|
Fauxtool posted:my understanding is that if your cores running the game are maxed but your gpu isnt then your cpu may be holding back the gpu in that specific game. If you are getting poor fps and your gpu load isnt very high then your cpu is probably the issue. Yeah, this is the caveat, because you can be CPU bottlenecked at 25% utilization. And it's not really a binary thing,T(frame) = T(GPU) + T(CPU) so making either the GPU or the CPU faster will always have some impact on framerate. What people perceive as a "bottleneck" is when (say) T(GPU) gets to be like 10 times as big as T(CPU) and even if you double CPU performance (cutting T(CPU) in half) it only makes like a 5% difference in overall frametime.
|
#
?
May 30, 2016 19:49
|
|
|
|
| # ? Apr 27, 2024 03:17 |
|
|
ijyt posted:Don't see how it's a big deal in the end Heatpipe without fluid is just a copper tube
|
|
#
?
May 30, 2016 19:54
|
|
|
The fluid is really important to making a heatpipe conduct heat like literal black magic. A copper pipe is orders of magnitude less effective.
|
|
#
?
May 30, 2016 19:59
|
|
|
Works for evga lol
|
|
#
?
May 30, 2016 20:04
|
|
|
Don Lapre posted:Works for evga lol
|
|
#
?
May 30, 2016 20:05
|
|
|
xthetenth posted:The fluid is really important to making a heatpipe conduct heat like literal black magic. A copper pipe is orders of magnitude less effective. Well I learned something today. Apparently it's not a very good overclocker either, but with the factory oc already in place is it really a concern?
|
|
#
?
May 30, 2016 20:06
|
|
|
Don Lapre posted:Works for evga lol Almost hurts to look at.
|
|
#
?
May 30, 2016 20:07
|
|
|
HalloKitty posted:Almost hurts to look at. It's almost like a How Too of bad design. 1) Where the pipes connect with the chip should be a single flat block to avoid pipe gap. 2) Try to put the chip in the middle of the plate. 3) Don't let your fluid escape. 4) Don't use too much TIM.
|
|
#
?
May 30, 2016 20:11
|
|
|
I don't understand why this is difficult after years of making these things
|
|
#
?
May 30, 2016 20:14
|
|
|
TBF the tim is fine. Its not spread because it spread out that far, its just a thin square. The problem with tim is usually too little, not too much.
|
|
#
?
May 30, 2016 20:14
|
|
Ak Gara posted:It's almost like a How Too of bad design. You want to use plenty of TIM with an exposed die. Also "too much" TIM makes little to no difference in temps, too little on the other hand is quite bad.
|
|
|
#
?
May 30, 2016 20:18
|
|
|
Josh Lyman posted:To be fair, the die appears to only span 2.5 heatpipes. Additionally, the space between the heatpipes on the Asus is much smaller than the EVGA (which also left a heatpipe hollow, so that pipe was worthless). So, a good deal of heat should be able to conduct to those heatpipes on the end of the Asus cooler. With such a small die, I'm not sure what else can be done. I guess you could layer a vapor chamber between the die and the heatpipes, or just use a vapor chamber with an open-fan cooler (some cards do that, right?) Edit: Huh, it just dawned on me that the GTX 1070 is a $379 MSRP card with 8 GB VRAM. With all the other stuff going on, that's easy to overlook. It's neat to have a midrange card with a more usable VRAM than the Xbone, PS4, and PS4.5. Maybe now we'll see some developers really go ham on textures. Zero VGS fucked around with this message at 20:32 on May 30, 2016 |
|
#
?
May 30, 2016 20:22
|
|
|
Zero VGS posted:Huh, it just dawned on me that the GTX 1070 is a $379 MSRP card with 8 GB VRAM. With all the other stuff going on, that's easy to overlook. To be fair, the 390 is around $300 and also has 8 gigs of VRAM. It would actually pretty awful if the new, high-end cards had any less.
|
|
#
?
