discussion The GTX 970's Memory Explained & Tested (Comment Thread)

[Fragman]

We covered frame times in the initial SLI review and didn't find any issues. Personally, I wasn't going to spend time searching high and low for games that exhibited issues. It would be like beating a dead horse.

You said there was an article in the works covering GTX 970 frametimes when VRAM usage went over 3.5 GB but that never happened.

http://www.hardwarecanucks.com/foru...970-memory-allocation-issue-3.html#post785710

The very fact that you argued thats its a non issue in most cases while now saying it would be like beating a dead horse is contradictory. The games which are hitting above 3.5 GB VRAM usage are well known. Anyway how you run your website is upto you. But imo the tech press in general has not been as active in exposing the shortcomings of GTX 970 memory partition as they were with the HD 7900 frametimes when Nvidia pushed FCAT testing . Looks like when you are the market leader everyone takes a lenient view on your product's shortcomings or drawbacks and a stronger view when the drawback is with the smaller company's products.

 

[SKYMTL]

Yeah, the whole issue has been beaten like a dead horse. There were too many other things piling up as I spent hours trying to find exactly the right settings to replicate any issues whatsoever, so I pushed it aside. Call it creative license or a lack of time if you like.

 

[Bond007]

Yeah, the whole issue has been beaten like a dead horse. There were too many other things piling up as I spent hours trying to find exactly the right settings to replicate any issues whatsoever, so I pushed it aside. Call it creative license or a lack of time if you like.

Sounds reasonable. We all know now that it is not common that the odd configuration of memory causes issues, but there has to be some scenarios where issues will arise. As you said...its not worth any more of your time IMO.

 

[Sagath]

Evolve Benchmarked: Graphics & CPU Performance > Benchmarks: 2560x1600 - TechSpot

"Disappointingly, the GTX 970 averaged just 44fps, being only marginally faster than the old HD 7970 GHz at 41fps. We believe this massive reduction in performance is due to the GTX 970's partitioned memory configuration."

That makes no sense. The 780Ti gets better FPS with 1 Gig less ram (or 512+512 if you must). That clearly has nothing to do with the memory partitioning, as if it was caching down the PCIE bus to the RAM the 780TI would be even MORE affected then the 970.

Something else is going on.

 

[SKYMTL]

That makes no sense. The 780Ti gets better FPS with 1 Gig less ram (or 512+512 if you must). That clearly has nothing to do with the memory partitioning, as if it was caching down the PCIE bus to the RAM the 780TI would be even MORE affected then the 970.

Something else is going on.

Which situation are you referring to?

Remember, the GTX 780Ti's architecture has a massive advantage in Texture Unit count. In texture limited situations, Kepler will destroy Maxwell.

 

[Sagath]

Which situation are you referring to?

Remember, the GTX 780Ti's architecture has a massive advantage in Texture Unit count. In texture limited situations, Kepler will destroy Maxwell.

Sorry, edited my comment for clarity. I was refering to the quote in the evolve review that states its the memory configuration causing the 970's problems in Evolve at 1600p. Your comment makes sense to my 'something else is going on'. So I'm not sure how they come to the conclusion that the 970's bandwith is the problem. There is not comparison of why this is so that I understand.

 

[SKYMTL]

Sorry, edited my comment for clarity. I was refering to the quote in the evolve review that states its the memory configuration causing the 970's problems in Evolve at 1600p. Your comment makes sense to my 'something else is going on'. So I'm not sure how they come to the conclusion that the 970's bandwith is the problem. There is not comparison of why this is so that I understand.

Pushing that situation purely towards the memory allocation is preposterous and it highlights some of the misunderstandings about what's happening. Sadly, the media isn't making the situation any better since so many of them are using it as a "crutch" now when they don't know how to explain some numbers. A game can easily bottleneck the TMUs on Maxwell while consuming high amounts of memory but it is impossible to point towards the memory allocation being the culprit of poor performance versus Kepler cards.

