RTX 2080 memory bandwidth issue

[dj-electric]

Hi w1z, seems like this RTX 2080 turned into an 8800GT

 

[bug]

Maybe it deactivates bandwidth when it doesn't need it?
You know, like engines shut down cylinders?

 

[mobiuus]

yeah i noticed it too, very curious...maybe a gpuz needs an update maybe it is like bug said...

 

[bug]

yeah i noticed it too, very curious...maybe a gpuz needs an update maybe it is like bug said...

No way it's like I said, I was just poking fun.

 

[PerfectWave]

maybe u need multiply that for 6 cos it use gdr6 XD

 

[Ruru]

It just shows the "true", it doesn't recognize it as DDR. So multiply that with 4.

 

[Assimilator]

The formula for calculating memory bandwidth is (bus bandwidth / 8) * (memory clock speed) * (memory type bits per cycle). So for stock GTX 2080 that's (256 / 8) * 1750 * 8 = 448 000 MB/s = 448 GB/s. My guess is that @W1zzard 's calc for memory doesn't know about GDDR6 (or isn't detecting it correctly) and is thus defaulting to standard DDR, which has only 2 bits/cycle versus GDDR6 which has 8 bits/cycle.

All the GPU-Z screenshots in the GTX 2080/2080 Ti reviews have the same issue... Wiz is gonna have to redo all of those once he fixes this bug!

 

[Ruru]

The formula for calculating memory bandwidth is (bus bandwidth / 8) * (memory clock speed) * (memory type bits per cycle). So for stock GTX 2080 that's (256 / 8) * 1750 * 8 = 448 000 MB/s = 448 GB/s. My guess is that @W1zzard 's calc for memory doesn't know about GDDR6 (or isn't detecting it correctly) and is thus defaulting to standard DDR, which has only 2 bits/cycle versus GDDR6 which has 8 bits/cycle.

All the GPU-Z screenshots in the GTX 2080/2080 Ti reviews have the same issue... Wiz is gonna have to redo all of those once he fixes this bug!

So in a nutshell, this means that it's quadruple like GDDR5?

 

[dj-electric]

Memory bandwidth throughput is memory bandwidth throughput. It does not need calculations, multiplier or anything else. Never did.
I just put it here so w1zz could fix it. That's all. Basically what @Assimilator said

 

[Assimilator]

So in a nutshell, this means that it's quadruple like GDDR5?

Nope, octuple. GDDR5 is 4 bits/cycle which is why GTX 1070 needs to run its 256bit GDDR5 at 2GHz to reach "only" 256GB/s bandwidth, while RTX 2080's 256bit GDDR6 purrs along at a much lower 1.75GHz yet is able to reach almost double the bandwidth of 1070 @ 448GB/s.

On the other hand, GDDR5X (which GDDR6 is essentially the successor of) is also 8 bits/cycle, hence why GTX 1080 can get 320GB/s bandwidth (far higher than GTX 1070) at only 1250MHz memory clock (far lower).

 

[kastriot]

O man i feel sorry for you, wasted money eh?

 

[Ruru]

Nope, octuple. GDDR5 is 4 bits/cycle which is why GTX 1070 needs to run its 256bit GDDR5 at 2GHz to reach "only" 256GB/s bandwidth, while RTX 2080's 256bit GDDR6 purrs along at a much lower 1.75GHz yet is able to reach almost double the bandwidth of 1070 @ 448GB/s.

On the other hand, GDDR5X (which GDDR6 is essentially the successor of) is also 8 bits/cycle, hence why GTX 1080 can get 320GB/s bandwidth (far higher than GTX 1070) at only 1250MHz memory clock (far lower).

Ah, ok. It's always nice to be a little wiser.

 

[W1zzard]

Just a calculation bug, will be fixed in next release

 

[bug]

Just a calculation bug, will be fixed in next release

*next build multiplies all bandwidths 4 times*

 

[Vayra86]

*next build multiplies all bandwidths 4 times*

Only under load!

 

[dj-electric]

O man i feel sorry for you, wasted money eh?

Not a dime (media), but thanks for the care.

 

[EarthDog]

Will the Texture Fillrate GT/s issue on the 2080's also be fixed with the memory bandwidth correction? The 278.8 for the 2080 does not match the NV whitepapers.

 

[EarthDog]

BUMP. @W1zzard

(cant seem to edit this post??????)

Also had the question in another thread since wedesnday. Confirmation or if its just me would be great!

 

[dj-electric]

Also here's one from an RTX 2080 Ti:

 

[Tomgang]

Nvidia has done it again. They gimped turing like they gimped Maxwell (GTX 970) if any one remember. This time its just more official

 

[Sebastian-san]

Can you guys try using Aida 64 gpu test and looking at the real time measures?

 

[EarthDog]

Also here's one from an RTX 2080 Ti:

Thats the same texture fill rate I see with a 2080.

https://www.techpowerup.com/forums/threads/techpowerup-gpu-z-v2-11-0-released.247661/#post-3915605

 

[EarthDog]

@W1zzard

..... is this fix on the list as well?


TIA for the acknowledgement.

 

[dj-electric]

W1zz, GPU-Z has to be fixed to suite RTX cards.

Base clock slot became extremely irrelevant this time around.
Having default and current is fine when it is accurate - it isnt as for 2.11.0
No mention of RT or Tensor cores in any way.

Boost is just the official stuff, and while being in the same realm of Intel's CPU boost clocks, the numbers were not relevant since GTX 900 four years ago.
Maybe time to move those slots into real-time reading? or at least real time vs official in the box below.

No RTX 2080 Ti is going to operate at 1350Mhz core under 3D load.

Stuff has to change.

 

[bug]

W1zz, GPU-Z has to be fixed to suite RTX cards.

Base clock slot became extremely irrelevant this time around.
Having default and current is fine when it is accurate - it isnt as for 2.11.0
No mention of RT or Tensor cores in any way.

Boost is just the official stuff, and while being in the same realm of Intel's CPU boost clocks, the numbers were not relevant since GTX 900 four years ago.
Maybe time to move those slots into real-time reading? or at least real time vs official in the box below.

No RTX 2080 Ti is going to operate at 1350Mhz core under 3D load.

Stuff has to change.

Oh, good. For a moment there I was worried we ran out of people that can tell others what they need to do