| Page 247 of 772 |
Kaltern
Posts: 5859
Location: Lockerbie, Scotland
|
 Posted: Sat, 24th Jan 2015 22:08 Post subject: |
|
 |
I was about to replace my 280X with a 970... not sure I should now 
Playing Valheim every weekday at 10pm GMT - twitch.tv/kaltern
Follow me on Twitter if you feel like it... @kaltern
My system: Ryzen 7 3700x|Gigabyte RTX 2080 Super Windforce OC|Vengeance 3000Mz 16Gb RAM|2x 500Gb Samsung EVO 970 M.2 SSD |SanDisk SSD PLUS 240 GB + OCZ Vertex 2 60Gb SSD|EVA Supernova 650W PSU|Logitech G27 Wheel|Logitech G19 Gaming Pad|SteelSeries Arctis 7|Logitech G502 Proteus Spectrum Mouse + Logitech MX Master Mouse|Razer Blackwidow Chroma X Keyboard|Oculus Quest 2 + Link|Pixio PX7 Prime 165hz HDR & 1x Samsung 24FG70FQUEN 144Hz curved monitor
-= Word to the wise: Having a higher forum post does not mean you are right. =-
|
|
| Back to top |
|
|
|
|
| Posted: Sat, 24th Jan 2015 22:09 Post subject: |
|
|
| Morphineus wrote: | So does that mean I get a 980 now?  |
980 shows same for some it seems if you check guru forums. you get the odd one guy who says im running 4k and its fine on ram use etc. But majority still saying they have same reporting to do regarding drop in last 500mb of ram bandwidth speed. |
|
| Back to top |
|
|
ixigia
[Moderator] Consigliere
Posts: 65097
Location: Italy
|
| Posted: Sun, 25th Jan 2015 00:19 Post subject: |
|
|
| consolitis wrote: | | ixigia wrote: |
Yep, I've read a similar thread at Guru3D last week, pretty curious (strange) results to be honest, I wonder if Nvidia has some official answer about this.
I made a few tests after I saw it (Far Cry 4/Unity, both with DSR 4x, Ultra Details and TXAA) and both the games used up to 3900+ish MB of VRAM, but the performance was obviously so poor that it was impossible to determine if there was the degradation mentioned there.  Will post the pics when I get home in the weekend. |
nvidia just responded that it's true the cards can more easily access ~3.5GB than the full 4 that eg the 980 can, but the performance drop when using the full 4GB should be pretty small. I'd like to see independent benchmarks though.
| Quote: | The GeForce GTX 970 is equipped with 4GB of dedicated graphics memory. However the 970 has a different configuration of SMs than the 980, and fewer crossbar resources to the memory system. To optimally manage memory traffic in this configuration, we segment graphics memory into a 3.5GB section and a 0.5GB section. The GPU has higher priority access to the 3.5GB section. When a game needs less than 3.5GB of video memory per draw command then it will only access the first partition, and 3rd party applications that measure memory usage will report 3.5GB of memory in use on GTX 970, but may report more for GTX 980 if there is more memory used by other commands. When a game requires more than 3.5GB of memory then we use both segments.
We understand there have been some questions about how the GTX 970 will perform when it accesses the 0.5GB memory segment. The best way to test that is to look at game performance. Compare a GTX 980 to a 970 on a game that uses less than 3.5GB. Then turn up the settings so the game needs more than 3.5GB and compare 980 and 970 performance again. |
| Quote: | | On GTX 980, Shadows of Mordor drops about 24% on GTX 980 and 25% on GTX 970, a 1% difference. On Battlefield 4, the drop is 47% on GTX 980 and 50% on GTX 970, a 3% difference. On CoD: AW, the drop is 41% on GTX 980 and 44% on GTX 970, a 3% difference. As you can see, there is very little change in the performance of the GTX 970 relative to GTX 980 on these games when it is using the 0.5GB segment. |
http://www.pcper.com/news/Graphics-Cards/NVIDIA-Responds-GTX-970-35GB-Memory-Issue |
Thanks for the update^^
Seems like the issue isn't huge, but there wasn't much transparency from mum Nvidia, and this kinda sucks.
The most intensive games do use all the memory available for me (Left: Far Cry 4, Ultra, TXAA, DSR4X - Right: Dragon Age Inquisition Ultra, AA4X, DSR4X) :
 
Though from the bandwidth test, the drop in those last 500MB is big, which (from what I can understand) may or may not cause issues (like stuttering for example) in games that use up to 3500MB of VRAM despite not influencing the average framerate itself (those 1%-5% losses are quite negligible).
