Do amd cards actually last longer?

[NicholasNicholasNicholas]

So do they? Well...
Gpu aging is a combination of things which makes it complex, not a black or white issue. (Sorry it ended up quite a long post!).

1. AMD's GCN architecture is brute power, very high TFlops but difficult to extract peak performance, especially in DX11 which AMD is running crippled with single-threaded draw call submission. But overtime, there's opportunities for AMD to fine-tune drivers for each specific game to avoid being driver bound. You could say there's more room to grow for AMD GPUs, while NV's GPU are operating close to peak in DX11 already.

2. Console GCN. Optimizations by developers for GCN specific cache, wavefronts and shader efficiency carry through to the PC port. There's good presentations from Game Developers Conferences on this topic. It's going to continue due to PS4 Pro and Scorpio using GCN Polaris.

3. GCN architecture is iterated, evolution rather than revolution. The basic hierarchy remains constant, each SIMD has the same layout of ALUs (Vector and Scalar), each Compute Unit consists of the same layout of SIMDs. The result is that code that's optimized to run on GCN is nearly always (there's exceptions, differences in Tessellation & Async Compute) optimized for all GCN. Thus, the older GCN based cards like the 7970 and 290/X still powers through modern games.

4. DX12 &Vulkan allows developers closer access and importantly, rendering/draw calls can be multi-threaded (Async Compute is another bonus on top). This removes AMD's weakness of single-thread DX11. Thus, as more modern games use these new APIs, the better AMD GPU's look in comparison. Example, a 390X has similar compute performance as a 980Ti in Tflops, and it's only in these new API where AMD's GCN can really hit their peak. Hence, don't be surprised if some of these next-gen API games have AMD GCN cards punching above their weight (390X ~ 980Ti, Fury X ~ 1080 etc).

5. Usually more VRAM. Example, 7970 3GB vs GTX 680 2GB. There's games in recent times where the 2GB is a severe bottleneck and the 7970 3GB auto-wins, irrespective of #1-4. Likewise, 290/X 4GB vs 780/Ti 3GB. This is repeated recently with the 1060 3GB vs 470/480 4GB and 1060 6GB vs 480 8GB. Some posters falsely claimed that 3GB is enough for 1080p gaming but a recent review studying frame times find that to be utterly wrong. 3GB stutters in most modern games even on the 2nd highest settings (not maxed).

6. NV architectures evolve and also has revolutionary changes. Kepler -> Maxwell was a big leap, not only with the tile-based rasterization, but also the SMX layout, CUDA cores per SM went from 196 -> 128. This meant that games optimized for Maxwell's architecture would run un-optimized on Kepler, reducing much of it's shader utilization. It's why the 980, which on release, was only slightly above a 780TI, which was faster than a 970 by ~10%, but over time, we see the 780TI behind the 970 by ~10% or sometimes, even more.

7. Some folks have mentioned driver neglect from NV, that their "Game Ready" drivers only optimize non-legacy GPUs, ie, their latest & greatest. This is not gimping older stuff, that's a incorrect myth, it's more than NV focus optimizations on more recent stuff only.

All of these results in a potential for a big shift in performance over time. The 290X at the start of it's life was 10-15% behind a GTX 780Ti. These days it's very common to see it 10-15% ahead, with outliers much higher. If you read reviews over the years, you would have noticed the GTX 980 made the 290X (even custom models which run at ~390X performance) look like shit, often 20% lead. You would have noticed the 390X vs 980 situation is very different in recent times.

Before some of you accuse me of being an AMD fanboy, let me be clear. I am a fan of my money, it's important to me, how much value I can get out of my hard earned $. I have seen for myself since 2011, how well AMD GPUs aged in comparison with the NV GPUs that I've owned (GTX 670 and 780Ti) as I've multiple rigs for the family. It is this exact reason why I have an RX 480 now and will get Vega for the other rig, instead of going with Pascal.

[themctipers]

6 minutes ago, Nickathom said:

So do they? Well...
Gpu aging is a combination of things which makes it complex, not a black or white issue. (Sorry it ended up quite a long post!).

1. AMD's GCN architecture is brute power, very high TFlops but difficult to extract peak performance, especially in DX11 which AMD is running crippled with single-threaded draw call submission. But overtime, there's opportunities for AMD to fine-tune drivers for each specific game to avoid being driver bound. You could say there's more room to grow for AMD GPUs, while NV's GPU are operating close to peak in DX11 already.

2. Console GCN. Optimizations by developers for GCN specific cache, wavefronts and shader efficiency carry through to the PC port. There's good presentations from Game Developers Conferences on this topic. It's going to continue due to PS4 Pro and Scorpio using GCN Polaris.

3. GCN architecture is iterated, evolution rather than revolution. The basic hierarchy remains constant, each SIMD has the same layout of ALUs (Vector and Scalar), each Compute Unit consists of the same layout of SIMDs. The result is that code that's optimized to run on GCN is nearly always (there's exceptions, differences in Tessellation & Async Compute) optimized for all GCN. Thus, the older GCN based cards like the 7970 and 290/X still powers through modern games.

4. DX12 &Vulkan allows developers closer access and importantly, rendering/draw calls can be multi-threaded (Async Compute is another bonus on top). This removes AMD's weakness of single-thread DX11. Thus, as more modern games use these new APIs, the better AMD GPU's look in comparison. Example, a 390X has similar compute performance as a 980Ti in Tflops, and it's only in these new API where AMD's GCN can really hit their peak. Hence, don't be surprised if some of these next-gen API games have AMD GCN cards punching above their weight (390X ~ 980Ti, Fury X ~ 1080 etc).

