AMD GPU gets better with 'age'

Tests for draw calls with different CPUs, for AMD and nvidia cards. A couple of very strange results.

http://www.eurogamer.net/articles/digitalfoundry-2015-why-directx-12-is-a-gamechanger

Quoting myself to point out that their table is flat out wrong and was probably put together by someone very drunk.

There are a couple of strange results still, but not as much as before.

https://docs.google.com/spreadsheets/d/1NoxzsMt7-inVKjGHw1IOW3fyPq0WaHNHFiBPZh5-EM4/edit?pli=1#gid=0
 
It depends. GPUs are bad at very small draw calls. AMDs current sweet spot seems to be around 256 vertices per draw call (~256 primitives). Anything below that slows down your triangle rate. ...
I was speaking more from an API perspective and the advice I remember reading wayback in the early/mid 2k's was that small batches of triangles were bad, thanks for putting some numbers on the hardware side of the subject.
 
Another game where AMD GPUs fare worse with slow CPUs:

The results here are fascinating: AMD's Radeon R7 260X is the better, slightly faster, and cheaper card. However, owing to the less efficient driver, the card performs under-par when combined with a budget CPU. This is most noticeable in the third segment of the benching sequence, where we see noticeable stutter. Meanwhile, performance between i3 and i7 when paired with the GTX 750 Ti sees no material difference. This strongly suggests that the Nvidia card will provide the best overall sustained performance throughout the game.

http://www.eurogamer.net/articles/digitalfoundry-2015-the-best-pc-hardware-for-the-witcher-3
 


I dont want to look like an Anti Gamework , or TWITTMP games, but do you have games who show the same behavior who are'nt part of them ? because i cant imagine how big should be the overhead on thoses games right now for the AMD driver for solve problems, they should not have to solve primary.

The next one for you to investigate: PC cars ... With a nearly 200% performance difference between the AMD and Nvidia gpu's ( 780TI-290X ) i think the AMD driver will need some miracle for go close of it, and so increase so much the overhead driver that it should be funny to see. ( WHen i think we have been thousands and thousands of AMD gpu's owner to test this games between 2011 and 2015, report everything to WDM ( slighty mad studio ) .. the result is somewhat funny,.
 
Last edited:
While is seems AMD is having some problems on recent (nvidia GW) games....i saw this summary graph
kflxGyB.png


Pretty shocking the modest 290X is hanging with the once more expensive...big Keplar!
 
While is seems AMD is having some problems on recent (nvidia GW) games....i saw this summary graph

Have a link to the actual review? Kepler issues are known and a fix is imminent .... sounds bugs have been identified that affect multiple games.
A few days ago Nvidia reported on working on driver updates to address Kepler issues with The Witcher 3. It appears though, that Kepler owners will enjoy a performance boost in several games with the new driver. Nvidia's ManuelG noted that the upcoming drivers will offer a universal fix for Kepler GPUs that will boost performance in various PC games.
http://www.guru3d.com/news-story/the-witcher-3-no-xp-glitch-fix-coming-on-june-1st.html
 
Last edited by a moderator:
More shocking imo is the difference between the 770 and 280X.

It's a 2560*1600 test, and predictably the GPUs hurting the most are the ones with only 2GB of memory.
Which is probably the reason why the R9 280 is ahead of the 285 and the GTX 960.
 
Did you happen to notice that every single AMD card, regardless of CPU had stuttering issues in that game? They specifically mention the latency spikes over and over again.

This isn't a "slow CPU" issue.

I'm playing in 2560x1440 on my R9 290X and TBH I don't see any latency spikes during gameplay but there are some in the cut scenes. It happens very rarely though. To me BF4 was much worse initially, especially in Mantle ...
 
Amid the 980Ti reviews, from AT and TR testing, i am surprised 290X is hanging in there with the 780Ti. On cases, it wins despite the crazy frame times...i think that is impressive considering AMD card came in at $150 less and was touted to compete with only the 780....AMD cards do get better with age!? It could also be the 512bit 4GB VRAM leading the way.

I look forward to 390X at $550 for history to repeat?? I won't mind the water cooled version at $600...will there be a 390 with unlocked shaders again?? The Fiji SKU seem to be 390X aircooled and Radeon Fury AIO cooled, this round. The other SKU are all rebrands..
 
I'm pretty sure it's the influence of console games being optimised for GCN. We probably haven't seen the end of those optimisations yet either so it's likely AMD GPU's will continue to catch up with Nvidia GPU's that were previously faster. To what extent we have already covered what's possible and what yet remains I have no idea though. Also, the further AMD moves away from GCN 1.1 (we're expecting Fiji to be 1.3) the less those optimisations should apply in theory. Although that certainly doesn't seem to have hurt the 285 in any way so far.
 
It's also worth noting that all the recent benchmarks are at 4K and occasionally 2560x1440 and these higher resolutions, especially 4K were always more biased in favour of the 290x. The earlier benchmarks around the 780Ti's release were generally at 1080p primarily and so made it's lead look much larger. I guess it all comes down to what resolution you want to game at.
 
Performance per watt is currently where AMD is technologically behind nVidia. GCN 1.0, 1.1 and 1.2 all have major issues on this. Didn't VLIW4 have some as well?
 
Last edited:
Back
Top