Puget Systems print logo

https://www.pugetsystems.com

Read this article at https://www.pugetsystems.com/guides/518
Article Thumbnail

Impact of PCI-E Speed on Gaming Performance

Written on November 13, 2013 by Matt Bach
Share:

Introduction

CPUs and chipsets only have a certain number of PCI-E lanes (or "paths" for data transfer), but with the number of features and options that manufacturers are cramming onto modern motherboards, there are times when there are simply not enough PCI-E lanes to run everything at full speed. We see this most often with the mainstream series of CPUs (currently Haswell) and chipsets (Z87, H87, etc.) where if you want to use multiple video cards (or install any other type of PCI-E card in the second x16 slot), the video card(s) will only run at x8 speeds rather than the full x16 speed you would otherwise get.

Normally, if you want to run a pair of video cards at full x16 speeds, you need an enthusiast motherboard and CPU (such as an Ivy Bridge-E CPU on a X79 motherboard). The problem with this is that NVIDIA video cards currently do not officially support PCI-E Gen3. You can run an .exe that NVIDIA has provided to gain Gen3 capabilities, but NVIDIA does not offer any guarantee of stability if you use that utility.

Many hardware sites (such as TechPowerUp and AnandTech) have shown in the past that most video cards do not show any performance decrease by running in x8 mode and cannot utilize the larger bandwidth provided by the latest Gen3 specification. However, video cards are getting faster and faster so we felt it was worth revisiting to find out if the fastest video cards available still do not have any performance advantage running at PCI-E x16 Gen3 versus PCI-E x8 Gen2. In addition, multiple GPU setups have not been throughly tested and with the gaining popularity of 4k displays, we felt it was important to see if the PCI-E revision/speed would affect a dual GPU setup at the much more demanding 4k resolution.

Test Setup

To see how the chipset and CPU affects our results, we used two separate testing platforms consisting of the following hardware:

We will be using three different benchmarks at two different resolutions: 1080p and 4k. At 1080p we will only be using a single GTX Titan, but at 4k we will be using a pair of GTX Titans in SLI. The settings for the benchmarks were chosen to give us an average 40-50 FPS.

To test both PCI-E 2.0 and PCI-E 3.0, we simply changed the PCI-E revision setting in the BIOS to either Gen2 or Gen3.

Finally, to switch between x16 and x8 modes a piece of insulating material (actually the sticky part of a post-it note) was used to cover half of the contacts on the cards which forces it to run in x8 mode. Note that we did not have to do this on the Z87 motherboard when we used two GTX Titans in SLI as that CPU/chipset can only run multiple video cards at dual x8 speeds, not dual x16.

1080p Results

Starting with Unigine Heaven 4.0, lets first take a closer look at the Z87 test system. On that system, PCI-E 3.0 was very slightly faster than PCI-E 2.0, but oddly we saw higher scores in x8 mode than in x16 mode. The biggest variance was still only 1 FPS, however, which we would call within our margin of error.

On the X79 system, there is really nothing to discuss. The biggest variance was only .4 FPS which is well within our margin of error.

Our second benchmark - Hitman: Absolution - has some very interesting results. This time, the Z87 test system showed pretty much no performance difference across any of our PCI-E combinations. On the other hand, the X79 system gives us very mixed results. In x16 mode, PCI-E 2.0 is faster than PCI-E 3.0, but only by .6 FPS. But in x8 mode, PCI-E 3.0 is 1.2 FPS faster than PCI-E 2.0. 1.2 FPS is still not very much, but it is noteworthy.

For our final benchmark - Metro: Last Light - our results are again unremarkable. On the Z87 test system, the results were all essentially identical. The X79 system was slightly faster when running PCI-E 3.0 in both x8 and x16 mode, but only by .3-.7 FPS.

4k Results

Running our benchmarks at 4k resolutions with two video cards should stress the PCI-E bus more than a single card at lower resolutions, but at least for Unigine Heaven there is no noteable difference in performance based on either the PCI-E revision or the number of PCI-E lanes.

Unlike Unigine Heaven 4.0, Hitman: Absolution does give us some interesting data to go over, at least for the X79 test system. On that system, at x16 speeds PCI-E 2.0 is actually 1.2 FPS faster than PCI-E 3.0. At x8 speeds, however, PCI-E 3.0 is actually much faster than PCI-E 2.0 by 1.5 FPS which is the biggest variance we saw in any of our tests.

Our final benchmark is actually the first one that gave us the results we originally expected. The Z87 system again shows no performance variance, but for the X79 system PCI-E 3.0 is either faster or the same as PCI-E 2.0 at both x8 and x16 speeds. At the same time, x16 outperforms x8 by as much as 1.5 FPS.

Conclusion

Our testing has pretty clearly shown that for gaming using either PCI-E 2.0 or PCI-E 3.0 will give you nearly identical performance. Oddly, in some benchmarks PCI-E 2.0 was actually faster than PCI-E 3.0. At the same time, x16 was not consistantly faster than x8. Again, x8 was actually faster than x16 in many cases. So unless you care about getting up to 1.5 FPS better performance, you might actually want to manually set your video cards to operate at x8 speeds - although we really would not recommend doing so.

This isn't to say that PCI-E 3.0 is not faster than PCI-E 2.0, or that x16 is the same as x8, but rather that current video cards and games are simply not able to utilize the additional bandwidth they provide. In fact, we recently showed that the performance of a Xeon Phi card is greatly reduced if you run it at x8 speeds in the blog post Performance of Xeon Phi on PCIe x8.