It's not a tragedy, but..well, looks like those 4GB aren't exactly 4GB after all. More like 3500MB + some spare (slower) 500MB. We'll see what happens in the future, there's still a lot of confusion at the moment. Hopefully our tech experts will give us more thorough technical details |
|
| Back to top |
|
|
tonizito
VIP Member
Posts: 51445
Location: Portugal, the shithole of Europe.
|
| Posted: Sun, 25th Jan 2015 00:28 Post subject: |
|
|
For those facing some issues with this I'll post my address soon so you can send me those 970's instead of tossing them in the bin 
| boundle (thoughts on cracking AITD) wrote: | | i guess thouth if without a legit key the installation was rolling back we are all fucking then |
|
|
| Back to top |
|
|
|
|
| Posted: Sun, 25th Jan 2015 01:08 Post subject: |
|
|
where is the nvidia link not the pcper link please 
According to many, they asked nvidia direct and they said they know nothing about those revelations.
|
|
| Back to top |
|
|
ixigia
[Moderator] Consigliere
Posts: 65097
Location: Italy
|
|
| Back to top |
|
|
|
|
| Posted: Sun, 25th Jan 2015 07:35 Post subject: |
|
|
This memory issue is hilarious, when the memory really runs out (a.k.a goes over 3,5GB) then the card would be anyway at unplayable frame rates (same goes for 980). I havent seen anywhere near that kind of VRAM usage with highest settings with 2560x1080 resolution paired with some quality SMAA or so. Then this shit becomes issue if you are planning to play 4K resolution (or DSR) with taxing AA on. Bad shit from Nvidia not telling this before, but i really dont care since we cant do anything about it. 970 is still like the best card what i have owned performance and price wise.
|
|
| Back to top |
|
|
|
|
| Posted: Sun, 25th Jan 2015 10:47 Post subject: |
|
|
| Breezer_ wrote: | | I havent seen anywhere near that kind of VRAM usage with highest settings with 2560x1080 resolution paired with some quality SMAA or so. |
I think that'll happen very soon even at 1080p, especially considering how the new console ports are.
|
|
| Back to top |
|
|
|
|
| Posted: Sun, 25th Jan 2015 11:15 Post subject: |
|
|
Then its probably time to buy new GPU since these new ones are so outdated already, something like Titan 2 or AMD new flagship with 6GB VRAM or more. Gotta love PC gaeming.
|
|
| Back to top |
|
|
|
|
| Posted: Sun, 25th Jan 2015 11:43 Post subject: |
|
|
You were saying games don't even use the full VRAM, so I assume it doesn't bother you. When a GPU is advertised with 4GB RAM, you expect you can use all of that VRAM just fine. Many people buy a GPU for 3-4 years.
|
|
| Back to top |
|
|
|
|
| Posted: Sun, 25th Jan 2015 12:42 Post subject: |
|
|
Having 0 problems with MSI kombustor - 3824 allocated vram. Everything runs smooth - no stutters, no hangups.
|
|
| Back to top |
|
|
JBeckman
VIP Member
Posts: 35025
Location: Sweden
|
|
| Back to top |
|
|
|
|
| Posted: Sun, 25th Jan 2015 14:05 Post subject: |
|
|
If some unbiased source confirms that the frame pacing doesn't suffer and it's only a 1-3 fps difference, I'l agree that it's not a big deal.
As of now, users are complaining about stuttering (not about low average framerates) and nvidia responds with average framerate diagrams. Doesn't look too convincing so far.
|
|
| Back to top |
|
|
|
|
| Posted: Sun, 25th Jan 2015 14:29 Post subject: |
|
|
The fact nvidia admits they are doing their best in order to avoid using the last 0.5GB means there is indeed a non insignificant issue with that 0.5 or they wouldn't have bothered. On the other hand the fact it has gone unnoticed so long to me points that the issue is not catastrophic or at least not in all cases.
In any case a) nvidia should be blamed for not being transparent about it and b) the 970 remains a very good card and the one I would recommend to someone looking to buy a new GPU right now (those that can wait, should wait).
TWIN PEAKS is "something of a miracle."
"...like nothing else on television."
"a phenomenon."
"A tangled tale of sex, violence, power, junk food..."
"Like Nothing On Earth"
~ WHAT THEY'RE TRYING TO SAY CAN ONLY BE SEEN ~
http://www.youtube.com/watch?v=CHTUOgYNRzY
|
|
| Back to top |
|
|
|
|
| Posted: Sun, 25th Jan 2015 16:26 Post subject: |
|
|
played 1 hour watch dogs 3504MB usage, Titanfall 3620MB... so?