 

[Fragman]

Pushing that situation purely towards the memory allocation is preposterous and it highlights some of the misunderstandings about what's happening. Sadly, the media isn't making the situation any better since so many of them are using it as a "crutch" now when they don't know how to explain some numbers. A game can easily bottleneck the TMUs on Maxwell while consuming high amounts of memory but it is impossible to point towards the memory allocation being the culprit of poor performance versus Kepler cards.

Evolve Benchmarked: Graphics & CPU Performance > Benchmarks: 2560x1600 - TechSpot

If you look at the techspot 1080p and 1440p numbers and you can see the 970 goes from being 11% slower at 1080p wrt GTX 980 to 20% slower at 1440p. Against the GTX 780 Ti it goes from being 6% faster at 1080p to 14% slower at 1440p. The GTX 970 has 81.25% of the TMUs and cores of a GTX 980 and 87.5% of ROPs of a GTX 980. As you might know performance never scales perfectly (1:1) with extra resources so even worst case we are looking at 15 - 17% lower performance on GTX 970 wrt GTX 980. The 20% lower performance is even more than the loss of TMUs and cores (in terms of %). The fact that Evolve is using 4 GB VRAM at 1440p and the drastic fall in perf from 1080p to 1440p makes us look at the memory partition as the culprit.

The fact is GTX 970 is being exposed in these latest games of 2015. The disconcerting point is this is happening within 6 months from launch. So it will only get worse from here for the most demanding games of 2015 and 2016. The GTX 970 is now a poor competitor against R9 290X which has that massive 512 bit memory bus and 4GB VRAM. The R9 290X continues to get better as time passes.

Nvidia cards like 780 Ti are getting hammered in TWIMTBP / Gameworks titles like Evolve. It must be embarassing for Nvidia to see their former flagship being schooled in a Gameworks title. R9 290X even beats the GTX 980 and thats even more surprising. I am sure the R9 3xx series will bring the much needed competition in the GPU market. Nvidia has roughly 3 months to milk the GTX 980 before its brought to the mid range segment and priced at around USD 350.

 

[Sagath]

Evolve Benchmarked: Graphics & CPU Performance > Benchmarks: 2560x1600 - TechSpot

If you look at the techspot 1080p and 1440p numbers and you can see the 970 goes from being 11% slower at 1080p wrt GTX 980 to 20% slower at 1440p. Against the GTX 780 Ti it goes from being 6% faster at 1080p to 14% slower at 1440p. The GTX 970 has 81.25% of the TMUs and cores of a GTX 980 and 87.5% of ROPs of a GTX 980. As you might know performance never scales perfectly (1:1) with extra resources so even worst case we are looking at 15 - 17% lower performance on GTX 970 wrt GTX 980. The 20% lower performance is even more than the loss of TMUs and cores (in terms of %). The fact that Evolve is using 4 GB VRAM at 1440p and the drastic fall in perf from 1080p to 1440p makes us look at the memory partition as the culprit.

So wait. I want to be sure I have this right...A card with 81.25% of the TMU's that an 980 has is getting 80% of the performance of a gtx980 at 1440p?

Well then...allow me to retort: As you might know, performance never scales perfectly (1:1) with extra resources, so even worst case we're looking at a perfect example of why this card is performing 20% slower then a card with 20% more TMUs, which is pritty damn ******* close to 1:1, and doesnt show a goddamn thing about it being the memory partition as the culprit, but probably being almost exactly 1:1 performance limited on TMU's and ROP's by 20%.

 

[YukonTrooper]

Those charts say to me the memory bus is coming into play, rather than the memory size. The 290, 780 Ti and Titan all end up surpassing the 970 on the high resolution benchmark, and the 290X surpasses both the 970 and 980. Furthermore, if the memory size was coming into play, you'd expect all the 3GB cards to plummet, but the 780 Ti stays strong and the rest of the 3GB cards gain considerably, including the 780 coming to within a couple FPS of the 970.