{"video_id":"rSfifE2Domo","title":"Radeon HD 6990 Bandwidth Comparison Test 16x vs 8x vs 4x 3DMark 11 Linus Tech Tips","channel":"Linus Tech Tips","show":"Linus Tech Tips","published_at":"2011-05-08T14:53:29Z","duration_s":308,"segments":[{"start_s":0.24,"end_s":8.559,"text":"So, as a follow-up to my recent video where I took two GTX 580s, threw them on","speaker":null,"is_sponsor":0},{"start_s":5.92,"end_s":15.2,"text":"the Gigabyte G1 Gorilla motherboard, and I tested in dual 16x mode in SLI and","speaker":null,"is_sponsor":0},{"start_s":12.08,"end_s":17.92,"text":"dual 8x bandwidth mode in SLI. I am","speaker":null,"is_sponsor":0},{"start_s":15.2,"end_s":23.119,"text":"going to be taking an even higher an even more demanding card, the Radeon","speaker":null,"is_sponsor":0},{"start_s":21.0,"end_s":30.08,"text":"HD6990, and I'm going to be running it in a single card configuration in a 16x","speaker":null,"is_sponsor":0},{"start_s":25.84,"end_s":32.64,"text":"slot, an 8X slot, and finally a 4X slot","speaker":null,"is_sponsor":0},{"start_s":30.08,"end_s":37.36,"text":"to find out what difference there is. In terms of performance, I'll be using the","speaker":null,"is_sponsor":0},{"start_s":34.0,"end_s":39.76,"text":"3D Mark 11 extreme preset. Now, please","speaker":null,"is_sponsor":0},{"start_s":37.36,"end_s":45.92,"text":"note that the extreme preset will not show as much of a difference as one of","speaker":null,"is_sponsor":0},{"start_s":42.399,"end_s":48.879,"text":"the lower quality presets, uh, such as","speaker":null,"is_sponsor":0},{"start_s":45.92,"end_s":53.6,"text":"the entry or performance presets. But what it does give us is a more realistic","speaker":null,"is_sponsor":0},{"start_s":51.44,"end_s":61.28,"text":"real world scenario because most people are not going to be using a 6990 to run","speaker":null,"is_sponsor":0},{"start_s":57.12,"end_s":62.559,"text":"Quake 3 at 1024 x 768. So, you're going","speaker":null,"is_sponsor":0},{"start_s":61.28,"end_s":67.92,"text":"to be turning up the details, you're going to be turning up the res. So, that is the scenario that we will be looking","speaker":null,"is_sponsor":0},{"start_s":65.84,"end_s":73.119,"text":"at. Something to bear in mind before I get too far into my testing here is that","speaker":null,"is_sponsor":0},{"start_s":70.4,"end_s":78.159,"text":"the top slot at 16 and the second slot at 8X are both running off the PCI","speaker":null,"is_sponsor":0},{"start_s":75.68,"end_s":82.64,"text":"Express controller integrated into the Sandy Bridge CPU on this board which is","speaker":null,"is_sponsor":0},{"start_s":80.56,"end_s":87.6,"text":"clocked at 4.7 GHz by the way to eliminate any CPU bottlenecking or as","speaker":null,"is_sponsor":0},{"start_s":84.96,"end_s":93.64,"text":"much as possible. The bottom slot actually runs off of the chipset at 4x","speaker":null,"is_sponsor":0},{"start_s":91.119,"end_s":98.0,"text":"speed. So, it has an additional disadvantage that is not shared by the","speaker":null,"is_sponsor":0},{"start_s":96.32,"end_s":101.6,"text":"top two slots. So, just bear that in mind. Take the results with a bit of a","speaker":null,"is_sponsor":0},{"start_s":99.36,"end_s":107.36,"text":"grain of salt, but I don't have another platform handy where I can show you 16x,","speaker":null,"is_sponsor":0},{"start_s":103.759,"end_s":111.88,"text":"8x, and 4x where they will all be equal.","speaker":null,"is_sponsor":0},{"start_s":107.36,"end_s":115.04,"text":"So, my 16x result is a score of","speaker":null,"is_sponsor":0},{"start_s":111.88,"end_s":117.68,"text":"3,330. So, I'll be back with 8x and 4x","speaker":null,"is_sponsor":0},{"start_s":115.04,"end_s":122.399,"text":"results shortly. Now, this may look like a stupid configuration. I've got my 6T90","speaker":null,"is_sponsor":0},{"start_s":120.159,"end_s":126.32,"text":"and a GTX 580 here. But the reason I've installed the GTX 580, which won't","speaker":null,"is_sponsor":0},{"start_s":124.32,"end_s":132.52,"text":"affect my benchmark scores in any way, is to force these PCIe lanes to go from","speaker":null,"is_sponsor":0},{"start_s":128.959,"end_s":136.48,"text":"16x to 8x8x. So, you can see here within","speaker":null,"is_sponsor":0},{"start_s":132.52,"end_s":139.76,"text":"GPUZ, the 6990 reports as running at","speaker":null,"is_sponsor":0},{"start_s":136.48,"end_s":141.52,"text":"16x. And while that may be true because","speaker":null,"is_sponsor":0},{"start_s":139.76,"end_s":146.959,"text":"it's