While we recommend using the latest PCI-E revision whenever possible, if your motherboard or video card only supports PCI-E 2.0 our results show that this really is not a problem. At the same time, if you want to install a sound card into your Z87 system but doing so would limit your video card to x8 speeds, that is also not a very big problem. At most you may see ~1.5 FPS drop in performance, but that change is so small that it is very unlikely to ever be noticeable.

Tags: PCI-E, PCIe, revision, 2.0, 3.0, Gen2, Gen3, x8, x16
Kenneth Crippen

There needs to be more of these articles. I appreciate the information Matt. Personally, using this ASUS Sabertooth 2.0 Mobo....I love having options. I'm going with PCI 2.0 whenever given the option. Who knows how long it will be before games can actually and fully utilize all the extra bandwidth these new cards are bringing to the table.

Posted on 2013-11-13 23:54:27

It's funny, but often these types of articles don't actually require that much time or effort on our part since we already have the hardware on hand, but its good to hear that people find them useful. Its a good reminder that the usefulness of an article doesn't have to be defined by how long it took to put it together. So thanks for letting me know that you liked it!

Posted on 2013-11-14 00:41:41
phiillyd

You should do a three card text on x79 vs z87/z77 (with a PLX chip, like the G1.Sniper 3).

Posted on 2013-11-14 01:21:57
Angel Stewart

Extremely useful. It is good to know what to spend on, and what can be avoided as a cost when building a system :-)

Perhaps by the time we can max out the bandwidth it will be time for an all new system anyways.

Posted on 2013-11-17 16:30:07
Gord Bestwick

I once saw an article which did this same thing but right down to 1x performance (16x,8x,4x,1x) And it was very interesting what happened around the 4x/1x mark. You should consider doing a follow up with this.

Where they started to show the cracks was with texture/bandwidth heavy uses. Most things barely noticed the change.

Posted on 2014-11-16 05:38:28
Jonathan

I have an ASUS Z-87K mobo. It has PCI E 3.0 x16 and can run the second GPU at PCI 2.0 x4 (no idea why it's not x8). Do you think I'll see performance decrease with dual Gigabyte R9 290's 4 GB running in crossfire? If so, how much? Is it worth it buying a new mobo with x8 x8. I have an i7 4770k 4.3 GHz also. When I bought PC I bought cheaper mobo and cheap ass GT 640 because it's the money I had at time. Now have single R9 290 and want to go Crossfire but just realized my second PCI is 2.0 x4.

Posted on 2015-03-01 18:46:00
Miroslav Lutsyk

Yes, agree with that is extremely useful. I have x99 Sabertooth + i7 5820k and 28 PCI-e lines that supports my CPU's, and now i don't worry about that if i'll do SLI i will lose some speed!

Posted on 2016-01-10 00:33:57
Raymond Ellis

Sense there is not much difference with 2.0 vs 3.0.

What about modern cards on PCI-E 1.0? (Providing you can get similar cpu/memory on older motherboard)

Posted on 2013-11-14 01:23:27

We originally was going to include PCI-E 1.1, but took it out since it added too much data which made it hard to interpret it easily. Since there really isn't any sort of equivalent CPU/motherboard available with PCI-E 1.1 since PCI-E 2.0 was launched over 6 years ago, we opted to keep it out of our results and instead focused on more modern systems. I would suggest taking a look at the TechPowerUp article we linked to in the introduction if you are interested in that since they included PCI-E 1.1

Posted on 2013-11-14 01:47:26
Raymond Ellis

I figured it had something to do with the age, and therefor lack of similar CPU.
I forgot about those links by the time I got to the bottom.. It was intrusting to see that there really isn't 'too' much of an improvement to 2.0.
Do you think we are just not using 2.0 to it's full potential yet?

Posted on 2013-11-14 09:00:18

Personally, I think it is just the nature of gaming. PCI-E generations and slot sizes are all about bandwidth: the amount of information that can pass back and forth between the system and the expansion card (in this case, a video card). With games, there is very little info going from the card to the system - the main output from the card is to the monitor, not back to the computer itself. The data coming in to the card is going to be raw information about where the player is looking and moving, which is pretty minimal in size, and then the graphics resources for the card to display. That definitely needs some bandwidth, but still not a ton... and so we see that games do okay with older generations of the PCI-E standard and even lower speed connections.

Now if you switch and look at other types of expansion cards, specifically co-processors like the Intel Xeon Phi, you can see more of an impact. It depend on exactly what you are doing there too, though: if you are simply running code that keeps it data mostly in the card's own memory, and just sends results back to the computer, that is not affected - but if you are moving data between the system and card for calculations then it can impact things. Dr Kinghorn, our resident HPC expert, did a little testing on that recently:

http://www.pugetsystems.com...

Posted on 2013-11-14 16:12:45
lehpron

I think that is the caveat many mistakenly attribute considering how many years these PCI Express specifications last in relation to CPU IPC differences, that we will bottleneck to CPU long before saturating the PCIe spec.

Just because the performance difference with Titan on PCIe 2.0 is negligible with 3.0 doesn't mean I can add Titan to a 7-year-old X38/780i system running a Core 2 Quad, even though there is a PCIe 2.0 controller and at least one x16 slot.

Considering the differences in IPC over the years, a Core 2 Quad would need 5.3GHz just to match stock i7-4770K Haswell, which wasn't practical then without LN2. On the other hand an X58 rig could probably support multiple Titans if their CPU overclock was 4.5GHz or so-- golden batch territory.

That TechPowerUps article that did include PCie 1.1 was flawed in the sense that the PCI Express spec was adjusted in the BIOS. Meaning the same Ivy Bridge quad was used, and thus the PCIe 1.1 results are not reflective of an older board with PCIe 1.1 and an older CPU t its highest possible overclock.