May 30, 2016 21:02
|
|
|
repiv posted:https://twitter.com/AMD/status/735902495952777216 At this point I fully expect them to put some empty suit on stage who just stands there and waves for an hour. Maybe they'll actually have useful info... but I am worried that it's going to be a bunch of fluff and poo poo about how great Polaris is in theory, without actually providing any info. Right now AMD is letting themselves be drowned out thanks to Pascal hype, and they seem to be fine with it. That's either confidence or complete incompetence, but I don't know which. I just want Polaris info. 
|
|
#
?
May 30, 2016 21:16
|
|
|
Anything but incompetence earns you 2 dollars in the state lottery
|
|
#
?
May 30, 2016 21:24
|
|
|
ijyt posted:Well I learned something today. Apparently it's not a very good overclocker either, but with the factory oc already in place is it really a concern? Depends on how good the end user overclocks are. End users could take take a 980 Ti a pretty good way from most factory OCs. And yeah, basically the heatpipe isn't a copper pipe, but a pipe lined with a wicking material and then an empty core. The goal is the cold end is cold enough the liquid is liquid at that temperature and the hot end is hot enough it evaporates. So the hot end's a source of vapor pressure that pushes the vapor to the cold end, and dries out the wicking material on that end, which draws liquid from the cold end to it, creating a cycle by which the liquid circulates and evens out the temperatures really well.
|
|
#
?
May 30, 2016 21:25
|
|
|
Yea I dont really know why amd isn't letting people talk about the macau event. Must be bad
|
|
#
?
May 30, 2016 21:36
|
|
|
At this point I wonder how much of it is the aforementioned rumored yield issues, vs how much of it is AMD vastly overestimating others' expectations
|
|
#
?
May 30, 2016 21:42
|
|
|
I just hope Polaris turns out to be good, I want to see prices pushed down. Not that I have much faith in AMD at this point. 
|
|
|
#
?
May 30, 2016 21:44
|
|
|
Char posted:How do you establish if a CPU can bottleneck a GPU? I've got to upgrade, at least, my 6890, but I don't know what to do with my i5-760. CPU bottlenecking a GPU is game-dependent. If the game is a simulator or poorly-optimized, your CPU will always weight in heavily. Otherwise, not so much generally speaking. Both will always have some weight either way. Ideally most of the heavy lifting in a game should be done by the GPU.
|
|
#
?
May 30, 2016 22:19
|
|
|
I have a theory on why the Macau NDA date is the end of June. AMD's next fiscal quarter begins July, so if there *is* any damage by what they're revealing under NDA, it doesn't hurt the current quarter, and they have the entire next one devoted completely to Zen and its foibles.
|
|
#
?
May 30, 2016 22:19
|
|
|
SwissArmyDruid posted:I have a theory on why the Macau NDA date is the end of June. AMD's next fiscal quarter begins July, so if there *is* any damage by what they're revealing under NDA, it doesn't hurt the current quarter, and they have the entire next one devoted completely to Zen and its foibles. this is akin to rearranging deck chairs on the rms lusitania. nvidia is gonna launch a gtx 1070 right through amd's hull.
|
|
#
?
May 30, 2016 22:31
|
|
|
SwissArmyDruid posted:I have a theory on why the Macau NDA date is the end of June. AMD's next fiscal quarter begins July, so if there *is* any damage by what they're revealing under NDA, it doesn't hurt the current quarter, and they have the entire next one devoted completely to Zen and its foibles. Makes sense. This is going to be disappointing, isn't it?
|
|
#
?
May 30, 2016 22:36
|
|
|
Smithereens posted:this is akin to rearranging deck chairs on the rms lusitania. nvidia is gonna launch a gtx 1070 right through amd's hull. Hey, I never said it was a *happy* theory.
|
|
#
?
May 30, 2016 22:39
|
|
|
AMD's real reason for the July launch is because they know they have an ace up their sleeve and want the petty Nverds to get high on their $400-700 purchases before releasing a $300 card that stomps them both 
|
|
#
?
May 30, 2016 23:00
|
|
|
Smithereens posted:this is akin to rearranging deck chairs on the rms lusitania. nvidia is gonna launch a gtx 1070 right through amd's hull. Last I checked the 300+ and sub 300 dollar segments of the market were different parts but ha loving ha people are intent on a monopoly.
|
|
#
?