running internal Crossfire where each GPU is getting 16x bandwidth from","speaker":null,"is_sponsor":0},{"start_s":144.48,"end_s":152.16,"text":"the uh from the lane spplitter that's on the PCB","speaker":null,"is_sponsor":0},{"start_s":149.239,"end_s":157.28,"text":"itself. The actual lane splitter itself is only","speaker":null,"is_sponsor":0},{"start_s":153.56,"end_s":159.04,"text":"getting if we look here at the GTX 580.","speaker":null,"is_sponsor":0},{"start_s":157.28,"end_s":164.72,"text":"So you can see that the actual bandwidth to the slot is only 8x because it is","speaker":null,"is_sponsor":0},{"start_s":162.08,"end_s":172.04,"text":"correctly split at this time. So my benchmark results are","speaker":null,"is_sponsor":0},{"start_s":168.519,"end_s":176.08,"text":"3,312 3D marks. So we have seen","speaker":null,"is_sponsor":0},{"start_s":172.04,"end_s":179.2,"text":"a difference in performance between 16x","speaker":null,"is_sponsor":0},{"start_s":176.08,"end_s":179.2,"text":"and 8x","speaker":null,"is_sponsor":0},{"start_s":179.64,"end_s":188.84,"text":"312. Okay, very small, very small. And I","speaker":null,"is_sponsor":0},{"start_s":183.28,"end_s":191.2,"text":"will be back with PCIe 4X results very","speaker":null,"is_sponsor":0},{"start_s":188.84,"end_s":197.2,"text":"shortly. So at the end of the day, this video becomes more about how if you","speaker":null,"is_sponsor":0},{"start_s":193.68,"end_s":200.239,"text":"internally bottleneck a card um","speaker":null,"is_sponsor":0},{"start_s":197.2,"end_s":202.64,"text":"according to the computational resources","speaker":null,"is_sponsor":0},{"start_s":200.239,"end_s":208.159,"text":"located on the card that the amount of PCI Express bandwidth you supply it with","speaker":null,"is_sponsor":0},{"start_s":205.2,"end_s":212.799,"text":"is actually going to make very little difference in terms of performance. Now,","speaker":null,"is_sponsor":0},{"start_s":210.239,"end_s":217.2,"text":"something to bear in mind too about this particular testing setup today. You can","speaker":null,"is_sponsor":0},{"start_s":214.64,"end_s":220.799,"text":"see I had to nudge my motherboard over a little bit to get the uh 6T90 to fit","speaker":null,"is_sponsor":0},{"start_s":219.2,"end_s":227.44,"text":"into that bottom slot on my tech station. But so, I've got it in the PCIe","speaker":null,"is_sponsor":0},{"start_s":223.959,"end_s":230.08,"text":"4x slot. Now, think about it this way.","speaker":null,"is_sponsor":0},{"start_s":227.44,"end_s":240.4,"text":"Basically, I've got two GPUs. I have on a 6990, I pretty much have two 6970","speaker":null,"is_sponsor":0},{"start_s":234.08,"end_s":240.4,"text":"GPUs. Okay? So, so bear with me here.","speaker":null,"is_sponsor":0},{"start_s":240.76,"end_s":248.04,"text":"So, by the comparison I did the other day where I was comparing","speaker":null,"is_sponsor":0},{"start_s":245.799,"end_s":253.439,"text":"16x6 and 8x8x SLI, this is basically running it","speaker":null,"is_sponsor":0},{"start_s":251.599,"end_s":259.4,"text":"on this motherboard, this card in 16X mode is pretty much like running two","speaker":null,"is_sponsor":0},{"start_s":255.84,"end_s":264.52,"text":"cards, two high-end single GPUs at","speaker":null,"is_sponsor":0},{"start_s":259.4,"end_s":268.56,"text":"8x8x, 4x4x in the 8x slot, and then at","speaker":null,"is_sponsor":0},{"start_s":264.52,"end_s":269.84,"text":"2x 2x in that 4x slot at at the very","speaker":null,"is_sponsor":0},{"start_s":268.56,"end_s":276.0,"text":"bottom that that's essentially the amount of bandwidth we have available","speaker":null,"is_sponsor":0},{"start_s":271.68,"end_s":277.52,"text":"per GPU and we still see very very","speaker":null,"is_sponsor":0},{"start_s":276.0,"end_s":282.24,"text":"little fluctuation in terms of the overall score. So, what I may have to do","speaker":null,"is_sponsor":0},{"start_s":279.04,"end_s":284.96,"text":"in the coming uh next little while is do","speaker":null,"is_sponsor":0},{"start_s":282.24,"end_s":291.84,"text":"an updated video about this where I find a slightly less uh GPU speed","speaker":null,"is_sponsor":0},{"start_s":289.28,"end_s":295.84,"text":"bottlenecked benchmark and uh demonstrate maybe more of a more of a","speaker":null,"is_sponsor":0},{"start_s":294.0,"end_s":300.0,"text":"typical or worst case scenario for how much the PCI Express bandwidth