It showed the true technical impact of how a GTX680 works in PCIe 1.1, but again, not reflective of real-world of someone with an aging system being cheap and only changing the graphics card-- not a scenario worth advocating except for academics.

Posted on 2014-03-21 01:59:20
Andrew Fox

I have an i7 2600k so I am limited to pci2 even though I'm using a Z77 board. I had an offer of a free 3770k but passed on it because real world benefit is at most 5%. I agree that pci 1 data is useless but pci 2 data is great as it re-enforced my choice that there's nothing wrong with pci2 for a video card.

Posted on 2015-03-04 18:47:44
Robodang

Are you using a PCE3 card on your system? I'm running a i7-2700k on a Z68 board and am looking to upgrade to a newer gpu. Wondering if a R9 380 will work on my setup. Thanks!

Posted on 2015-10-06 16:22:58
Jason Copas

I have a system that has a 1.0 slot and a 3.0 card will not work in it, ive heard 1.0 and 2.0 are compatiable and 2.0 and 3.0 are also but a 1.0 slot will not run a 3.0 card. I use to think it would til i got 1 like this.

Posted on 2017-08-14 06:48:23
cri8tive08

boy I would like to have that

Posted on 2013-11-14 02:35:33
Neil Welch

Awesome test, guys! Good practical information. Definitely a must share.

Posted on 2013-11-14 03:08:13
Imminent

The marketing behind PCIE 3.0 looks good on paper but has translated into minimal gaming results for just about every test. However, I wonder if the PCIE 3.0 performance is any faster with the same cards when video transcoding. Sure, PCIE 3.0 might favor enterprise configurations but it should also give regular consumers something extra. Right?

Posted on 2013-11-14 06:30:39

It all depends on how much data needs to move between the computer and the video card. I'm not sure what that looks like when using a card for GPU acceleration of video editing & transcoding, but it is something we could probably test - I'll ask Matt and see what he thinks.

Posted on 2013-11-14 16:14:10
Imminent

I would be interested to see where PCIE 3.0 benefits most. I know what I've seen here but I often wonder would it be considered worth the money to pro transcoders to spend more on higher end cards. That said, very rarely would spending another $300 be worth almost 20 frames more per second from card to card just to play games.

Posted on 2013-11-19 22:51:13
Dave4321

Would 3.0 make a difference on a future card that took advantage of it? Or would that not even matter?

Posted on 2013-11-14 15:22:29

Other articles (like the TechPowerUp we linked to in the Intro) found that they started getting a performance hit around PCI-E 2.0 x4, which operates at about 2GB/s. Considering that PCI-E 2.0 x16 equates to 8GB/sec of bandwidth, and PCIE-3.0 x16 equated to 16GB/sec, I think it will be at least a couple more years before PCI-E 3.0 becomes a big deal for gaming.

Posted on 2013-11-20 20:34:06
Malcolm Galloway

Touche'.... (Clappin' hands)

Posted on 2013-11-18 21:42:09
Sybreed

My motherboard is using PCI 2.0 and I recently bought an NVIDIA 780GTX TI... everyone I knew told me I was practically buying this for nothing and that I would not benefit from my new graphic card at all. I'm glad to know it isn't entirely true and that I won't have to build an entirely new computer from scratch! It's especially crazy considering I got my computer merely 2 years ago.

Posted on 2013-12-03 22:30:04
sam

Are u facing any problem with this card now ?

Posted on 2014-11-26 07:19:28
nate call

i'm sure not and i'm on a dell xps 8300 with a i7-2600 which does not support 3x. i have a 780 hawks by msi its so fast. but unity on release day was the first time it didn't totally kick a.. in ultimate way it was still high performance and great looking but ran kinda choppy with vsync.. anyways have not played since but i can vouch i never noticed a diffrence in other machines with same card.. anyone know where i can buy a
Z77 Extreme11 ? in the states? I have rar amd fatality pro that u can't get in the states... trade?

Posted on 2014-11-27 10:11:05
Andrew Fox

2600 is an insane CPU and I had a 670 on mine but this year upgraded to a GTX 970 and it almost doubled my performance and I play at 2440p. There's no problem with pci2 16x for a single card setup. The speed decrease only hits if you go tri SLI and even then it's around 4% loss at worst.

Posted on 2015-03-04 18:52:12
Sofia Lucifairy

Very interesting post!

Posted on 2014-03-28 20:08:32
lowrizzle

Just curious, but have you considered that the PCH does not contain a discreet PCIe 3.0, 2.0, 1.1 and 1.0 stack? Why would a firmware engineer include all three when each downlinks as backwards compatible?

In other words, how did you validate that you were actually on a 3.0 versus a 2.0 link? By setting the BIOS option? I think your nearly invariable results are actually the result of an incomplete testing methodology.

Posted on 2014-04-28 20:09:53
Robin Kleven

MY question is, when does the bottle necks appear. How much longer will 8x keep up?

Posted on 2014-06-23 21:06:22
gadgety

This is a great piece. I'd like to see you revisit it annually, because, as you say GPUs are getting faster and faster. There are dual cards as well, R9 295x2 from AMD, and similar solutions from Nvidia.

Posted on 2014-08-16 07:51:58
AquaVixen

I do wish you would of tested to see what the effect x4 does on things. Some motherboards that run both gpu's at dual-8x will drop 1 or more cards down to 4x if you add other pci-e devices. That would also be interesting data.

Posted on 2014-08-22 15:16:54

We didn't go down to x4 because usually when you get to that point you are running triple SLI/Crossfire and the scaling from that is going to be much more significant than the loss of bandwidth with PCI-E x4. The one article I know that has tested x4 (and even x2) is from AnandTech (http://www.anandtech.com/sh.... It only covers PCI-E 3.0, but you can see that x4 does see a performance loss compared to x8 or x16.