5. Usually more VRAM. Example, 7970 3GB vs GTX 680 2GB. There's games in recent times where the 2GB is a severe bottleneck and the 7970 3GB auto-wins, irrespective of #1-4. Likewise, 290/X 4GB vs 780/Ti 3GB. This is repeated recently with the 1060 3GB vs 470/480 4GB and 1060 6GB vs 480 8GB. Some posters falsely claimed that 3GB is enough for 1080p gaming but a recent review studying frame times find that to be utterly wrong. 3GB stutters in most modern games even on the 2nd highest settings (not maxed).

6. NV architectures evolve and also has revolutionary changes. Kepler -> Maxwell was a big leap, not only with the tile-based rasterization, but also the SMX layout, CUDA cores per SM went from 196 -> 128. This meant that games optimized for Maxwell's architecture would run un-optimized on Kepler, reducing much of it's shader utilization. It's why the 980, which on release, was only slightly above a 780TI, which was faster than a 970 by ~10%, but over time, we see the 780TI behind the 970 by ~10% or sometimes, even more.

7. Some folks have mentioned driver neglect from NV, that their "Game Ready" drivers only optimize non-legacy GPUs, ie, their latest & greatest. This is not gimping older stuff, that's a incorrect myth, it's more than NV focus optimizations on more recent stuff only.

All of these results in a potential for a big shift in performance over time. The 290X at the start of it's life was 10-15% behind a GTX 780Ti. These days it's very common to see it 10-15% ahead, with outliers much higher. If you read reviews over the years, you would have noticed the GTX 980 made the 290X (even custom models which run at ~390X performance) look like shit, often 20% lead. You would have noticed the 390X vs 980 situation is very different in recent times.

Before some of you accuse me of being an AMD fanboy, let me be clear. I am a fan of my money, it's important to me, how much value I can get out of my hard earned $. I have seen for myself since 2011, how well AMD GPUs aged in comparison with the NV GPUs that I've owned (GTX 670 and 780Ti) as I've multiple rigs for the family. It is this exact reason why I have an RX 480 now and will get Vega for the other rig, instead of going with Pascal.

Invisible test, pls use Tx

[DIV1D3]

As you said, your post is a long one, so I haven't read it yet. But I'd like to say first off that on behalf of everyone using the Night Theme, please use paste without formatting, otherwise you have to highlight the text to be able to read it. 

[R3ep3r]

1 minute ago, themctipers said:

Invisible test, pls use Tx

Also I can't even read the text since I use the black theme.. and I totally hate when people decide to use a dark colored font.

PS: @Nickathom both cards get older at the same time and both will last longer if you buy what's that moments' HIGH END.

An example: I bought a GT430 GDDR3 2GB 6 years ago and I used it since two months or three ago when I bought a 1050ti.

[Blebekblebek]

Try to use HD 5000 or 6000 and let me know what you think.

Oh wait, they stop driver support since 2015.

[NicholasNicholasNicholas]

3 minutes ago, Blebekblebek said:

Try to use HD 5000 or 6000 and let me know what you think.

Oh wait, they stop driver support since 2015.

and lets see the competition... oh... i see... 200 series... no thank you. 

[Enderman]

Dude don't set it to black...

Set it to auto, otherwise it's still hard to see.

[Blebekblebek]

2 minutes ago, Nickathom said:

and lets see the competition... oh... i see... 200 series... no thank you. 

See, you already break your own theory.

at least 200 series still have driver till dec 2016, also proper windows 10 drivers support.

 

Also let's forget the 400 series even exist at all, sure that would make your point even better... not.

[NicholasNicholasNicholas]

Just now, Blebekblebek said:

See, you already break your own theory.

at least 200 series still have driver till dec 2016, also proper windows 10 drivers support.

 

Also let's forget the 400 series even exist at all, sure that would make your point even better... not.

thats not what i was saying. the 6000 series has aged well performance wise. 

[Blebekblebek]

1 hour ago, Nickathom said:

thats not what i was saying. the 6000 series has aged well performance wise. 

I thought this thread is all about AMD GPU Longevity.

Sure I'll give credits where it's due.

 

AMD had proper (at least for me) driver support only with GCN architecture because it's one architecture, and probably polaris as well.

But prior to that... meh.

it's even worse for people who bought enthusiast card such as HD 3890 HD 4890 HD 5990, and HD 6990 (not even considering the X2 card)

These cards released almost a year after official release and yet killed prematurely.

 

For me who owns HD 3000/4000/5000/6000 it's like dejavu every time  they announce news to stop driver support for their line up.

 

Point is, The GCN was all they've had and they bank on it hard.

Rehash, rebadge, rebrand.... it's like watching what nvidia does with their G92 chipset.

 

 

[xAcid9]

I'm not surprise if AMD drop support for pre-Polaris card when Vega drop in a few month. *knock on wood*

 

Also, give credit when you copy paste someone else opinion from interweb.

 

[Frankenburger]

It's probably a combination of GCN being ahead of its time and the fact Nvidia has been selling 256bit 80 class GPUs since 2012. If the GTX 680 was 384bit, like the HD7970, I'm pretty sure it would be holding up a lot better than what it is now.

 

Also, AMD is far more quick to drop support on older cards than Nvidia. The GTX400 series gpus are still getting driver updates, whereas the HD6000 series is not. I love how everyone who argues driver support in favor of AMD conveniently forgets this fact.

[jjohnthedon1]

I think amd has better driver improvements compared to nvdia

there cards launch below nvdia performance then after a year there 10 percent above, which is impressive.

but driver support no,

i still get driver updates on a laptop with a 420mx ffs not that they improve anything but I still get them