affects","speaker":null,"is_sponsor":0},{"start_s":298.0,"end_s":302.639,"text":"single card performance. So, thank you for checking out this little episode of","speaker":null,"is_sponsor":0},{"start_s":301.36,"end_s":307.88,"text":"Linus Tech Tips. Don't forget to subscribe for more unboxings, reviews, and other computer videos.","speaker":null,"is_sponsor":0}],"full_text":"So, as a follow-up to my recent video where I took two GTX 580s, threw them on the Gigabyte G1 Gorilla motherboard, and I tested in dual 16x mode in SLI and dual 8x bandwidth mode in SLI. I am going to be taking an even higher an even more demanding card, the Radeon HD6990, and I'm going to be running it in a single card configuration in a 16x slot, an 8X slot, and finally a 4X slot to find out what difference there is. In terms of performance, I'll be using the 3D Mark 11 extreme preset. Now, please note that the extreme preset will not show as much of a difference as one of the lower quality presets, uh, such as the entry or performance presets. But what it does give us is a more realistic real world scenario because most people are not going to be using a 6990 to run Quake 3 at 1024 x 768. So, you're going to be turning up the details, you're going to be turning up the res. So, that is the scenario that we will be looking at. Something to bear in mind before I get too far into my testing here is that the top slot at 16 and the second slot at 8X are both running off the PCI Express controller integrated into the Sandy Bridge CPU on this board which is clocked at 4.7 GHz by the way to eliminate any CPU bottlenecking or as much as possible. The bottom slot actually runs off of the chipset at 4x speed. So, it has an additional disadvantage that is not shared by the top two slots. So, just bear that in mind. Take the results with a bit of a grain of salt, but I don't have another platform handy where I can show you 16x, 8x, and 4x where they will all be equal. So, my 16x result is a score of 3,330. So, I'll be back with 8x and 4x results shortly. Now, this may look like a stupid configuration. I've got my 6T90 and a GTX 580 here. But the reason I've installed the GTX 580, which won't affect my benchmark scores in any way, is to force these PCIe lanes to go from 16x to 8x8x. So, you can see here within GPUZ, the 6990 reports as running at 16x. And while that may be true because it's running internal Crossfire where each GPU is getting 16x bandwidth from the uh from the lane spplitter that's on the PCB itself. The actual lane splitter itself is only getting if we look here at the GTX 580. So you can see that the actual bandwidth to the slot is only 8x because it is correctly split at this time. So my benchmark results are 3,312 3D marks. So we have seen a difference in performance between 16x and 8x 312. Okay, very small, very small. And I will be back with PCIe 4X results very shortly. So at the end of the day, this video becomes more about how if you internally bottleneck a card um according to the computational resources located on the card that the amount of PCI Express bandwidth you supply it with is actually going to make very little difference in terms of performance. Now, something to bear in mind too about this particular testing setup today. You can see I had to nudge my motherboard over a little bit to get the uh 6T90 to fit into that bottom slot on my tech station. But so, I've got it in the PCIe 4x slot. Now, think about it this way. Basically, I've got two GPUs. I have on a 6990, I pretty much have two 6970 GPUs. Okay? So, so bear with me here. So, by the comparison I did the other day where I was comparing 16x6 and 8x8x SLI, this is basically running it on this motherboard, this card in 16X mode is pretty much like running two cards, two high-end single GPUs at 8x8x, 4x4x in the 8x slot, and then at 2x 2x in that 4x slot at at the very bottom that that's essentially the amount of bandwidth we have available per GPU and we still see very very little fluctuation in terms of the overall score. So, what I may have to do in the coming uh next little while is do an updated video about this where I find a slightly less uh GPU speed bottlenecked benchmark and uh demonstrate maybe more of a more of a typical or worst case scenario for how much the PCI Express bandwidth affects single card performance. So, thank you for checking out this little episode of Linus Tech Tips. Don't forget to subscribe for more unboxings, reviews, and other computer videos."}