Widescreen Gaming Forum

it's 16%. so really not very small to nonexistent.

Please read the following:

We have put forth a great effort to get to the bottom of the PCIe 2.0 versus PCIe 3.0 debate. We put a lot of time into testing performance and verifying that our data is accurate. Except for a couple of specific scenarios, most of the performance advantage had under PCIe 3.0 was well under 10%. This actually falls in-line with the kind of performance advantages one might expect using n Ivy Bridge CPU clock-for-clock compared to a Sandy Bridge CPU. The IPC can affect performance by as much as 4-7% in favor of Ivy Bridge easily. As you noticed, most of our data when we experienced an improvement on the Ivy Bridge system was in this range of improvements. There were a few specific cases of 11% in The Witcher 2 in one test, and 19% in Batman (for part of the game only) and 14% when we cranked up the settings to unplayable levels in Max Payne 3. For the most part, at the real-world highest playable settings we found playable, all performance advantages were under 10%.

With real-world gameplay performance advantages under 10% it doesn't change the actual gameplay experience. It in no way allows us to improve in-game quality settings nor does it give us any advantages over the PCIe 2.0 system. As we've stated previously in this evaluation, the technical performance advantages are "benchmarkable" but not relating to the gameplay experience.

It is also very clear from our testing that the NVIDIA GeForce GTX 680 receives an overall higher percentage of improvements with Ivy Bridge than the Radeon HD 7970 does. It is possible that similar to our past CPU frequency testing, NVIDIA GeForce GTX 680 GPUs are simply more sensitive to CPU clock speed and IPC, especially when you scale these upwards. We've done testing in the past that also shows NVIDIA GPUs are more sensitive to CPU clock speed than AMD GPUs are as you scale those up to dual and triple-GPUs. Therefore, we are not shocked to find that one brand might benefit with a technology more than another. It is an interesting result that we didn't expect when we started testing.

So do not fret if you are on a Sandy Bridge PCI Express 2.0 system, you aren't missing out on a bunch of performance compared to an Ivy Bridge PCI Express 3.0 system. Most of our readers will likely benefit from higher CPU overclocks on Sandy Bridge anyway if you are truly pushing the CPU clock and this alone will likely negate any "advantages" from PCIe 3.0 or Ivy Bridge IPC when it comes to real-world gaming scenarios. PCIe 3.0 is a great evolution, one day it may actually support a better gameplay experience compared to PCIe 2.0, but that day is not today.

So what they are saying is that the difference in performance between PCIe 2.0 versus 3.0 has to do with the difference in platform (sandy bridge vs ivy bridge). Take away the difference in performance ivy bridge produces and for my system its really is very small to nonexistent. Furthermore I am using an AMD gpu and compared to NVIDIA you see a lot less increase in performance.

_________________
Philips BDM4065UC(3840x2160) Acer Z35(2560x1080@200hz); 980 Ti Hybrid @stock ; 6700K 4.6ghz (1.35v)/D15; 16GB 3200mhz; Asus Maximus Ranger VIII; AX860; 1TB 960 EVO; 750GB 840 EVO; Teufel Concept D 500; Sennheiser HD6XX; Windows 10 (latest build)

Yes, because they say it is the difference between sandy and ivy bridge, the ONLY way to test if PCI-e 2.0 is really holding the card back is to use a PCI-e 2.0 interface and make it spit out more bandwidth.

HardOCP used 2 different motherboards with 2 different types of CPU. They could of just used a 2600K running at 40x100 and 38x105 on the Z77 board, that would make the PCI-e slot run at 100% and 105% bandwidth PCI-e 2.0 and the 3770K at 40x100 and 38x105 which would make it run at 100%/ 105% PCI-e. They could run a control test multiple times at the 100x40 settings to see how much IPC made a difference at said resolution, then see if PCI-e 2 or 3 benifit from more bandwidth seperately. It is potentially possible that say you get 7% more perfoamnce on IB setup, but you get 14% better when you also increase the bus to 105x38, compared to the 40x100 SB, then you must be able to say that the extra difference comes purely from the bus speed. For example

This is only to feed my curiosity, nothing more, don't feel obligated to do this, but all HardOCP proved was that with 2 different CPU's and boards which are completely different that you get different results, which is a bit like comparing apples to oranges...

I also personally think that differences of 4-7% are quite big when you are talking about the high end hardware actually, when you are spending £500 on a graphics card, you can potentially get another 7% performance just like that. Or look at it another way, that you can lose 7% performance by using the wrong hardware.