May 30, 2016 23:05
|
|
|
I do wish people would stop saying that the FTC cares in the least about nVidia's business practices - especially in the context of AMD folding or getting out of the enthusiast graphics market. The Steam Hardware Survey alone shows that AMD's already dead (at least in the enthusiast space), and it's not because of mediocre products or missteps, it's because the vast majority of people still view their computers as appliances, not ~speshul snowflakes~ arranged into a perfect snowball that runs games at 144FPS. The FTC would not give one solitary speck of poo poo if nVidia started charging $500 for an "x70" card, $750 for an "x80" card, and $1000 and $1500 for the Ti and Titan respectively. And why? Because what we're engaged in is - wait for it - a hobby. Our custom-built towers are fundamentally no different than that Uncle of yours who throws thousands of dollars a year into ~HIS BOAT~ or ricers buying mods for their WRXes. AIB cards are a *luxury item*, the same as anything that only carries intrinsic value. It's around this time usually one would bring up the classic 'chicken vs. steak' example from ECON 102. Consoles are and always will be the 'chicken' option, while enthusiast PC gaming will always be 'steak.' It's only when people's Dells and Macbooks start to fail because nVidia used substandard soldering that anyone at the government level gives a drat, and not really even then, because laptops with discrete graphics are a shockingly small portion of the market. AMD's possible/eventual/imminent passage will resound like a fart in the wind to everyone but us and stock insiders who short them as they die - OEM computers sold with discrete nVidia graphics will go up $200-500 from what they sell for now, and outside of our little clique, the only reaction will be 'wow, it seems like computers have gotten more expensive all of a sudden.' BIG HEADLINE fucked around with this message at 00:13 on May 31, 2016 |
|
#
?
May 30, 2016 23:19
|
|
|
Can we kill x86 while we're at it
|
|
#
?
May 30, 2016 23:38
|
|
|
xthetenth posted:Last I checked the 300+ and sub 300 dollar segments of the market were different parts but ha loving ha people are intent on a monopoly. honestly, i don't think that any (realistic) pc enthusiast believes amd is gonna come anywhere close to pascal. even though amd hasn't released any specs, you think they would've hinted at competition when pascal was released? it's been pretty quiet on their front and, for now, it seems clear the nvidia is way ahead and will have a dominant market share on the upper echelons of graphics cards going forward. fyi, i'm really not a team red/green person; i just go with the better tech. if amd actually made decent cpus or graphics cards i would definitely get them over intel/nvidia, but at this point amd looks like toast
|
|
#
?
May 30, 2016 23:54
|
|
|
We should buy and no matter what poo poo they release to protect them. Reward mediocrity
|
|
#
?
May 31, 2016 00:02
|
|
|
Don Lapre posted:We should buy and no matter what poo poo they release to protect them. Reward mediocrity Anime Schoolgirl fucked around with this message at 00:05 on May 31, 2016 |
|
#
?
May 31, 2016 00:03
|
|
|
Anime Schoolgirl posted:Like buying Xeons for HPC clusters "I'll have you know that Xeons are manufactured with strict guidelines in place for operations under extreme conditions and *furthermore*..." 
|
|
#
?
May 31, 2016 01:07
|
|
|
In this increasingly divergent timeline, Radeon is bought by Intel and becomes a successful HPC competitor to Nvidia, while AMD's Zen and sale of Radeon provides the cash injection necessary for AMD to move far into the black. Later on, they buy out PowerVR who struggles with the limited and shrinking revenue from Apple iPads. The widespread adoption of Vulkan and passaage of time results in ARM and POWER becoming strong second tier competitors to x86 µarchs. This change in fortunes results in IBM acquiring Nvidia sometime in 2027, and by 2029 there is a strong 4 way competition between Intel/Radeon, IBM/Nvidia, AMD/PowerVR and Qualcomm in HPC, server, desktop and mobile. Via and S3 remain irrelevant because they refuse to make chips out of silicon instead of woodpulp. No one cares. Thanks for reading my futuretech fanfiction full of boundless optimism.
|
|
#
?
May 31, 2016 01:29
|
|
|
Powervr is really making that Dreamcast money last
|
|
#
?