Posted on 2014-08-25 17:19:44
AquaVixen

One thing I have been curious on. Some platforms for LGA-2011 on certain motherboards (That I won't mention) claim to be able to run 4 x PCI-E 16x video cards at all 4 16x at the same time, (Which would be 64 PCI-E Lanes!), even though the CPU's specifications only state 40 lanes. Do you know of any article that has tested such a configuration (4 gpu's on those 4-16x boards) verses a "normal" motherboard where all 4 would run at 8x? I've been considering upgrading to newer systems and not sure if it's worth the price premium for a 4-16x board.. if it would be any faster for a 4-way GPU setup or just go with some of the cheaper 4-8x boards. I've tried googling but not finding much in the way of information on this.

Posted on 2014-08-30 16:03:40
Papizach Akbar

agree with the statement gpu would scale onto cpu, I run 2 gtx580 in asus Rampage4 formula for final render, save me seconds and even minutes when I replaced i7 3820 to i7 4930K. Im planning to add titan black and beefier psu for animation works. Any1 here knows link that shows how it goes in 3d apps rather than gaming..

Posted on 2014-09-02 13:36:12

Dude, that's a really handy article, thanks! I was wondering if the i7 5820K only having 28 PCI channels was going to be an issue if I ever wanted a second 780 GPU (the 5820 can only support two GPUs since it runs 8,8,8,4 when you have the two cards in). I believe your work has answered that question! I don't think possibly dropping a frame here or there is worth splashing out the extra £300. Really appreciate it, nice one. :)

Posted on 2014-09-02 23:50:15
Edward Casey

I am very pleased to have read this test data. I am thinking about getting an Asrock Z97 Extreme6 mother board, paired up with an Samsung XP941 M.2 x4 card, but putting it in that slot drops the x16 slot for the video card to x8. I do think that combination will give me very fast boot times using Windows 8.1. I am relieved that there will be no throttling of the performance of the graphics card for all intents and purposes running in x8.

Posted on 2014-09-10 21:55:26
Marcio

Thank you so much for this information. i have a Z98 motherbord 2600k. altho the G1 Sniper supports pci 3.0 the 2600k cpu doen't. and i'm looking at maybe going for a gtx 970 sli configuration. This information was is very useful. thank you once again.

Posted on 2014-09-24 20:45:36
Marcio

sorry Z68 motherbord.

Posted on 2014-09-24 20:50:14
jim greene

So it is almost a year after this article was written.
PCIe 3.0 is more common on mobos.
h97 and z97 chipset is out. (True, not much change from 87 series)
New, more powerful video cards are out (ex. r9 290 and gtx 900 series).
Does a higher end gaming machine push the limits yet of what the mobo buses (expansion slots) can handle?

At the level (Serenity) I am looking at, I see a mobo (SABERTOOTH Z97 MARK 2) with
2 x PCIe 3.0/2.0 x16 (Single at x16, dual at x8/x8) 1 x PCIe 2.0 x16 (max at x4 mode, black)
3 x PCIe x1.

I have one GPU, one HHD and one SSD to add to the machine.

I guess you put the gtx 970 in one of the PCIe 3.0 x16. Where does the 7200 SATA III HDD and the SSD attach?

There is lots of talk about how the PCIe widths available to individual GPUs are cut, but not too much talk about when other devices are added to the mobo. In my above example, there are three devices to attach, only on is a GPU.

How will you put these three devices on the mobo? What will be the resultant theoretical throughput for each (x16, x8, x4, x1)? Do any realize an actual bottleneck?

If the above does not push limits, what if I choose 2 top GPUs and a HDD and a SSD?
(that would be getting a Serenity now and adding a like GPU later.)

Posted on 2014-10-30 02:31:06

For gaming, no - video cards still don't need more than PCI-E x8 connectivity for full game performance. Select other applications with video cards, like GPGPU computing, can use the full bandwidth of a x16 slot though.

The vast majority of drives attach through a completely separate interface - SATA, rather than PCI-Express. Motherboards have 6 or more SATA ports, for connecting SSDs, HDDs, and even optical drives (CD / DVD / Blu-ray).

Now there are a select few SSDs that use PCI-Express rather than SATA, in order to reach higher speeds. We have a couple of articles on those, if you are interested:

http://www.pugetsystems.com...

http://www.pugetsystems.com...

If you really want, you can get two GPUs (each in a PCI-E x16 slot but running at x8) and a SSD that uses a PCI-Express x4 slot (nothing higher is needed) and a HDD on SATA and fit that all in the Serenity :)

Posted on 2014-10-30 03:58:53
Andrew

Is there somewhere to look up how much bandwidth a GPU puts through PCI-E? I've only found memory bus bandwidth listed in specs for GPUs, i don't think memory bandwidth has to do with whats going though PCI-E.

Thanks for any answers!

Posted on 2014-10-30 21:47:38

There isn't a specific attribute of a video card that tells how much bandwidth it can / will use on the PCI-E bus... aside from the generation of PCI-E it uses (1.0, 2.0, 3.0, etc). The amount that is actually used depends on the application and situation. It is just that games currently don't need enough bandwidth between the video card and the rest of the system to fully saturate the standard x16 connection (even at PCI-E 2.0). Some other applications that use the video card probably can, even on the same sort of video card... it all depends on your needs.

Posted on 2014-10-31 05:09:55
robt

Interesting; In a similar vein, I thought in the case of having a PCI-E 2.0 motherboard for my 2.0 x16 card would show a big improvement over running the 2.0 x16 card in a PCI-E 1.1 motherboard. I was surprised to see no difference and wondered if the settings were bad or I had power supply problems. Thanks for the insight.
I am assuming the relative difference in 2.0 to 1.1 is the same as 3.0 to 2.0.

