So, what this essentially means is that one can use a PCIe 3 x8 slot for any current GPU without worrying about bandwidth constraints, since offers the same bandwidth as a PCIe 2 x16 slot.
I'll be curious to see Thunderbolt 2 scaling, since even though there's not much difference between PCI E 2 and 3 here the difference between PCI and Thunderbolt is still huge.
Nope, Thunderbolt 2 is still 4 lanes of PCIe 2.0, the only difference between v1 and v2 of Thunderbolt is that v2 allows for channel-bonding at the logical level, so you can use those same 4 lanes to connect to a single device, when w/ v1 you were limited to 2 lanes (single channel) per device.
I'm not sure how useful the AVGs are in this case Ryan. Don't you think it would be more informative to give some type of FPS vs. time graph, to show the peaks and valleys in framerate? Or even some type of percentile measurement, giving how much time is spent above certain framerate thresholds. I think I'd be more interested to see if PCIe 2 v 3 results in a significant difference in minimum FPS mainly, because slight dips are going to get lost in a sea of frametimes in an average, but will be very noticeable IRL.
In other recent gfx threads Anandtech editors have said that accumulating and processing time per frame data is too time consuming to make it into their current review/benchmarking workflows.
Unfortunately, until the toolchain improves that sort of data appears likely to remain the province of more specialized review sites.
This is one of those cases where I'm just throwing what data I have on the wall. I didn't put much time into the subject since it wasn't possible to get PCIe 3 working on the third slot, so this was all that was collected. It wasn't originally intended for publication.
Not necessarily disagreeing with waiting for Ivy Bridge-E but last I checked the i7-4960X will be six cores..... best not to include them with ultra high spec 12 core xeon's that may or may not be unlockable and even if so will cost an obscene amount. ;)
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
21 Comments
Back to Article
JMC2000 - Friday, June 21, 2013 - link
That is interesting. Ryan, do you think that a triple 30" setup could show even a larger difference?dishayu - Friday, June 21, 2013 - link
So, what this essentially means is that one can use a PCIe 3 x8 slot for any current GPU without worrying about bandwidth constraints, since offers the same bandwidth as a PCIe 2 x16 slot.Granseth - Friday, June 21, 2013 - link
I would be interested in knowing if PCIe 2 or 3 affects frame times and min fps as well.fokka - Sunday, June 23, 2013 - link
i don't think so, if it's not even influencing fps in a significant way.tipoo - Friday, June 21, 2013 - link
I'll be curious to see Thunderbolt 2 scaling, since even though there's not much difference between PCI E 2 and 3 here the difference between PCI and Thunderbolt is still huge.sna1970 - Saturday, June 22, 2013 - link
Thunderbolt 2 is around 3-4 PCIe 3.0 lanesMartinT - Saturday, June 22, 2013 - link
Nope, Thunderbolt 2 is still 4 lanes of PCIe 2.0, the only difference between v1 and v2 of Thunderbolt is that v2 allows for channel-bonding at the logical level, so you can use those same 4 lanes to connect to a single device, when w/ v1 you were limited to 2 lanes (single channel) per device.SantaAna12 - Friday, June 21, 2013 - link
Thank you. Disappointing, but good to know.chizow - Friday, June 21, 2013 - link
I'm not sure how useful the AVGs are in this case Ryan. Don't you think it would be more informative to give some type of FPS vs. time graph, to show the peaks and valleys in framerate? Or even some type of percentile measurement, giving how much time is spent above certain framerate thresholds. I think I'd be more interested to see if PCIe 2 v 3 results in a significant difference in minimum FPS mainly, because slight dips are going to get lost in a sea of frametimes in an average, but will be very noticeable IRL.DanNeely - Friday, June 21, 2013 - link
In other recent gfx threads Anandtech editors have said that accumulating and processing time per frame data is too time consuming to make it into their current review/benchmarking workflows.Unfortunately, until the toolchain improves that sort of data appears likely to remain the province of more specialized review sites.
MartinT - Saturday, June 22, 2013 - link
I call BS. I fully expect frame times to make it into Anandtech reviews by Aug 1st.Ryan Smith - Saturday, June 22, 2013 - link
They're already in our reviews, starting with GTX 780: http://www.anandtech.com/show/6973/nvidia-geforce-...MartinT - Saturday, June 22, 2013 - link
Ah, thanks. Did you ever go back and update the 7990 review?Ryan Smith - Friday, June 21, 2013 - link
This is one of those cases where I'm just throwing what data I have on the wall. I didn't put much time into the subject since it wasn't possible to get PCIe 3 working on the third slot, so this was all that was collected. It wasn't originally intended for publication.ZeDestructor - Friday, June 21, 2013 - link
"It wasn't originally intended for publication." We appreciate it though. The trend is obvious here.Remarius - Saturday, June 22, 2013 - link
I assume a motherboard with a PLX PEX8747 PCIe 3.0 like the MSI Z87 XPOWER will bypass the Haswell lane limitations?sna1970 - Saturday, June 22, 2013 - link
dont bother with Haswell at all ..soon in sept. we will have Ivy-Bridge-E with 40 PCIe3.0 lanes , 4,6,8,10,12 cores options , quad channel memory at 1866 DDR3
who cares about Haswell ? it is only 3 months to wait.
Remarius - Saturday, June 22, 2013 - link
Not necessarily disagreeing with waiting for Ivy Bridge-E but last I checked the i7-4960X will be six cores..... best not to include them with ultra high spec 12 core xeon's that may or may not be unlockable and even if so will cost an obscene amount. ;)R3MF - Tuesday, July 2, 2013 - link
won't we still have the same X79 chipset?and don't the socket 2011 CPU's run the PCIe from the chipset, in which case we'll have the same slightly flakey v3.0 support that we have today?
sna1970 - Saturday, June 22, 2013 - link
@ Ryan Smithyou should try Xeons on a 2011 socket , they support 40 lanes PCIe 3.0 officially ...
to compare , just use a Xeon 4 cores at stock speed of the same Haswel/Ivy i7
JeBarr - Monday, June 24, 2013 - link
So now you all know why kingpin decided to take the record with two titans instead of three :D