In no way does any of the following reflect the word or opinions of BagoGames as a company nor does this article reflect on any of Bago’s other employees. This is article is based on the opinions of Brett Madigan, a BagoGames employee.
If you aren’t up to date on the whole Nvidia GTX 970 issue then here are my previous articles on the matter, Memory Allocation issue, The Lawsuit.
Ok now that’s out of the way. Nvidia ohhh Nvidia, what have you got yourself into? First we find out that you misled us about the 4GB VRAM total, in the initial specs list about the ROPs, as well as the L2 Cache being cut down, now there’s a pending class action lawsuit against you. What more could go wrong?
The Nvidia CEO Jen-Hsun came out with the following public apology.
“Hey everyone,
Some of you are disappointed that we didn’t clearly describe the segmented memory of GeForce GTX 970 when we launched it. I can see why, so let me address it.
We invented a new memory architecture in Maxwell. This new capability was created so that reduced-configurations of Maxwell can have a larger framebuffer – i.e., so that GTX 970 is not limited to 3GB, and can have an additional 1GB.
GTX 970 is a 4GB card. However, the upper 512MB of the additional 1GB is segmented and has reduced bandwidth. This is a good design because we were able to add an additional 1GB for GTX 970 and our software engineers can keep less frequently used data in the 512MB segment.
Unfortunately, we failed to communicate this internally to our marketing team, and externally to reviewers at launch.
Since then, Jonah Alben, our senior vice president of hardware engineering, provided a technical description of the design, which was captured well by several editors. Here’s one example from The Tech Report.
Instead of being excited that we invented a way to increase memory of the GTX 970 from 3GB to 4GB, some were disappointed that we didn’t better describe the segmented nature of the architecture for that last 1GB of memory.
This is understandable. But, let me be clear: Our only intention was to create the best GPU for you. We wanted GTX 970 to have 4GB of memory, as games are using more memory than ever.
The 4GB of memory on GTX 970 is used and useful to achieve the performance you are enjoying. And as ever, our engineers will continue to enhance game performance that you can regularly download using GeForce Experience.
This new feature of Maxwell should have been clearly detailed from the beginning.
We won’t let this happen again. We’ll do a better job next time.
Jen-Hsun”
Let’s begin by breaking down what Jen-Hsun said here paragraph by paragraph and try to piece together each statement with the current practice of Nvidia as a company.
“Some of you are disappointed that we didn’t clearly describe the segmented memory of GeForce GTX 970 when we launched it”
This would imply that you didn’t mislead us initially with the specs list you provided on your website for the GPU and actually told us about this memory allocation technique from the beginning. I would believe his statement if you actually mentioned it at all, which they didn’t until called out. So some clear misinterpretation is present here yes because if it wasn’t mentioned how would we know about it unless we looked for it, even after reviewers didn’t notice it?
“We invented a new memory architecture in Maxwell. This new capability was created so that reduced-configurations of Maxwell can have a larger framebuffer – i.e., so that GTX 970 is not limited to 3GB, and can have an additional 1GB.”
So apparently Nvidia is now claiming they created the very first GPU to ever use more than 3GB of VRAM… umm wait what. No Nvidia you were not the first to use 4GB’s of VRAM, that honor would go to AMD simply due to the R9 270x, the AMD vendor Sapphire has a 4GB model which came out in October 2013 whereas the GTX 970 was released in September 2014. For safe measure we could even mention the release date of the reference R9 290 which is a 4GB GPU which was released along side the R9 270x. The R9 290 is relevant as well because it’s performance is quite similar to that of the GTX 970. So again NO Nvidia you didn’t invent anything. AMD has been making 4GB native GPU’s since around 2013, in this case you are still behind as the vendor’s themselves have to add more VRAM to the cards.
To further investigate this “new invention” claim how about we take a look at the GTX 770 which also had a 4GB native model released by a few Nvidia vendors such as ASUS and EVGA slightly after the launch, the reference model was 2GB though. So Nvidia has allowed 4GB GPU’s in the past so for them to say they “invented” something to allow the Maxwell based cards to support 4GB’s of VRAM is crazy talk… also the GTX 980 doesn’t have this issue at all.
“GTX 970 is a 4GB card. However, the upper 512MB of the additional 1GB is segmented and has reduced bandwidth. This is a good design because we were able to add an additional 1GB for GTX 970 and our software engineers can keep less frequently used data in the 512MB segment.”