Posted on 2014-11-05 21:30:20
kanuj

Very interseting article hats off to pugetsystem now i am great fan :)

kanuj
tech enthusiast
Candytech.in

Posted on 2014-11-29 07:31:24
Eddie

It is certainly an
interesting hypothesis to test, but I'm not quite sure how they
expected some last generation game-engine powered games running on
systems with their total video memory(dedicated + physical VRAM) to
be in excess of 9GB (on the non-SLI titan configuration) to provide
any type of bandwidth issue for these systems; whether in Full HD or
4K.

I would have thought
that better test conditions would come from using the Unreal Engine 4
SDK rendering at extreme settings and using high-end video cards just
above the Steam average (something like a GTX 660), but certainly not
using cards like Titans that could store an entire game section in
VRAM, as the test really needs to have a workload that generates
plenty of contention for VRAM on a second by second basis, so that
the PCIe bus is thrashed at each speed and could illuminate results
inline with the conditions the average steam user will face because
their setups will face high contention for available VRAM when
gaming, with these games, and the coming games, as once again, the
games will be designed around the strengths of the games consoles,
and this generation the consoles' strengths are HSA and high memory
bandwidth.

Posted on 2014-12-20 01:59:39
00blahblahblah00

It would have been nice to throw in PCIe 2.0 x4 in there since many mainstream Z-series and H-series boards have one PCIe 3.0 x16 and the 2nd PCIe x16 slot is 2.0 @ x4 for whatever reason. These boards are not SLI compatible but many of them are Crossfire compatible. It would be interesting to see how an AMD GPU like the R9 290/290x would do in 2-way Crossfire on a board with a 2.0 @ x4 slot. Obviously, this is going to be a bottleneck, the question is how much? If it isn't that much worse than the same GPUs on PCIe 2.0 x8, x8 then getting one of these cheaper, non-SLI motherboards could make a pretty sweet 'budget' multi-GPU rig. Also, this info would be beneficial to people who bought a motherboard like this and only had one GPU for a while but now are considering a 2nd.

Posted on 2015-02-21 16:02:16
koenrad

Thank you, I appreciate this. I found this because, unfortunately my GTX 970 does not fit in the main PCIe slot on my motherboard, and I am forced to put it in the second, x8 one.

Posted on 2015-05-29 20:20:59
Jose

At what point does a PCIe 2.0 motherboard hold back a PCIe 3.0 GPU? Wouldn't newer games also be an issue?

I have:

AMD Phenom II X4 955

MSI 790FX-GD70 (2009)
http://www.newegg.com/Produ...

Evga 760 2g Super Clocked

Posted on 2015-06-13 19:01:14

For gaming, at least today, using a PCIe 2.0 slot does not impact game performance at all. However, a system that is old enough to have PCIe 2.0 instead of 3.0 may have other aspects of the computer that would inhibit performance. For example, that CPU is starting to show its age when it comes to gaming: a newer Intel Core i5 or i7 would be up to twice as fast.

Posted on 2015-06-14 06:09:19
Jose

Thanks for your reply. I'm hoping to get the new Z170 motherboard as soon as it comes out for Skylake. I think the PCIe 2.0 mobo vs PCIe 3.0 GPU issue will be more of an issue when the next generation of GPU's and games come out with more 4k support and HBM memory.

Posted on 2015-06-14 15:32:28
cat1092

While I don't feel that PCIe 2.0 will hold back performance, some CPU's may. Just my opinion, if possible, would upgrade that CPU to a newer & more powerful model, and check your PSU to ensure it'll handle both. My main two self built PC's has the EVGA Supernova G2 650W installed, and another has a Seasonic M12 Edition Bronze 550W, which am working on now. All are Full Modular, which I prefer.

Good Luck!

Cat

Posted on 2016-09-26 10:02:32
GuccizBud

Interesting that the PCIe 3.0 x8 configuration prevailed over PCIe 3.0 x16 on the whole, more or less proving that it's the cards themselves (and/or the software) that aren't using the additional bandwidth. Those tests were 1–2 years ago though, would be interesting to see if/how that situation has changed.

Posted on 2015-06-18 19:52:43
Anders tn

Any chance you will repeat this test with the new GPU's from AMD? Especially that dual Fiji card.

Posted on 2015-06-18 21:15:29
anandcg

I've always wondered if there is any GFX card for the PCI-E slot. My motherboard is old so I've always done a lot of research. It looks like the power issue will always be there. Anyways, thanks for the information dude.

Annu
www.techteria.com

Posted on 2015-07-25 19:33:44

Thank you - Your article cleared up a big concern I had in building on my MSI X99 Gaming 7 board with an Intel i7-5820k, which only has 28 PCIe lanes enabled (or 12 lanes crippled). I was going to move up to a higher-end 40-lane CPU, then read your article. I found you out of frustration due to the addition of PCIe x1 cards to the system changing the video card from x16 to x8.

Would switching to a 40-lane i7 allow the video card to run in x16 mode? Does this afect GPU rendering for video editing apps? I am using this setup as a video capture and edit workstation.

Posted on 2015-08-11 14:41:37
cat1092

Switching to a 40 lane CPU has a lot of benefits, including the ability of running several GPU's, as well as having more than one NVMe SSD. Some MB's has these along with the SATA connectors (such as Intel), these SSD's are identical to the 2.5" versions, only a little larger. Plus can run another on both the MB with M.2 slot & probably x4 slot also, as long as it's PCIe 3rd gen (2nd gen will run, just slower in reads).

