3 hours ago, Rob_Ainscough said:

I agree monitoring GPU memory "allocation" may not be 100% accurate, but I'm willing to bet it's at least 90% accurate.

Very interesting blog Rob, thanks for sharing it.  I am as far from a coder as you can get, but it seems that there's a bit of a contradiction in that statement.  According to the blog post .....

" Before CUDA 10.2, the only way to implement this concept in CUDA was to either use cudaMalloc, cudaFree, and cudaMemcpy, or to use cudaMallocManaged and cudaPrefetchAsync to commit the memory as you need it."

Also, a bit further down,

"While the implementation is fairly straightforward,"

(referring to the method above)

" there are a number of performance implications.

• The cudaMalloc function allocates more than what it needs to grow the allocation. To grow, you need to keep the old allocation and allocate a new allocation with enough room for the old allocation and the extra space, which significantly reduces how much you can grow. If the device only has 2 GiB of memory and you already have a 1 GiB vector, you can’t grow it any larger, as you would need 1 GiB plus how much you need to grow. Effectively, you can’t grow a vector that is larger than half of the GPU memory.
• Each allocation must be mapped to all peer contexts, even if it is never used in those peer contexts.
• The cudaMemcpy call adds latency to the growing request and uses precious memory bandwidth to duplicate data. This bandwidth could be better spent elsewhere.
• The cudaFree call waits for all pending work on the current context (and all the peer contexts as well) before proceeding."

So it seems to me that the article is stating that there are methods to distribute memory resources more efficiently and precisely than the quoted context.  Are you saying that optimistically, there is only 10% more accuracy to be gained with optimizations.  I'm not being contrarian, just curious.  

What are your thoughts on what performance would be like with the RTX3000 "current" line and DX12 and Vulcan when it comes to simulators that use those engines?  Seeing the DX11 sim results, I'm not very stoked.   

 

Edited September 18, 20205 yr by jalbino59

25 minutes ago, Rob_Ainscough said:

I'm not a graphics expert either, hopefully it doesn't come across that way.

Not at all, your insights are appreciated, I've gathered a wealth of good information from videos or posts you've made. Keep the information coming!

 

26 minutes ago, Rob_Ainscough said:

I think MSFS needs to get onto DX12 ASP, but I have zero experience with any 3000 series GPU, hopefully next week but who knows if I can defeat the buy Bots.

As do I.  Good luck with those bots! If you do manage to get your hands on one, please post your impressions.  

20 hours ago, micstatic said:

I mean it's not total nonsense.  I'm personally looking forward to bring my texture resolution back to 4096 instead of 2048.  that won't be a challenge on the 3090

Obviously you are right that the 3090 is the thing to have but at more than twice the price of a 3080, the latter is going to be the choice for many.

On 9/17/2020 at 2:24 PM, Chapstick said:

Not enough VRAM unless you can afford the 3090.

See https://videocardz.com/newz/gigabyte-confirms-geforce-rtx-3060-8gb-rtx-3070-16gb-and-rtx-3080-20gb

2 hours ago, pgde said:

See https://videocardz.com/newz/gigabyte-confirms-geforce-rtx-3060-8gb-rtx-3070-16gb-and-rtx-3080-20gb

Yes, those are the cards could be worth waiting for if you're using V5. 

I'm getting a lot of flak for saying the current 3000 series SKUs doesn't have enough VRAM, but I'm right. If you're on an 8GB card, spending $700+ for 2 measly GB more isn't worth it. 

Well, you are in good company. Virtually all reviewers that I have seen/read ding Nvidia for only 10Gb of VRAM which is why I think the AIBs will be adding VRAM to their cards. I agree that DX12 will be much more VRAM hungry!

3 hours ago, pgde said:

Well, you are in good company. Virtually all reviewers that I have seen/read ding Nvidia for only 10Gb of VRAM which is why I think the AIBs will be adding VRAM to their cards. I agree that DX12 will be much more VRAM hungry!

The AIB`s will wait for Nvidia to release the next cards then mod them.