http://www.hardocp.com/article/2012...gpu_gaming_performance_review/14#.UVCW8RfbhBk
Yeah, that was their theory...
-
Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!
PCI Express 3.0
- Thread starter Lionstarr
- Start date
Yeah, that was their theory...
I also have a theory on why AMD video cards did not benefit from the PCI-E 3.0 as much as Nvidia cards did at Hardocp review. The nvidia cards needs more buss speed do to the lack of 2GB Vram compaird to AMD 3GB so the nvidia cards use the PCI-E buss bandwidth more.
Hardocp theory is SB vs Ivy Bridge system does have a better IPC, somewhere in the neighborhood of 4% to 7% by there calculations and the nvidia card can take advantage of that however AMD cant.
Hardocp theory is SB vs Ivy Bridge system does have a better IPC, somewhere in the neighborhood of 4% to 7% by there calculations and the nvidia card can take advantage of that however AMD cant.
- Joined
- Feb 6, 2011
The X79 MoBo's are HW ready for PCIe-3.
Interesting... however, wouldnt a card with more throughput, like the 384bit AMD cards tend to process more data than a 256bit Nvidia card? Look at GPUz for a 680 and a 7970 and the bandwidth available... the difference is significant.
Yes the AMD cards have more bandwidth on the Vram memory bus to the GPU however the PCI-E 3.0-2.0 bus is from system memory and is much slower than the memory bus on the video card, so as my theory goes if the AMD already has all the data in the 3GB of Vram then it does not depend on the PCI-E 3.0 bus as much as nvidia with the 2GB Vram. So nvidia will receive more data from the PCI-E 3.0 bus do to the lack of data in the 2GB of Vram and see more improvement with the PCI-E 3.0 bus than AMD, this acording to what I see with Hardocp review.
My feeling is that is why AMD has done so much better with the driver updates by utilizing more bandwidth and Vram that they have.
One thing I found interesting At Hardocp PCI-E 3.0 review was the AMD GPUz take a look at it's only reporting 32bit bus.
http://www.hardocp.com/article/2012/07/18/pci_express_20_vs_30_gpu_gaming_performance_review/
That GPUz, I'm guessing, is simply an error? No clue as all others say 384bit at that time (just google 7970 GPUz and click on images).
PCIe lanes and speed are on the CPU die...Not sure it has much if anything to do with system memory(RAM)... but again, I don't know. I just looked that this.
That said, I dont understand what you are trying to say.. it doesnt make sense to me. What you are saying would be situation specific in that 2GB would have to be saturated to experience what I think you are trying to say. I can tell you memory use has not increased from the 12.1b drivers compared to before though...
EDIT: I see now...ish.
PCIe lanes and speed are on the CPU die...Not sure it has much if anything to do with system memory(RAM)... but again, I don't know. I just looked that this.
That said, I dont understand what you are trying to say.. it doesnt make sense to me. What you are saying would be situation specific in that 2GB would have to be saturated to experience what I think you are trying to say. I can tell you memory use has not increased from the 12.1b drivers compared to before though...
EDIT: I see now...ish.
So wouldnt that occur only if you break the 2GB barrier... which with the resolution listed, they may? But that is good thinking assuming I understand it... and assuming they broke 2GB... lots of assumptions though.. but a logical jump for sure.
Last edited:
Yes I was point that out because I was wondering if there was a problem in his testing, I know the AMD bus is 384bit
Yes everything goes through the CPU I just did not want to make it complicated because the data for the PCI bus comes from memory and the CPU can call for data in blocks and sends it directly to the GPU in big blocks of data with using the memory bus and PCI-E bus, however it's slower than the Vram memory bus.
That said, I dont understand what you are trying to say.. it doesnt make sense to me. What you are saying would be situation specific in that 2GB would have to be saturated to experience what I think you are trying to say. I can tell you memory use has not increased from the 12.1b drivers compared to before though...
EDIT: I see now...ish.
So wouldnt that occur only if you break the 2GB barrier... which the games listed and resolutions do not?
like system memory the more you have the more some games will use. when some games load they adjust to the amount of Vram you have and most games have a minim requirement for Vram however I see it all the time in the forms people playing the same games at the same resolution and they are using different amount of Vram like my GTX 570 with 1.2GB with BF3 maxed 1920X1080. So what i'm trying to say if you have 2GB of Vram the game maybe will only use 1.5GB if you have 3GB of Vram the same game could use 2GB of vram. So in Hardocp review some games benefited with nvidia 2GB vram with the PCI-E bus 3.0 where as AMD did not with 3GB of vram and PCI-E 3.0.
