average clock frequency 2,639 MHz
Average temperature 34 °C which implies water.
https://www.3dmark.com/pr/638640

This is 2.6GHz air https://www.3dmark.com/pr/671189

This is a rare maximum overclock. The 6900xt wont reach higher atm.

6900xt cant do 2.7GHz stable. Its only stable enough to pass Time Spy at this frequency. The argument that these cards will get faster and faster. Is non sense. The higher average in Time Spy is not the 6800xt or 6900xt's real performance. They both can't match an overclocked RTX 3080 in DX12. Only the highest overclocked 6900xt's can match a rtx 3080 FE in port royal. Both AMD 6000 series cards are slower. The real stable max OC could be between 2.4GHz-2.6GHz. Getting more rare as you move towards 2.6GHz.

For Rt games or apps AMD cards slower thats accepted. Also 6900xt seems to have problem with higher frequencies. Maybe it needs more power don't know.

But for me 2750mhz is stable with my 6800xt.

Here is 4k ultra Valhalla: https://www.youtube.com/watch?v=36y2x39C35E As you can see only 72c and fans are auto my daily settings. Never got crash before.

It's rare to boost that much with other games. For ex; cyberpunk averages about 2730mhz. Using MTP for more power also increases frequencies but only up to 2750mhz not more.

I believe if there is no 2800mhz barrier at time spy 6800xt can beat ln2'd 3080. (already beat every other 3080's)

 

[Grim5]

You cant boost to 3GHz on a 6800xt or a 6900xt. With the 6800xt its bios limited to 2.8GHz. To get 3GHz on a 6900xt you need LN2. Both cards like 2.7GHz maximum at the moment in time spy benchmarks and much lower in port royal. This means that the max stable OC frequency is about 2.5GHz air and 2.6GHz water because thats the maximum for RT games. This is why the Time spy scores are so high but the cards are not stable in all games at that frequency. Thus the preformance of the 3080 and 3090 once overclocked is far higher. This is why the average for time spy for the 6800xt and 6900xt is misleading.

RT reduces max frequency on all of the cards. 6800xt/6900xt dont go much above 2.5GHz. This is port royal, control reduces frequency even more on all RT cards.

I disagree again. If it's power limited why If you set max 2600mhz at radeon software it boost to 2550mhz?

Please check picture above. It is not a power limit there are artificial limit for card. Otherwise no one can create a constant line like this.

A clock speed over time graph is a good way to see if your card is power limited - when it's limited the clock speed line will be very unstable, when you are not power limited the clock speed line appears very stable and flat

So yea if you are seeing 6900xt with flat clock speed at 2.7ghz then it's not power limited but clock speed limited in its bios

 

[xcx xcxvgyt]

A clock speed over time graph is a good way to see if your card is power limited - when it's limited the clock speed line will be very unstable, when you are not power limited the clock speed line appears very stable and flat

So yea if you are seeing 6900xt with flat clock speed at 2.7ghz then it's not power limited but clock speed limited in its bios

That is correct but as you can see at second video basic formula is (max freq.set at radeon software - ~50mhz). After 2750mhz it never boost above even with more power. I tried very silly powers with MCP but nothing changed. Also decreasing power also didn't change frequency after a certain level.

Just compare what i set at radeon software and how much it boost as a result: I don't think at 2500mhz it was power limited. https://www.youtube.com/watch?v=Ozu0SAXV9mM

It is obious for me card wants to boost higher but bios or driver pulls it back to formula.

 

[xcx xcxvgyt]

In addition 3dmark avg clocks is not %100 correct. Again for ex; https://www.3dmark.com/spy/16240104 Average clock frequency 2.716 MHz

But reality: https://hizliresim.com/wFXy2j Except the mid of GT2 almost have a flatline again. Don't understand why it get lower anyway. Average must be around 2730-40mhz again.

Now imagine without limitation of the score.