_________________
P8Z68-V Pro | 2600K | HR02 | HD5850 | 2x4GB Vengeance LP | 128GB M4 + 6TB | X-Fi > HD595 | AX850 | Tai Chi | PB278Q | G110 + Deathadder 2013
P8Z77-V | 3570K | Mugen 2 | HD5850 | 2x4GB Vengeance LP | 500GB | X-750 | Fractal R3 | U2212HM | G110 + G400
P8H77-I | G860 | 4650 | 2x2GB XMS | 320GB | CX500 | Prodigy | T22B350EW | MX518
DC3217IYE | 1x4GB Vengeance | 64GB M4 | TX-42VT20E

uhm.... Tried to set PCI-E Gen 1.0 in BIOS, Seems to be recognized by GPU-z. But still didn't change anything!

Next try is gonna be a paper around PCI-E Slot and see if that slows down the GPU ^^



Edit: Tried to go up from 100mhz BCLK.... minimal voltage increasement, froze at 103mhz.


Edit2 : You fkn kidding me ? Running on PCI-E 3.0x1, blocked manually via Tape! No real Performance loss!

Attached files

_________________
We gonna send it to outa space!

So what you are saying is that since PCI-E 1.0 won't slow down a PCI-E 3.0 card, running it in a PCI-E 2.0 system won't slow it down either.

_________________
Philips BDM4065UC(3840x2160) Acer Z35(2560x1080@200hz); 980 Ti Hybrid @stock ; 6700K 4.6ghz (1.35v)/D15; 16GB 3200mhz; Asus Maximus Ranger VIII; AX860; 1TB 960 EVO; 750GB 840 EVO; Teufel Concept D 500; Sennheiser HD6XX; Windows 10 (latest build)

So what you are saying is that since PCI-E 1.0 won't slow down a PCI-E 3.0 card, running it in a PCI-E 2.0 system won't slow it down either.

He tested using:



Ungine follows the same trend, but not to the same extent.

You can clearly see that by running the card in a PCI-e 3.0x1 slot, which is a scenario that may happen using CrossfireX, or SLI for example with multiple cards, that instead of the control 100% performance you would expect, you get 92.3% of the performance.

Again, going back to the PCI-e 1.1x16 you would get 97.5% of the performance which is within a margin for error, but still would be a reasonable overclock for free on any given card.

Assuming these results are accurate, this difference would almost be like downgrading the 7970 to a 7950 in many situations.

I don't think you realise, a lot of people go for the new i7 CPU's etc, and slap in 32GB or 64GB of RAM, and can justify the extra cost for the extra performance, but then most people tend to brush off the umportance of losing 7% of your FPS just because their card is in the wrong socket, or running at wrong speed.

Based on these 3 results, it is hard to say where the line comes in where bandwidth stops playing a roll, and he wasn't using surround resolutions either, so makes it even harder to see that.

I for one am not going to post anymore here now, you made it perfectly clear you are not going to run these benchmarks that would show the lack/ the increase in performance, I personally do not have the hardware to run them, so I guess we will never know.

_________________
P8Z68-V Pro | 2600K | HR02 | HD5850 | 2x4GB Vengeance LP | 128GB M4 + 6TB | X-Fi > HD595 | AX850 | Tai Chi | PB278Q | G110 + Deathadder 2013
P8Z77-V | 3570K | Mugen 2 | HD5850 | 2x4GB Vengeance LP | 500GB | X-750 | Fractal R3 | U2212HM | G110 + G400
P8H77-I | G860 | 4650 | 2x2GB XMS | 320GB | CX500 | Prodigy | T22B350EW | MX518
DC3217IYE | 1x4GB Vengeance | 64GB M4 | TX-42VT20E

I for one am not going to post anymore here now, you made it perfectly clear you are not going to run these benchmarks that would show the lack/ the increase in performance, I personally do not have the hardware to run them, so I guess we will never know.

In a couple of weeks I have some free time and might test your theory. Don't take this the wrong way, but you seem to be put off by people that have (strong) opinions different from yours.

_________________
Philips BDM4065UC(3840x2160) Acer Z35(2560x1080@200hz); 980 Ti Hybrid @stock ; 6700K 4.6ghz (1.35v)/D15; 16GB 3200mhz; Asus Maximus Ranger VIII; AX860; 1TB 960 EVO; 750GB 840 EVO; Teufel Concept D 500; Sennheiser HD6XX; Windows 10 (latest build)

Code:

PCI-e 1.1 16x (2.5GT/ lane (250MB)) = ~4GB/ sec                                         Crysis FPS: 59.15
   
   PCI-e 3.0 1x (8GT/ lane (<span style="color: #900;">1GB</span>)) = <span style="color: #900;">985 MB/s</span>                                                           55.98
   
   PCI-e 3.0 16x (8GT/ lane (<span style="color: #900;">1GB</span>)) = <span style="color: #900;">15754 MB/s </span>                                                       60.67

Ungine follows the same trend, but not to the same extent.