As far as your question goes, you can run a single GPU on a normal quad core (28 lane) CPU, it's all of the extras that you can add that makes the 40 lanes a great setup, even if the GHz level is a bit lower than the 4.0GHz of popular quads. With the unlocked CPU, you'll easily be able to go past 4.0GHz, as long as there's adequate cooling (Noctua has high quality air coolers for these sockets). Liquid cooling is also an option, though will never be installed in any of my computers, all it takes is a leakage & there goes a $2,000+ investment down the drain.

Noctua has premium coolers that'll keep the CPU running at safe temps.

Cat

Posted on 2017-02-28 23:36:55
Steve Smith

Hi Matt have you performed any of these comparison with a z97 board and would conecting a pci-e ssd like sm951 affect once a gtx960 is connect to port 1? reason for asking is that i want to buy a Asrock extreme 6 and that comes with a ultra m.2 socket and i already have a samsung sm951 card. thanks

Posted on 2015-09-18 02:29:44

I don't think you should see any real difference in performance for either device. I couldn't find in the manual exactly what happens if you populate the PCI-E 3.0 M.2 slot, but I would assume that it reduces the primary PCI-E 3.0 x16 slot to x8 and the secondary PCI-E 3.0 slot to x4 (with the M.2 running at x4). A GTX 960 won't be able to saturate anything near a PCI-E 3.0 x8 slot and the SM951 is only a PCI-E 3.0 x4 capable card so you should see pretty much full performance on both cards.

One thing I will suggest you read is out SM951 review: https://www.pugetsystems.co... specifically about how hot that drive runs. I would highly recommend not using a M.2 slot if you can avoid it and instead use a PCI-E to M.2 adapter that includes a heatsink. If you can't do that, I would get some small heatsinks (GPU VRAM heatsinks work pretty good if you cut them down to size) to stick onto the controller chip to help keep it cool.

Posted on 2015-09-18 18:49:27
cat1092

Can this be done w/out removing the Samsung label, which may void the warranty? I purchased 3 small copper heatsinks for the native MB install, though chose to go with PCIe 3.0 & using an Addonics adaptor instead, basically the same as Lycom that Amazon sells, these looks so identical that could had been manufactured at the same facility.

I believe that the NVMe SSD runs cooler on these alone, or at least on mine, considered installing a small fan that fits in the PCI slots next to it, which in powered by a Molex plug. On modern computers, it's highly likely that a SATA to Molex adaptor will be needed, though shopping on eBay, these can be found for as little as $5. The fan will cost more, though if can be pointed to blow directly on the SSD, will be worth it, some has adjustable speeds to meet whatever application is needed, usually a component attached to a PCIe slot, such as a PCIe GPU that has passive cooling & running rather hot.

If it'll cool a GPU, then certainly should a NVMe SSD.

Cat

Posted on 2017-02-28 23:25:09
Georg Zechozkyy

i have an ZH77-G43 mainboard and i am thinking of probably adding a second r9 290 to crossfire it, but the second pcie slot is electrical x4 would this be ok, or pointless? I dont realy think of upgrading cpu + mainboard + ram, so would the crossfire would be worth?

Posted on 2016-01-06 10:50:09

Yes, it would be worth it. If you haven't already done so, you should.

Posted on 2016-05-09 04:17:48
John Myers

I want to thank you guys for doing all these articles and studies... and sharing them in such an accessible format! I wish more companies would share this type of well thought out testing.

Posted on 2016-01-16 20:56:23
Peepfishes2

Great article. THANK YOU!

Posted on 2016-01-19 18:08:18
Van Stamelen Tot Vlottemond

Nice!!! Just the article that I was looking for. Presented data about the tests just fills in ALL the blanks I had about what GPU to add to my Z97 Extreme6 mobo (due to the sharing of lanes with M.2 PCIe3 x4). My budget was limited anyway, but this way I know any high end card of last few years won't be bottlenecked by the limit of PCIe3 x8. Cheers!!!

Posted on 2016-03-20 15:30:10
FirestoneX .

Does anyone know if in 2016 when the next Nvidia graphics cards are released (presumably the gtx 1080) if pci 2.0 will start bottle necking that card? I I have a i7 2600k and my motherboard only has pci 2.0 and I'm not yet ready to upgrade to a new pc for at least 2-3 more years, though want to get the newest future graphics card for VR gaming. Will pci 2.0 be a bottleneck then (now if you are reading this in the future)

Posted on 2016-03-26 13:28:30

No, PCIe 2.0 will not be a bottleneck to any newer GPU for at least a few more years. As indicated by the article, most are seeing speed differences between 3.0 and 2.0 of <4%. Heck, most GPUs cant even take full advantage of the 2.0 bandwidth now. Just make sure the card is compatible with 2.0 as there is a good chance that, soon, cards will require 3.0 regardless if they can utilize the bandwidth capacity.

Posted on 2016-05-09 04:16:00
Jesse Albright

this is just like the damn multi core support architecture for video games currently. they are designing hardware that our software and pipeline architecture cannot even utilize yet. most games today only utilize 2-4 cores MAX this is just another example of building for a future that we are slowly crawling towards. in other words save your money.

Posted on 2016-03-29 10:25:29
ZMe Ul

can this be retested with a GTX1080? I suspect the impact will be different this time around

Posted on 2016-07-24 12:05:28

It might be interesting to re-test with the new 1000-series at some point, but I don't think it is so much the video card that impacts this. It seems instead to be the game, or game engine, as there are a couple of isolated cases where recent games showed a performance difference even on previous-generation cards. They are the edge cases, though, rather than the norm:

https://www.techpowerup.com...

