WCCftech: Memory allocation problem with GTX 970 [UPDATE] PCPer: NVidia response

[Head1985]

http://wccftech.com/users-reporting...970-graphics-cards-games-utilizing-full-vram/

I must say i have GTX970 and i have that problem.And we also discovered on other forum, that GTX970 is not 256bit card.
Thats i think is why GTX970 have some problems utilizing full 4096 Vram in games.
its similar to GTX660TI

We benched aida64 memory bandwidth and Both gtx980 and 970 if they have 256bit interface will have same results.
But we found results are not same.Look at MEMORY COPY thing
GTX970

GTX980

We think thats why GTX970 have those problems with memory allocation.

EDIT:Screenshots from far cry4 1920x1080 all max+8xMSAA
scene1
GTX980 http://abload.de/img/farcry4_2015_01_10_19f1ym3.jpg
GTX970 http://abload.de/img/farcry4_2015_01_10_23diuev.png

Scene2
GTX980 http://abload.de/img/farcry4_2015_01_10_1910a3l.jpg
GTX970 http://abload.de/img/farcry4_2015_01_10_232ik3k.png

 

[cmdrdredd]

No problem here with memory usage.

Full 4GB in use. 4k Crysis 3 with 8x MSAA(unplayable as hell but illustrates that I have no problems)

 

[Head1985]

Try shadow of mordor 1920x1080 Ultra textures.
GTX980 uses full 4GB Vram

Or try far cry4 1920x1080 all max + 8xMSAA
GTX980 again uses full 4GB Vram

watch Dogs 1920x1080 all max+ Ultra textures + 8xMSAA
Again GTX980 full 4GB Vram

We test it in same game in same setting and same scene on GTX970 and 980 and gtx970 alloacate onl 3500-3600MB in those games and GTX980 always full 4GB Vram

 

[cmdrdredd]

You do realize that if Crysis can eat 4GB that any other game could do the same right?

If it was a card problem it wouldn't be able to use 4GB at all.

 

[ShintaiDK]

Nothing wrong with the memory copy. Remember its an OpenCL test. So it depends a lot on the driver support. Example from a 384bit card:

I wonder if someone pulled a joke on wccftech again.

 

[Head1985]

You do realize that if Crysis can eat 4GB that any other game could do the same right?

If it was a card problem it wouldn't be able to use 4GB at all.

just try it pls

 

[ShintaiDK]

Here is one from a GTX980 Superclocked.

 

[cmdrdredd]

just try it pls

There's no point to. At no time anywhere does anyone state a specific part of the game or map that a 970 uses only 3.6GB compared to a maxed out 4GB on the 980. When I play Farcry 4 it hardly goes above 3.3GB until I up the resolution to 2560x1440 and then it rises up near 3.6GB with 8x MSAA. This is at the beginning of the game. What does that tell us? Nothing. I need to go above 3k resolution to max the memory at 4GB at which time the game becomes unplayable or locks up.

In all honesty I don't think you know what you are talking about. Shadow of Mordor loads up the map's texture assets to use your available memory on purpose. It doesn't dynamically stream it. It's an inefficient use of resources.

 

[Head1985]

Here is one from a GTX980 Superclocked.

Yeah we already know that.GTX 970 didnt use full 256bit memory bus because its not 256bit card.
Your results are normal with GTX980.

i am curius if cmdrdredd test games in 1920x1080 and manage full 4GB Vram usage.

Btw on czech forum we have user with both GTX970 and 980 now and he test it so far only in far cry 4 1920x1080 with 8xMSAA and with GTX970 he never manage use FULL 4GB Vram(mostly 3500-3600MB) and now he have GTX980 and GTX980 uses in same computer in same game with same drivers in same scene Full 4GB Vram.

 

[ShintaiDK]

Yeah we already know that.GTX 970 didnt use full 256bit memory bus because its not 256bit card.
Your results are normal with GTX980.

i am curius if cmdrdredd test games in 1920x1080 and manage full 4GB Vram usage.

The memory copy you talk about is ~130GB/sec vs ~160GB/sec. the GTX980 is roughly 20% faster than the GTX970. I dont see the issue. Remember this is an OpenCL test, not some raw performance test.

Arcording to you, the 384bit HD7970 I linked with 110GB/sec is what then?

 

[Black Octagon]

I wonder if someone pulled a joke on wccftech again.

Whenever that site publishes anything I err on the side of caution and assume they are either partially or completely deluded as to the content of the article

 

[Cloudfire777]

What kind of BS is this? GTX 970 isnt a 256bit card? Yes it is. Could it be sonething wrong with the card that was tested? Maybe.
VRAM usage have nothing to do with bus width so why wccftech mention that is beyond me.

Seriously they need to get it together

 

[Head1985]

The memory copy you talk about is ~130GB/sec vs ~160GB/sec. the GTX980 is roughly 20% faster than the GTX970. I dont see the issue. Remember this is an OpenCL test, not some raw performance test.

Its memory bandwidth benchmark!!!Try oc GPU and you will see same rusults because its memory bandwidth.Shaders are not involved.
Both gtx 970 and 980 uses 256bit but why gtx980 have 160GB/s and GTX970 130GB/s ???
And aida also underclock memory on GT970 and 980 to 6Ghz.
This is proof that GTX970 dont use full 256bit interface.
Edit user with both gtx980 and 970 test it on same computer with same drivers

 

[cmdrdredd]

Its memory bandwidth benchmark!!!Try oc GPU and will see same rusults because its memory bandwidth.Shaders are not involved.
Both gtx 970 and 980 uses 256bit but why gtx980 have 160GB/s and GTX970 130GB/s ???
And aida also underclock memory on GT980 and 980 to 6Ghz.
This is proof that GTX970 dont use full 256bit interface.