You can clearly see that by running the card in a PCI-e 3.0x1 slot, which is a scenario that may happen using CrossfireX, or SLI for example with multiple cards, that instead of the control 100% performance you would expect, you get 92.3% of the performance.

Normal CF/SLI configuration on a Socket 1155 would be 8x/8x, or triple Fire 8x/4x/4x Even quad fire would be 4x/4x/4x/4x though... not sure if there are boards that support this.
BTW there was this news about a Geforce GT610 with PCI-E 1.0 .... well... not like it would be possible to slow down that card xD even running PCI-E 1.0 1x xD

I for one am not going to post anymore here now, you made it perfectly clear you are not going to run these benchmarks that would show the lack/ the increase in performance, I personally do not have the hardware to run them, so I guess we will never know.

There you go, Surround in Crysis and Unigine....

Still trying to figure out how i might reach BCLK higher than 103mhz... my old board went up to 104.7 without Voltage increasement. But yesterday i nearly grilled my board by trying 1.7V on the VTTCPU lane ^^

Attached files

_________________
We gonna send it to outa space!

Double post, i know. but i don't want the Pictures to mix!


So, there we go! 39x103 and 40x100 ! Eyefinity Resolution.

As you can clearly see 103mhz BCLK is to much for my mainboard to handle! Even heavily overvolted it doesn't run stable! therefore you need to find some1 else who can test your theory, sorry.

Edit: and for comparission. Real PCI-E 3.0x16

Attached files

_________________
We gonna send it to outa space!

[quote]I for one am not going to post anymore here now, you made it perfectly clear you are not going to run these benchmarks that would show the lack/ the increase in performance, I personally do not have the hardware to run them, so I guess we will never know.

In a couple of weeks I have some free time and might test your theory. Don't take this the wrong way, but you seem to be put off by people that have (strong) opinions different from yours.

I know I said I wouldn't post again here, so sorry for that.

I don't like it when people make other people out to be wrong WITHOUT EVIDENCE.

http://hardforum.com/showpost.php?p=1038584646&postcount=238

This is the reason I wanted to see how the PCI-e bandwidth effected things. Although it is with a different setup to yours, I wanted to see if with a single card if it made any difference worth talking about. That was a all. I was so insistent because of how I had seen a 4 GPU setup behave.

Sorry if I came across as being an arsehole, but to basically refuse to run setups based on your "gut feeling" or whatever it was and then say I have strong opinions is a tad rich :)

_________________
P8Z68-V Pro | 2600K | HR02 | HD5850 | 2x4GB Vengeance LP | 128GB M4 + 6TB | X-Fi > HD595 | AX850 | Tai Chi | PB278Q | G110 + Deathadder 2013
P8Z77-V | 3570K | Mugen 2 | HD5850 | 2x4GB Vengeance LP | 500GB | X-750 | Fractal R3 | U2212HM | G110 + G400
P8H77-I | G860 | 4650 | 2x2GB XMS | 320GB | CX500 | Prodigy | T22B350EW | MX518
DC3217IYE | 1x4GB Vengeance | 64GB M4 | TX-42VT20E

I don't like it when people make other people out to be wrong WITHOUT EVIDENCE.

I didn't say you were wrong, I just said that I didn't see the point. Look at Haldi's findings. If it is that hard to run ivy/sandy bridge cpu's at higher than 100 bclk, than even if you were right, I would be running my cpu at stock or use the multiplier to overclock anyway. I understand that it is interesting to know, but for me personally it would not matter.

This is the reason I wanted to see how the PCI-e bandwidth effected things. Although it is with a different setup to yours, I wanted to see if with a single card if it made any difference worth talking about. That was a all. I was so insistent because of how I had seen a 4 GPU setup behave.

Than why didn't you post that link earlier? I would have been interested in reading it.

Sorry if I came across as being an arsehole, but to basically refuse to run setups based on your "gut feeling" or whatever it was and then say I have strong opinions is a tad rich :)

I agree, it would have been a tad rich, but again I never said that. In my comment the one with the strong opinions was me, not you. So what I said was that you don't like people that have strong opinions (e.g. me) that aren't based on scientific findings. On the other hand none of the tests we or any review site (e.g. Tech Report, Hardocp) does uses the scientific method.

_________________
Philips BDM4065UC(3840x2160) Acer Z35(2560x1080@200hz); 980 Ti Hybrid @stock ; 6700K 4.6ghz (1.35v)/D15; 16GB 3200mhz; Asus Maximus Ranger VIII; AX860; 1TB 960 EVO; 750GB 840 EVO; Teufel Concept D 500; Sennheiser HD6XX; Windows 10 (latest build)