Sorry but in no way is this design a GOOD design by any modern GPU standards for both AMD and Nvidia, this is just slack on Nvidia’s part by not properly creating a 4GB VRAM memory pool but simply chopping it up into two separate pools, a 3.5GB pool alongside a significantly slower 512MB pool..
I also don’t understand where this magical “additional 1GB” came from because it was found that the main memory pool is 3.5GB so if you added 1GB ontop of that it would add up to 4.5GB + 512MB = 5GB… again you didn’t add 1GB if you are referring to your previous GPU’s which had 3GB then sure but you can’t just say we added a 1GB chunk and cut it up because either way it’s still 3.5GB as the main memory pool. So I have no idea what he is trying to say there.
Implying that “our software engineers can keep less frequently used data” in the 512MB pool is all good and great but that has to be programmed into the game or on the drivers side to allow less frequently used assets into that set of VRAM, which by the way when any game accesses that memory the game has a slight hiccup and framerate dip before regaining the previous stable FPS. How is that ideal for gaming if it is constantly trickling between the main memory pool and the slower one?
“Unfortunately, we failed to communicate this internally to our marketing team, and externally to reviewers at launch.”
Repeating yourself, but again you didn’t mention it therefore you can’t communicate something you CHOOSE to keep quite.
“Instead of being excited that we invented a way to increase memory of the GTX 970 from 3GB to 4GB, some were disappointed that we didn’t better describe the segmented nature of the architecture for that last 1GB of memory.”
AMD can do it with a 4GB – 8GB native memory pool (as for 8GB I don’t know if it is a 8GB pool or broken up into 2 sets of 4GB pools). Their way of dealing with 4GB seems to work fine without hindering the player experience… why did you have to make a stupid way of implementing 4GB to a single GPU. Explain the logic behind that choice?
Stop saying you invented this. NEWS FLASH YOU MAY HAVE INVENTED YOUR POINTLESS WAY OF DEALING WITH 4GB… for a single GPU but hold on why doesn’t this issue occur on the GTX 980. Explain that, why was this so called 3GB to 4GB “technology” only used thus far IN SECRET on the GTX 970? Hmmm.
“This is understandable. But, let me be clear: Our only intention was to create the best GPU for you. We wanted GTX 970 to have 4GB of memory, as games are using more memory than ever.”
You are so high up on that horse of your’s that you can’t see the real issue here… the issue is that you can’t take NO for an answer, you can’t accept your mistakes but decide to dig this hole further and further into the ground and make you guys seem like complete idiots.
“The 4GB of memory on GTX 970 is used and useful to achieve the performance you are enjoying. And as ever, our engineers will continue to enhance game performance that you can regularly download using GeForce Experience.”
I don’t know about you but my experience with Geforce Experience hasn’t helped any of my games, when I choose to use the settings optimizer why does it always choose the max settings or try to get me running games at 30FPS. Is it to much to ask for a frame cap which would set the settings to run in cohesion.
Also while I’m harping on Geforce Experience, just add multiple audio channels to ShadowPlay recordings OH MY LORDYYYY.
“This new feature of Maxwell should have been clearly detailed from the beginning.”
Correction, just like GPU Boost I DON’T WANT THIS FEATURE TO BE USED EVER. NOT ONCE.
“We won’t let this happen again. We’ll do a better job next time.”
What will it be, will you be better at lying to your customers or actually telling us what your doing because over the last 3 generations of GPU’s I have barely seen any game optimizations since mid 660ti era.
So that’s 700 and 900 series (800 I have left out due to being laptop oriented and I haven’t had experience with them) I haven’t seen any driver updates, excluding one of the the recent drivers, to have fixed any issues with games on Nvidia GPU’s nor have I seen Nvidia fix or mention better performance for any games.
This is called slacking off Nvidia, if you do not support your customers by trying to get the best performance possible out of the GPU’s that you created, why would I continue to support you after all this.
You have done nothing but ignore fans for the last few years and you do not care about fixing BSOD’s or even driver issues or issues related to any of your software suites, hell Physx hasn’t been fixed for the original Metro: Last Light yet nor has it had performance improvements.
Why claim things like realistic fire and water simulations running on 900 series GPU’s if you don’t even bother to fix Physx performance?
Source: Nvidia Blog, Memory Allocation Issue, The Lawsuit