You only test with the best equipment, there is allot of people playing new games 1920X1080 with lesser equipment like 1.0GB of Vram SLI and there playing fine maxed out. so I think there is allot more dynamic adjusting than what we think there is.
There has to be a reason why nvidia does better than AMD with some games with PCI-E 3.0 and like I said I don't think it because nvidia does better than AMD with IPC that is Hardocp theory.
Anyway most of the improvements with 3.0 is within a margin of error or so small of a % that there is no noticeable improvement in real life gaming.
Battlefield 3 GTX 550 Ti SLI Conquest Assault (Ultra Settings)
There has to be a reason why nvidia does better than AMD with some games with PCI-E 3.0 and like I said I don't think it because nvidia does better than AMD with IPC that is Hardocp theory.
Anyway most of the improvements with 3.0 is within a margin of error or so small of a % that there is no noticeable improvement in real life gaming.
Battlefield 3 GTX 550 Ti SLI Conquest Assault (Ultra Settings)
Last edited:
- Joined
- Nov 1, 2009
- Location
- Austin, TX
so far... the only benches i've found that actually show a benefit for pcie 3.0 on current systems, is aa slight benefit for CUDA computing on the Nvida side (didn't see tests on the ATI front) and then quad SLI (16x,8x,8x,8x) tri-monitor setups. like in the neighborhood of 4300x2300 resolution. the test i saw was a o/c'd 3960x with the Nvidia registry hack. heck, the guy's rig might have been on 120hz running 3D, too. here's the link: http://www.overclock.net/t/1220962/...y-and-computer-edition-2012/370#post_16915399
he was also using gtx 680 2G cards, so it's possible the Vram might have been choking his 2.0 tests, too
he was also using gtx 680 2G cards, so it's possible the Vram might have been choking his 2.0 tests, too
+1 That is what I was saying above, if you have less Vram 2.0 GB with a 4320X2304 res in BF3 you can go from 69 FPS to 133 FPS and heaven 3.0 66 FPS 94 FPS with going from PCI-E 2.0 to 3.0 because your more dependent on the PCI-E bus for data than Vram.
- Joined
- Oct 21, 2012
- Location
- Phoenix, Az
man it's threads like this that keep me coming back. I feel like i am getting one heck of an education in video cards right now. all sorts of good info here.
Yes I also like threads like this it reminds me of how the PC works and how it has progressed also things that are forgotten over time.
like from PCI video to the Accelerated Graphics Port , AGP was the fist time the texture maps could be directly loaded to the GPU from system memory.
Now with PCI-E 2.0-3.0 the drivers and software interact with the CPU to fetch texture maps and data much much faster from system memory to the VRAM.
QUOTE:As the GPU creates images, it needs somewhere to hold information and completed pictures. It uses the card's VRAM for this purpose, storing data about each pixel, its color and its location on the screen. Part of the VRAM will also act as a frame buffer, meaning that it holds completed images until it is time to display them. Typically, video RAM operates at very high speeds and is dual ported, meaning that the system can read from it and write to it at the same time.
Since the vram is dual ported the GPU and CPU can read and write to it at the same time that is where PCI-E 3.0 can really help out if you don't have allot of Vram.
So if you don't have enough Vram for the game your running the frame buffer will overload and your pc will crash or not allow you to run the game. Note I made 2 calls to nvidia to get some of this information and both of them said the same thing.
I found this how a nvidia card works if your interested. T&L is transform and lighting.
QUOTE:Ok, first, your CPU sends data to your video card to be cached in VRAM. This is what it's there for, to hold pre and post-processed data.
Next, Geometry and Commands are sent to the Pre-T&L Cache in VRAM. Vertex Shading (T&L) then takes place, and the data is sent to the Post-T&L Cache in VRAM. Then, this data is sent to Triangle Setup and Rasterization.
Now, what about Textures? They're sent to the... take a guess... Texture Cache in VRAM. Fragment Shading and Raster Operations are then executed on those Textures and semi-processed Geometry from the Triangle Setup and Rasterization stages.
All this is then sent to the frame buffer, which is located within your VRAM. Now what? The RAMDAC sends this info out through your VGA/DVI port.
like from PCI video to the Accelerated Graphics Port , AGP was the fist time the texture maps could be directly loaded to the GPU from system memory.
Now with PCI-E 2.0-3.0 the drivers and software interact with the CPU to fetch texture maps and data much much faster from system memory to the VRAM.