May 31, 2016 01:37
|
|
|
In the far future, Via and S3 are the ONLY ones still making chips out of silicon. Everyone else has moved onto InGaAs or whatever.
|
|
#
?
May 31, 2016 01:39
|
|
|
Don Lapre posted:We should buy and no matter what poo poo they release to protect them. Reward mediocrity So that year or so when a 290 or 290X was the best way to turn dollars into framerate? Agreed, did buy AMD then. I mean really when cards like the 780 and 780 Ti existed and were the 290 and 290X's competition...
|
|
#
?
May 31, 2016 02:06
|
|
|
Paul MaudDib posted:Yeah, this is the caveat, because you can be CPU bottlenecked at 25% utilization. And it's not really a binary thing,T(frame) = T(GPU) + T(CPU) so making either the GPU or the CPU faster will always have some impact on framerate. What people perceive as a "bottleneck" is when (say) T(GPU) gets to be like 10 times as big as T(CPU) and even if you double CPU performance (cutting T(CPU) in half) it only makes like a 5% difference in overall frametime. This isn't true in most cases -- Amdahl's Law applies to serial workloads, not parallel ones. The real equation is "T_frame = MAX(T_gpu + T_cpu) + T_serial", where T_serial should be 0 for a well-designed engine. Ideally, your GPU will be working on rendering a submitted workload at the same time your CPU is putting together a command list for the next batch of GPU work, or doing non-rendering stuff for the frame (Physics, Gameplay/AI, whatever). In an ideal case both sets of tasks are well-balanced to your hardware, so T_gpu and T_cpu are pretty close to one another. When we talk about "bottlenecks", what is happening is that one of the two components (usually rendering) has a much higher workload than the other; however, since we are processing a given frame through both units, the faster workload (lower T_x) gets held up by the slower one (higher T_x), leaving that unit idle. One easy way to identify this is by changing resolutions. Does your framerate improve going from 1440p to 1080p? Then you are GPU-bound: you can potentially improve your 1440p performance by up to that difference by improving your GPU. This isn't a perfect test (there are other types of GPU workloads that aren't affected by resolution) but it's a quick and easy first stop. Another easy test is to change the clock speeds on your GPU if you have an overclocking tool. This can let you dial in exactly how GPU-bound you are by seeing how much increasing (or decreasing) your clock speed affects framerate. Gonkish posted:At this point I fully expect them to put some empty suit on stage who just stands there and waves for an hour. Maybe they'll actually have useful info... but I am worried that it's going to be a bunch of fluff and poo poo about how great Polaris is in theory, without actually providing any info. Hey, they had a T-shirt cannon at the Capcasin event Hubis fucked around with this message at 02:34 on May 31, 2016 |
|
#
?
May 31, 2016 02:30
|
|
|
Fauxtool posted:i2500k doesnt bottleneck a well OCed 980ti and that is comparable to a 1070. Some games it does!
|
|
#
?
May 31, 2016 04:14
|
|
|
I have a chance to straight trade my R9-290 for a GTX 970, do I want to do this? When I bought it about 1.5 years ago, 970s were ~$330 and this 290 was $210. I'm assuming that this will be a performance upgrade, but you guys have been circlejerking how good Hawaii is the last couple pages and talking about AMD cards aging more gracefully in general. Also the whole "3.5GB" thing happened. At the end of the day, a 970 is still a faster GPU, right?
|
|
#
?
May 31, 2016 04:47
|
|
|
|
| # ? Apr 27, 2024 03:17 |
|
|
Twerk from Home posted:I have a chance to straight trade my R9-290 for a GTX 970, do I want to do this? When I bought it about 1.5 years ago, 970s were ~$330 and this 290 was $210. I'm assuming that this will be a performance upgrade, but you guys have been circlejerking how good Hawaii is the last couple pages and talking about AMD cards aging more gracefully in general. Also the whole "3.5GB" thing happened. At the end of the day, a 970 is still a faster GPU, right? By little enough it matters a lot what games you play, and whether one/both can overclock.
|
|
#
?
May 31, 2016 04:57
|
|