Wrong...Read a bit about what ShintaiDK posted. Look up what OpenCL is and how it functions.

 

[Head1985]

you are wrong its pure memory bandwidth benchmark we also tried Oc memory on GTX970 and we have increase in results

Look only on MEMORY COPY results
GTX970 with 6GHZ memory have +- 130GB/s
GTX970 with 8Ghz memory have 170GB/s
GTX980 with 6Ghz memory have 159GB/s

why GTX970 have worse results than GTX980 if both uses 256bit with same memory speed?
i know why because GTX970 didnt use full 256bit interface

 

[ShintaiDK]

If it was a pure memory benchmark, the HD7970 would have the highest numbers of the 3 and not the lowest.

But most importantly, we're introducing a brand new benchmark panel that offers a set of OpenCL GPGPU benchmarks that you can launch from AIDA64 / main menu / Tools / GPGPU Benchmarks. These benchmarks are designed to measure GPGPU computing performance via different OpenCL workloads. Every benchmark methods are designed to work on up to 16 GPUs, including AMD, Intel and nVIDIA GPUs, in any combination. Of course CrossFire and SLI configurations, and both dGPUs and APUs are also fully supported. HSA configurations are handled via preliminary support. Basically any computing capable device will be benchmarked that appears as a GPU device among OpenCL devices.

The OpenCL benchmark methods currently offered are not specifically optimized for any GPU architectures. Instead, the AIDA64 OpenCL module relies on the OpenCL compiler to optimize the OpenCL kernel to run best on the underlying hardware. The OpenCL kernels used for these benchmarks are compiled in real-time, using the actual OpenCL driver the OpenCL GPU device belongs to. Due to that approach, it is always best to have all video drivers (Catalyst, ForceWare, HD Graphics, etc) updated to their latest & greatest version. For compilation the following OpenCL compiler options are passed: -cl-fast-relaxed-math -cl-mad-enable.

And my GTX980 with 7Ghz had 157GB/sec. Lower than your 6Ghz GTX980 at 159Gb/sec.

 

[96Firebird]

why GTX970 have worse results than GTX980 if both uses 256bit with same memory speed?
i know why because GTX970 didnt use full 256bit interface

I love when people answer their own questions!

 

[Cloudfire777]

1253MHz vs 1291MHz.
104TMU vs 128TMU

 

[Head1985]

If it was a pure memory benchmark, the HD7970 would have the highest numbers of the 3 and not the lowest.



And my GTX980 with 7Ghz had 157GB/sec. Lower than your 6Ghz GTX980 at 159Gb/sec.

Your gtx980 was on 6Ghz.Aida64 underlock memory to 6Ghz
And if its shader involved benchmark why you didnt see increase in memory copy when you overclock GPU???
we talking about same architekture GTX970 vs 980.
when you overclock GPU you dont see increase in memory COPY benchmark.Tell me why?If its shader involved,
When you overclock memory you must have hard 8GHz in bios otherwise aida underlock it to 6Ght you see Memory copy increse.Tell me Why if its shader involved.

i think its because GTX970 have less memory bandwidth

 

[96Firebird]

Quick note though, for this test my graphics memory clock only went up to 6000MHz, then bumped up to 7000MHz at the last couple seconds.

You don't think underclocking memory is a flaw with the test?

 

[Deders]

Is memory compression involved when doing these benchmarks? might explain inconstancies.

 

[KaRLiToS]

If it was a pure memory benchmark, the HD7970 would have the highest numbers of the 3 and not the lowest.

And my GTX980 with 7Ghz had 157GB/sec. Lower than your 6Ghz GTX980 at 159Gb/sec.

I really don't understand why you are talking about the HD 7970 or the GTX 980 when the article is about the GTX 970.

You already have 25% of the posts in this thread and you don't even own a GTX 970.

The article says some users experience it, while the writter of the article claims he doesn't have that issue with his GTX 970.

 

[96Firebird]

I really don't understand why you are talking about the HD 7970 or the GTX 980 when the article is about the GTX 970.

You already have 25% of the posts in this thread and you don't even own a GTX 970.

Really not that hard to understand him...

If, as the OP is claiming, the 970 has a lower "Memory copy" result because it is not 256bit bus width, the 384bit 7970 should be ahead of even a 980. Especially since this "test" downclocks the memory to 6GHz, so memory speed is not a variable.

Alas, that is not the case, so the OP's conclusion is not held up with other tests.

 

[cmdrdredd]

I really don't understand why you are talking about the HD 7970 or the GTX 980 when the article is about the GTX 970.

You already have 25% of the posts in this thread and you don't even own a GTX 970.

Simple, the OP implied that lower memory copy scores = lower than 256bit memory interface. This is flatly false as indicated by the 7970 being 384bit (higher than 256bit) yet having a lower speed in the test.

 

[KaRLiToS]

Simple, the OP implied that lower memory copy scores = lower than 256bit memory interface. This is flatly false as indicated by the 7970 being 384bit (higher than 256bit) yet having a lower speed in the test.

Oh I see, I stand corrected.