 

[zx128k]

A clock speed over time graph is a good way to see if your card is power limited - when it's limited the clock speed line will be very unstable, when you are not power limited the clock speed line appears very stable and flat

So yea if you are seeing 6900xt with flat clock speed at 2.7ghz then it's not power limited but clock speed limited in its bios

An average of 2.7GHz means higher or lower than 2.7GHz. These cards are boosting based on the load. We already know that ray tracing loads hit cards harder. The average drops to 2.6GHz and below for maximum clocks. This means an average clock is to be expected and will be based on the load, temperature and not just the power limit. This is why rage mode is not much of an increase in performance. A load like Time Spy could hit 2.7GHz max but only get 2.5GHz max in Ray Tracing. This makes the card seem faster in Time Spy but overall in RT the card clocks drop. The cards are not that fast, if you want a flat clock speed, then 2.4-2.6GHz (2.6GHz is approx the value in the top 100 cards for port royal, post royal is not as hard on the gpu as control can be which will lower clocks even more) is what is stable maximum for all loads. 6900xt is clock limited in bios at 3GHz. THe 6800xt is clock limited in bios at 2.8GHz. https://www.techpowerup.com/review/amd-radeon-rx-6900-xt/40.html

The techpowerup card OC'ed to 2.8GHz boost but ran at an average of 2.4GHz, after setting the power limit to maximum. A 6900xt @ 2.7GHz average is covered here https://www.igorslab.de/en/radeon-r...e-spy-thanks-water-cooling-and-morepowertool/ and here https://www.igorslab.de/en/red-bios...en-more-stable-overclocking-navi-unlimited/3/

So people are getting 2.7GHz by increasing the power limit in MorePowerTool and water cooling. This clock happens only in lighter loads and lower resolutions than 4k. This is why firestrike gets higher performance than time spy and RT.

Somehow, despite a permanent lack of time, my fingers itched very badly during the current water cooling test. And because I just screwed the card onto a water block anyway, it just had to be done. With the MorePowerTool I raised the TGP from 255 watts to 320 watts, increased the power limit in the Wattman by another 15% (320*1.15=368 watts), set the GPU clock to 2730 MHz and a run the benchmart. I was then quite satisfied with the first result. I also had a score over 19k, but only with the press drivers, which could not be validated. With the release driver and the same settings, it was suddenly just over 100 points less. Why ever.

In detail, 20,710 points in the Graphics Score are ok, but you could certainly have continued testing for hours and maybe even shimmyed over the 2.7 GHz limit. In simpler games without a heavy load, such clock rates are certainly possible in Full-HD and WQHD, but in Ultra-HD it gets tight.

So a 2.7GHz overclock is an extreme overclock for the 6900xt, with a massive increase in the power limit and water cooling. Even so 2.7GHz is only possible in lighter GPU loads.

In benchmarks a stock 6900xt from AMD barely out performs a stock RTX 3080 FE in raster and is beaten in RT games. https://www.techpowerup.com/review/amd-radeon-rx-6900-xt/35.html and https://www.techpowerup.com/review/amd-radeon-rx-6900-xt/38.html

 

[xcx xcxvgyt]

An average of 2.7GHz means higher or lower than 2.7GHz.

Igors 6900xt seems very bad binned chip. Means even a bad chip can do 2.7mhz with watercooling&power.

Also 3dmark averages is not really correct for example i did this 5min ego: https://www.3dmark.com/fs/24420118

Says avg: 2653mhz But in reality here is the 3dmark file you can download and analyze over 3dmark software.
https://dosya.co/zt4hkv94yj3s/result.3dmark-result.html


It always a flat line with little fluctuations. On air with a stock card only using MTP to provide enought power. Just imagine what can this card do with ln2 when unlocked. AMDs barrier over there, pulls frequency back. That why max 6800xt scores not good.

Note: This is not 6900xt but you know they are almost same chip. Same rules applied.

 

[nvidiamd]

Also 3dmark averages is not really correct for example i did this 5min ego: https://www.3dmark.com/fs/24420118

Damn! Don't do that because you will ruin 6800xt average score and then zx128k will talk again about how great the 3080 is and how bad the Bell curve looks on the 6800xt.

 

[xcx xcxvgyt]

Damn! Don't do that because you will ruin 6800xt average score and then zx128k will talk again about how great the 3080 is and how bad the Bell curve looks on the 6800xt.

I mean frequency averages on the score page not score averages