Ryzen 7 7800X3D, Noctua NH D15 Chromax, GIGABYTE RTX 5070 Ti Windforce, GIGABYTE X670 AORUS ELITE AX, Corsair 2x16GB DDR5 CL30
|
|
| Back to top |
|
|
ixigia
[Moderator] Consigliere
Posts: 65097
Location: Italy
|
| Posted: Mon, 26th Jan 2015 00:47 Post subject: |
|
|
| trollpower wrote: | | played 1 hour watch dogs 3504MB usage, Titanfall 3620MB... so? |
If you haven't noticed anything suspicious when trespassing the infamous ~3500 limit then you're good to go.
I haven't had major problems either (very similar result as Koroush Ghazi's acticle from tweakguides posted by JB), but I'll try to install Shadow of Murdhurr with the texture pack tomorrow, and see what happens once all those 4GB are completely filled without using artificial DSR' and the likes. Hopefully the card won't commit seppuku like my beloved 4870 did when trying RAGE. |
|
| Back to top |
|
|
|
|
| Posted: Mon, 26th Jan 2015 01:50 Post subject: |
|
|
guys please the guy that made the benchtest tool for what you are seeing has already said its not correct due to many reasons, every cuda card used produces the same readings 
and the tool itself is not even great for the ram monitoring but its the only thing available right now.
|
|
| Back to top |
|
|
Raccoon
Posts: 3160
Location: Poland
|
|
| Back to top |
|
|
Werelds
Special Little Man
Posts: 15098
Location: 0100111001001100
|
| Posted: Mon, 26th Jan 2015 10:36 Post subject: |
|
|
It all depends on what's in that last 512 megabytes of RAM; or more accurately, whether it is data that has to be shuffled around a lot.
With memory, it is divided into several chunks, one per chip. The physical chips have 32-bit interfaces and the memory controller has to have the "wires" for each of these interfaces to address that - it has an I/O interface for every 32 bits of memory bandwidth. That means that only multiples of 32-bit exist for bandwidth buses: 32, 64, 96, 128, 160, 192 and so on. The number of times 32 fits in there is the maximum amount of memory chips it can address (bare with me, more coming, because GDDR5 has a little trick). So, on a 256-bit memory bus, it can use 256/32 = 8 memory chips. On the 970/980 they use 4 Gbit GDDR5 - that's 512 MB per chip, times 8 means 4096 MB (4 GB) total.
Now, GDDR5 has a little trick up its sleeve, where it can be operated in what's known as "clamshell" mode. That means that on one of those 32-bit interfaces it operates 2 chips - it just alternates between addressing the 2 memory chips connected to it. That's how Nvidia have managed to put 4 GB of 2 Gbit memory on a 256-bit bus in the past and that's also why I've said that that is completely fucking pointless. You effectively cut the bandwidth to each chip in half if they're both in use. So yes, for those that have 4 GB 770's, if you go over 2 GB of memory usage, your memory bandwidth starts dropping drastically. There's no "ifs", no "buts" and you can claim that I'm wrong all you want, but the electrical construction simply does not allow for you to address all 4 GB at 224 GB/s. That is not technically possible, period.
Now, the memory controller sits inside the actual GPU and is connected to what's known as a crossbar. This crossbar essentially is just a buffered bus. Maxwell's SPs* are all grouped into SMMs** and then 4 of those SMMs in turn are grouped into GPCs - either the GPCs or the SMMs are each connected to the crossbar too (that's how they get to the memory). Now, it's impossible to see this on a die shot, but apparently Nvidia's crossbar is designed in such a way that it prohibits the 4 GPCs to fully address all of the memory controller's interfaces. What they've effectively done is reduce the memory bus to 192 bits, which means that the last 512 MB is more or less operating in clamshell mode.
Will that affect everything? No, it will only have a negative effect when that last 512 MB holds important data that has to be read/written to a lot. Since 90% of games barely even need 2 GB, it's going to do absolutely nothing. Again, I've said this before, but while in some games you will see all 4 GB in use, that doesn't mean all 4 GB are being used actively. 50% of it will merely be used as a cache and thus accessed very infrequently, if at all. You have to push to 4K to truly get close to 4 GB of VRAM usage and that's the only case where you'll see performance drops because then that last 512 MB at some point will hold the framebuffer and it has to start swapping that at half the bandwidth.
* SP = Streaming Processor..or we can use the marketing term, CUDA core.