Posted on 2016-07-25 06:18:54
ZMe Ul

me again, maybe now's the time since we have a new GTX Titan X in the wild
yay

Posted on 2016-08-26 17:05:24

We have done new testing no this with the Titan X: https://www.pugetsystems.co...

Posted on 2016-10-15 20:23:51
ZMe Ul

awesome, reading it right now

Posted on 2016-10-16 06:03:30
Jay Jardin

Thanks. You are pretty much alone answering this question. Is there a chance we can get the 4k tests done with two GTX 1080's? I don't want to continue to buy 40 lane processors that bring nothing to the gaming table. p.s. I hear the HB bridge also brings and fps difference on those cards.

Posted on 2016-09-15 23:55:02

We have a newer article about this now as well: https://www.pugetsystems.co...

Posted on 2016-10-15 20:23:14
cat1092

What about if a PCIe SSD, such as the Samsung 950 Pro M.2, is placed in the 2nd graphics slot?

Sure, I could had placed it in the x4 slot, and writes were on par with Samsung specs. Yet reads were 1,600MB/sec, rather than it's advertised 2,500MB/sec, placing it in the 2nd GPU lane cured it, while at the time, showed no issues with a MSI branded 2GB GTX 960. Now that I have an EVGA GTX 1070 FTW edition, maybe I should reconsider. First off, Speccy & other specs shows it as having 4GB VRAM, while only GPU-Z reports the correct 8GB of Micron memory.

What give here, Speccy or what? I can accept the 1,600MB/sec reads from the PCIe SSD & still be 3x faster than most SATA-3 SSD's. Yet having 4 out of 8GB memory is a different story. If I wanted a 4GB card, would had purchased one for much less. The latest NVIDIA 64 bit driver for Windows 8.1 x64 is installed, plus the DDU uninstaller was ran in Safe Mode to purge remnants of the GTX 960.

The fact that GPU-Z is reporting the correct amount of memory gives me hope, and it shows PCIe 3.0 x8, because of the PCIe SSD being in the 2nd slot.

I feel that these reporting apps should get their act together & do the job they're supposed to, or exit the market. The card is running fine, User Benchmark shows the GPU at 140% (UFO category), as well as the Desktop, only Workstation showed less at 95% & this may be due to an improperly tuned CPU. The CPU was the only component that showed to be running at less of it's potential. Oddly, though it has a 4.4Ghz Turbo Boost, have never reached it, except for a mild overclock which I had to get out of & fast. Need a better cooler, will be getting a Noctua one as funds permit, and no not the $60 one, rather the $90 model. Watercooling is out for me, have too much in the system to risk damage from leaks & cannot afford to replace.

http://www.userbenchmark.co...

Cat

Posted on 2016-09-23 07:40:42

With PCI-E SSDs, you really do need to have them installed into the correct slot in order to get full speed. In the case of the Samsung 950 Pro, that should be a PCI-E 3.0 x4 slot. If you were only getting 1,600 MB/s, my guess is that it was either installed in a PCI-E 2.0 x4 slot or a PCI-E 3.0 x2 slot (both would end up giving similar performance with that drive).

As for your GPU only showing 4GB, as far as I know the GTX 1070 is only manufactured with 8GB of VRAM. So if an application is only reporting 4GB either that application isn't reporting correctly or you have a defective card. We use GPU-Z pretty extensively here, however, and I've never seen it report an incorrect amount of VRAM so I would believe GPU-Z over Speccy. You could also confirm in the NVIDIA Control Panel by clicking on the "System Information" button in the bottom-left and seeing what is reported under "Dedicated video memory"

Posted on 2016-09-23 19:56:21
cat1092

The card is good, have since installed Linux Mint Cinnamon 18 & the full 8GB shows on that OS.

There's a catch to the full performance when in a x4 slot, even if nothing else is installed on the lane. The two features are reported by Magician as below, though both required a x16 slot (or x8) to enable these. Here's what the Magician snapshot showed before placing in the x16 slot.

Link Speed Cur:4Gbps, Max:8Gbps
Link Width Cur: x4, Max: x4
Bandwidth Cur: 16Gbps, Max 32Gbps

So something was up for Magician to be reporting this while in the x4 slot, now it shows all as max. Best yet, my EVGA 1070 hasn't taken a hit in performance, Speccy has either read the specs wrong, or Windows itself is, GPU-Z is calling it right, so is Linux Mint 18.

Being that I'm not on that OS at the moment, can't access it, though can provide a snapshot of what I have on Linux Mint Cinnamon x64, running in UEFI mode.

[img]http://i.imgur.com/jb0s1qs.png[/img]

Will also provide the one from the Samsung 950 Pro M.2 while in x4 slot.

[img]http://i.imgur.com/vCrbEZ6.png[/img]

Had I known the 960 Pro was coming out, would had waited a month, major performance increases!

Cat

Posted on 2016-09-26 09:46:05
cat1092

Matt, sorry about my misinformation, I was assuming the the x4 port was a 3.0 one, it's not. After re-reading the MB specs on the back of the package, it states that there's 2 x PCIe 3.0/2.0 x16 slot (single at x16 more or dual at x8 mode).

Then, it says '1 x PCIe 2.0 x16 slot (max. at x4 mode.....) & you likely know the rest. So my Samsung 950 Pro wasn't running at full speed on reads because it couldn't in that x4 slot. I do find it odd though, that the writes hit the max & didn't improve in the second x16 lane.

Sadly, I may have to move it back over because it may be what's causing my Linux Mint boot text & splash screen to be enlarged.