QUOTE:As the GPU creates images, it needs somewhere to hold information and completed pictures. It uses the card's VRAM for this purpose, storing data about each pixel, its color and its location on the screen. Part of the VRAM will also act as a frame buffer, meaning that it holds completed images until it is time to display them. Typically, video RAM operates at very high speeds and is dual ported, meaning that the system can read from it and write to it at the same time.
Since the vram is dual ported the GPU and CPU can read and write to it at the same time that is where PCI-E 3.0 can really help out if you don't have allot of Vram.
So if you don't have enough Vram for the game your running the frame buffer will overload and your pc will crash or not allow you to run the game. Note I made 2 calls to nvidia to get some of this information and both of them said the same thing.
I found this how a nvidia card works if your interested. T&L is transform and lighting.
QUOTE:Ok, first, your CPU sends data to your video card to be cached in VRAM. This is what it's there for, to hold pre and post-processed data.
Next, Geometry and Commands are sent to the Pre-T&L Cache in VRAM. Vertex Shading (T&L) then takes place, and the data is sent to the Post-T&L Cache in VRAM. Then, this data is sent to Triangle Setup and Rasterization.
Now, what about Textures? They're sent to the... take a guess... Texture Cache in VRAM. Fragment Shading and Raster Operations are then executed on those Textures and semi-processed Geometry from the Triangle Setup and Rasterization stages.
All this is then sent to the frame buffer, which is located within your VRAM. Now what? The RAMDAC sends this info out through your VGA/DVI port.
Last edited:
- Joined
- Apr 25, 2002
- Location
- San Antonio, TX
Nice explanation
One minor thing, RAMDAC isn't used on digital interfaces like HDMI/DVI. They use a TMDS module instead.No AA in that vid... we have always said you will need to turn it off with 1GB at 1080p...You only test with the best equipment, there is allot of people playing new games 1920X1080 with lesser equipment like 1.0GB of Vram SLI and there playing fine maxed out. so I think there is allot more dynamic adjusting than what we think there is.
There has to be a reason why nvidia does better than AMD with some games with PCI-E 3.0 and like I said I don't think it because nvidia does better than AMD with IPC that is Hardocp theory.
Anyway most of the improvements with 3.0 is within a margin of error or so small of a % that there is no noticeable improvement in real life gaming.
Battlefield 3 GTX 550 Ti SLI Conquest Assault (Ultra Settings)
. LIke I said, BF3 is the only game that seems to adjust vram on the fly like that. 1GB @ 1080p with 4xMSAA will hurt those cards a bit however. It was theorized that the game lowers IQ, but nobody could produce results (I didnt get a response from EA), nor post screenshots.http://www.overclockers.com/forums/showthread.php?t=704750
Ive tested this as low as a 7850/660...
Last edited:
- Joined
- Nov 1, 2009
- Location
- Austin, TX
Sticky! Lots of good info here!
- Joined
- Nov 1, 2009
- Location
- Austin, TX
Oh, and S_I_N and manu... brief history of the sandy bridge E. It came out before pcie 3.0 vid cards did, and launched before they could get certification as a result. Therefore, even though the sandy bridge E is capable of pcie 3.0 bandwidth, it's not certified, and therefore not officially supported by intel or nVidia. The gfx drivers only support 2.0, but NVidia released a patch that changes registry to allow 3.0. You also need a mobo that will allow it, too. Since some mobo's wont allow 3.0 on a sandy bridge E without official support. I'm on the mobile right now, but when I get home I can repost some links here. Or you can find my recent "updating!" Thread in general hardware.
So no, you can't hack 3.0 out of a 1.0 board. Its something specific to sandy bridge E due to bad luck at launch.
So no, you can't hack 3.0 out of a 1.0 board. Its something specific to sandy bridge E due to bad luck at launch.
I see a constant, it's the Drivers The improvement I see in 314.07 drivers is decreased timing latency on pci-e 3.0, it may be due to the interaction of my particular platform with the GPU's I'm using. It may be in the pipeline delay which is the length of time for a packet to traverse an empty switch or a function of the switch’s design. I do not know, however I do see an improvement overall, Gaming and Benching. With pci-e 3.0, 3 way SLI really come alive, in some Games and Benches I'm seeing at least a 8% to a 10% increase!
The improvement I see in 314.07 drivers is decreased timing latency on pci-e 3.0, it may be due to the interaction of my particular platform with the GPU's I'm using. It may be in the pipeline delay which is the length of time for a packet to traverse an empty switch or a function of the switch’s design. I do not know, however I do see an improvement overall, Gaming and Benching. With pci-e 3.0, 3 way SLI really come alive, in some Games and Benches I'm seeing at least a 8% to a 10% increase!