** For those that know what Fermi and Kepler look like: SMM = SM in Fermi = SMX in Kepler. Ironically, Nvidia called the SMMs "SM" in their own press statement about this issue.
|
|
| Back to top |
|
|
|
|
| Posted: Mon, 26th Jan 2015 11:03 Post subject: |
|
|
Last edited by paxsali on Thu, 4th Jul 2024 21:53; edited 1 time in total
|
|
| Back to top |
|
|
Werelds
Special Little Man
Posts: 15098
Location: 0100111001001100
|
| Posted: Mon, 26th Jan 2015 11:11 Post subject: |
|
|
This tech just interests me and I've been actively following it and reading up about it for 15 years 
And it's not insignificant really
|
|
| Back to top |
|
|
Raccoon
Posts: 3160
Location: Poland
|
|
| Back to top |
|
|
|
|
| Posted: Mon, 26th Jan 2015 11:12 Post subject: |
|
|
Last edited by paxsali on Thu, 4th Jul 2024 21:53; edited 1 time in total
|
|
| Back to top |
|
|
Raccoon
Posts: 3160
Location: Poland
|
| Posted: Mon, 26th Jan 2015 11:17 Post subject: |
|
|
I never claimed it was.
"Anatidaephobia is the fear that somewhere in the world there is a duck watching you."
Fuck Polish government for oppressing women!
"People are such awful conversationalists that they’ll interrupt you when you’re answering a question they asked."
|
|
| Back to top |
|
|
|
|
|
| Back to top |
|
|
Morphineus
VIP Member
Posts: 24883
Location: Sweden
|
| Posted: Mon, 26th Jan 2015 16:28 Post subject: |
|
|
You should try at least!
Most of Werelds posts are quite interesting, mostly filled with very specific information. It's an opportunity to learn some details. The scary part: those are just snippets. 
I thought it was quite good on conveying it in an understandable piece. Which is not an easy task for most technical people.
So...  for werelds!
|
|
| Back to top |
|
|
couleur
[Moderator] Janitor
Posts: 14369
|
|
| Back to top |
|
|
|
|
| Posted: Mon, 26th Jan 2015 20:27 Post subject: |
|
|
Details: http://www.pcper.com/reviews/Graphics-Cards/NVIDIA-Discloses-Full-Memory-Structure-and-Limitations-GTX-970
| Quote: | | First, despite initial reviews and information from NVIDIA, the GTX 970 actually has fewer ROPs and less L2 cache than the GTX 980. NVIDIA says this was an error in the reviewer’s guide and a misunderstanding between the engineering team and the technical PR team on how the architecture itself functioned. That means the GTX 970 has 56 ROPs and 1792 KB of L2 cache compared to 64 ROPs and 2048 KB of L2 cache for the GTX 980. Before people complain about the ROP count difference as a performance bottleneck, keep in mind that the 13 SMMs in the GTX 970 can only output 52 pixels/clock and the seven segments of 8 ROPs each (56 total) can handle 56 pixels/clock. The SMMs are the bottleneck, not the ROPs. |
Apparently nvidia claims they just now realized the marketing department made a "mistake" regarding the 970's ROPs and L2 cache All those months they had no idea the wrong numbers had been reported to the press 
| Quote: | | NVIDIA’s performance labs continue to work away at finding examples of this occurring and the consensus seems to be something in the 4-6% range. A GTX 970 without this memory pool division would run 4-6% faster than the GTX 970s selling today in high memory utilization scenarios. |
Anyway, no independent benchmarks found yet.
TWIN PEAKS is "something of a miracle."
"...like nothing else on television."
"a phenomenon."
"A tangled tale of sex, violence, power, junk food..."
"Like Nothing On Earth"
~ WHAT THEY'RE TRYING TO SAY CAN ONLY BE SEEN ~
http://www.youtube.com/watch?v=CHTUOgYNRzY
|
|
| Back to top |
|
|
|
|
| Posted: Mon, 26th Jan 2015 20:31 Post subject: |
|
|
|
|
|
| Back to top |
|
|
|
|
| Posted: Mon, 26th Jan 2015 20:40 Post subject: |
|
|
I think that now everyone can RMA their 970 claiming it has different than the reported specs. Since it's a #truthfact.
But it's still a really really good card, the problem is fucking nvidia who didn't have the balls to explain the situation from the beginning. People would still buy it, since it's excellent value for money. There was no need for a late reveal and all this drama.
TWIN PEAKS is "something of a miracle."
"...like nothing else on television."
"a phenomenon."
"A tangled tale of sex, violence, power, junk food..."
"Like Nothing On Earth"
~ WHAT THEY'RE TRYING TO SAY CAN ONLY BE SEEN ~
http://www.youtube.com/watch?v=CHTUOgYNRzY
|
|
| Back to top |
|
|
| Page 247 of 772 |
All times are GMT + 1 Hour |