Being that I've sent an ASRock Z97 Extreme6 for RMA, may just use it in the Ultra M.2 socket on that MB, which is one of the few that fully supports M.2 3rd gen SSD's. I have some heatsinks that can be used to cool it down, though I just don't like the idea of installing it on a MB. Had a mSATA SSD on my XPS 8700 that ran hot as fire, though all other components around the drive were well within range (upper 20's low 30's in C). The SSD was hitting close to 70C & throttling. That was the Crucial M550, Samsung 840/850 mSATA owners has reported the same.

Oddly, when I reised a Topic, Crucial moderators said that this is 'normal'. For a SSD to be running at over 2x the temps of the CPU, GPU, MB, other connected drives, I felt robbed, to put it mildly. Just wanted to make use of a spare port. To make it run cool, not only required placing in a SYBA 2.5" adapter, also placed it in a Rosewill 3.5" to 2.5" adapter with a small fan, that move brought the temps in line with the rest of the drives & probably added years to it's lifespan.

So I'd not expect a M.2 SSD to be different, at least away from the MB holds heat down some.

At any rate, sorry I jumped to conclusions, should had been suspicious myself when seeing the Samsung snapshot above. Had the write speeds also been low, I sure would had checked, though normally it's the reverse (with SATA-3 SSD's), the reads will be faster than writes, so I was happy until I seen that pic above, and had I not been paying close attention, would had missed that, normally with Samsung there's a bolder warning with in print (or Red) if not on the 'right' port for optimal speed.

On the other hand, I can place it back in the x4 slot & still enjoy 3x the performance of a SATA-3 SSD! As long as I don't connect anything else, well there goes the PCIe wireless/BT upgrade on this PC, if I keep the SSD on here.

Thanks for your input that led to the discovery of the issue.;-)

Now if only I can monitor temps...........

Cat

Posted on 2016-09-27 05:29:29
PhilsComputerlab

Isn't PCI-E x16 Gen3 versus PCI-E x8 Gen2 a difference of a factor of 4?

Posted on 2016-10-05 21:45:53

Yes. Compared to x16 Gen3, both x8 Gen3 and x16 Gen2 should provide half as much bandwidth - and x8 Gen2 should provide a quarter (half of a half).

Posted on 2016-10-05 21:49:43
PhilsComputerlab

Cheers.

Posted on 2016-10-05 23:40:15
DMM

I can see this is a rather old article but seriously thank you for it. Looking to upgrade and I was worried my current Sabertooth just wasn't going to work. Literally saved me several hundred helping me see I could continue rocking this motherboard and CPU.

Posted on 2016-10-15 15:44:49

We have a newer article on this topic now: https://www.pugetsystems.co...

Posted on 2016-10-15 20:24:38
Woody Manly

Hello all, for Senior and PC Expert.
I need some suggestion here for choosing 1150 socket motherboard to build for medium budgetgaming and planned to use dual AMD GPU / Crossfire feature. The thing which really confusing me is about the speed / bandwith on the PCIe 3.0 x 16 mode (main socket) and the others PCIe 2.0 x 4 mode.

My motherboard to choose which on my list is:

01.Gigabyte G1 Sniper B6
02.Gigabyte GA H97M-D3H
03.MSI H87-G41 PC Mate
04.ASUS B85 Pro Gamer
05.ASRock Z87M Extreme4

Honestly I really like the features and the nice looking G1 Sniper B6.
Now I already had HIS Radeon HD 7850 and if I decided which motherboard already I planned to get another HD 7850 (pair in Crossfire mode)

Please I need a suggestion to choose from my list (or other recommendation if still on the same price range/level). And thank you so much for reading my comment.

*Sorry for bad English language :)

Posted on 2016-10-16 07:59:32
alex seaton

Its pretty much within margin of error and or 1 fps so Great to know.

Posted on 2016-10-21 23:06:35
Jianju

Still a useful discussion 4 years on.

Posted on 2017-07-21 13:39:02
www.tricksclu.com

Good post https://www.tricksclu.com/

Posted on 2017-08-05 21:27:05
Chris Cambridge

Thanks for a really great article..

Would it be possible in 2018 to perhaps revisit this article and do another, with Pascal GPUs (etc) including not only gaming but also some situations such as in data processing using GPUs where a smaller data bus (width/gen) does actually have a critical impact?

Folding@home which is protein folding works at maximum on x8 or x16 gen3, but if you go lower than this you take a massive hit on performance (eg data processed). On BOINC which is similar to F@H, I have yet to see any project which cannot achieve pretty much the same results even running x1 Gen3 (using PCIE risers, as we run via a GPU rack using these).

I wanted to ask you to ask you a question if that is okay? On GPU-Z you have a sensor showing the GPU' Bus Interface load - am I right in thinking that if this ever hit 100% then that would clearly show a bottleneck with the PCIE slot width/speed/gen that you are using? Or is the only way to really tell to do what you have done and actually run a benchmark looking at the given results?

Posted on 2017-12-29 13:26:05
Pagad Ultimo

Thanks, very helpful!

Posted on 2018-01-21 14:01:29
Cimmay Rivard

Wow, that's amazing. Thanks, I had to do a double take.

Posted on 2018-03-05 16:54:55
Ryan Barber

I want to know the pin counts you used to cover up the card with, like in your example showing the card with the sticky note attached to it. I want to block part of my card to force it to run in gen3 x8 mode instead of x16 mode. I don't have any option in my bios to adjust this so If you could please tell me how many pins you covered up and if you covered both sides to get your x8 mode that would be awesome.

Posted on 2018-04-21 03:55:50

Most motherboards have at least one slot that is physically x16, but only has actual electrical connections for x8. What we typically do is stick the card into that slot, then mark with a sharpie on the pcb of the GPU which pins we should cover with the sticky note. Basically, any of the pins that don't have an electrical connection get covered. Easier this way than trying to count out the pins.

Posted on 2018-04-21 03:58:59