Overclock.net › Forums › Industry News › Hardware News › [ FORBES ] Crunching The Numbers: Can Dual $269 Nvidia GTX 760s Outperform A $999 Titan?
New Posts  All Forums:Forum Nav:

[ FORBES ] Crunching The Numbers: Can Dual $269 Nvidia GTX 760s Outperform A $999 Titan? - Page 14

post #131 of 234
Quote:
Originally Posted by fateswarm View Post

That's the core of the problem in your reasoning: You want FPS to include all. It can't and it won't no matter what those subjective graphs try.

That's why you have more than one measurements. Measure first the FPS and then measure the Artifacts of a certain kind.

Are you unable to absorb two different numbers? Are you forced to compress them to an unreadable subjective number?

That's why when you are told the Weather you are told a temperature and a wind direction, not temperaturewindirection.

So the problem is that you don't understand what those percentile graphs are telling us, not the graphs themselves.

Frame rate and frame time go hand in hand. They are just different ways to say the same thing. What you want is a steady frame rate, that is to say a frame rate that has a flat and linear percentile graph. You can demonstrate frame time issues perfectly well with frame rate numbers, just not average fps numbers. What we have in those percentile graphs by Guru3D is the frame rate information from all the frames, whereas an average fps number only tells the average frame rate.

That is why a frame rate percentile graph (based on data gathered by FCAT, not FRAPS) is a perfectly objective measure. Average fps is not, because it filters the data to only tell a part of the whole story. What PCPer is doing by rating the frames "good" or "bad" (a.k.a "runt" as they like to call them) is also a much more subjective thing than the percentile graphs by Guru3D. PCPer judges the data and filters it according to their judgement, Guru3D shows us the raw data and lets us do the judgement. I'm not saying that PCPer's judgement is flawed, I'm just saying that there is no such level of subjectivity in those Guru3D's percentile graphs.
The Air Tunnel
(10 items)
 
  
CPUMotherboardGraphicsRAM
i7-4820K (4500MHz@1.28V) P9X79 Sapphire R9 290 Tri-X New Edition Team Group Vulcan 4x4GB 2133MHz 
Hard DriveCoolingOSPower
2x240GB SSD @RAID 0 Noctua NH-D14 SE2011 Windows 10 Pro Corsair AX750 
CaseAudio
Rosewill Armor EVO hiFace+AudioGD+DT770Pro 
  hide details  
Reply
The Air Tunnel
(10 items)
 
  
CPUMotherboardGraphicsRAM
i7-4820K (4500MHz@1.28V) P9X79 Sapphire R9 290 Tri-X New Edition Team Group Vulcan 4x4GB 2133MHz 
Hard DriveCoolingOSPower
2x240GB SSD @RAID 0 Noctua NH-D14 SE2011 Windows 10 Pro Corsair AX750 
CaseAudio
Rosewill Armor EVO hiFace+AudioGD+DT770Pro 
  hide details  
Reply
post #132 of 234
Quote:
Originally Posted by thestache View Post

Unfortunatly this conversation has been had a million times. It's the same old thing. People without surround or experience claiming certain amounts of VRAM are enough (and linking stupid reviews that are wrong) and then the surround crowd that knows what they are talking about telling it like it is.

You can hit the hard limit with a 2GB card at 1080P in Crysis 3 maxed. Not to mention several games hit the hard limit in 1080P Surround with 2GB cards and without AA. 2GB is not enough for surround without AA. We came to this conclusion shortly after the GTX 690 released.

That's it, not saying anymore. If people are uneducated on this topic, go find the abundance of information on OCN about it. We have several experienced and educated users on the topic.

There are also many very well informed people saying that GK104 doesn't have the shader power needed to run settings that might require more than 2GB. IMO, for single chip they have a point. SLI is cutting it close, there are some cases where more VRAM would give a little more wiggle room for playable settings but only when you have tri-SLI GK104 will you really be bottlenecked by 2GB.
The Air Tunnel
(10 items)
 
  
CPUMotherboardGraphicsRAM
i7-4820K (4500MHz@1.28V) P9X79 Sapphire R9 290 Tri-X New Edition Team Group Vulcan 4x4GB 2133MHz 
Hard DriveCoolingOSPower
2x240GB SSD @RAID 0 Noctua NH-D14 SE2011 Windows 10 Pro Corsair AX750 
CaseAudio
Rosewill Armor EVO hiFace+AudioGD+DT770Pro 
  hide details  
Reply
The Air Tunnel
(10 items)
 
  
CPUMotherboardGraphicsRAM
i7-4820K (4500MHz@1.28V) P9X79 Sapphire R9 290 Tri-X New Edition Team Group Vulcan 4x4GB 2133MHz 
Hard DriveCoolingOSPower
2x240GB SSD @RAID 0 Noctua NH-D14 SE2011 Windows 10 Pro Corsair AX750 
CaseAudio
Rosewill Armor EVO hiFace+AudioGD+DT770Pro 
  hide details  
Reply
post #133 of 234
Quote:
Originally Posted by thestache View Post

Unfortunatly this conversation has been had a million times. It's the same old thing. People without surround or experience claiming certain amounts of VRAM are enough (and linking stupid reviews that are wrong) and then the surround crowd that knows what they are talking about telling it like it is.

You can hit the hard limit with a 2GB card at 1080P in Crysis 3 maxed. Not to mention several games hit the hard limit in 1080P Surround with 2GB cards and without AA. 2GB is not enough for surround without AA. We came to this conclusion shortly after the GTX 690 released.

That's it, not saying anymore. If people are uneducated on this topic, go find the abundance of information on OCN about it. We have several experienced and educated users on the topic.
Don't be unfair though. I mean a ton of people put an explicit disclaimer: "For most games". i.e. Yes, it will bottlneck on certain games, and it may even bottleneck on most games on insanely high settings but they aren't fools when they say for most users, it won't be horrible deficiency. I mean, people have budget limits. We could say, ye get 6GB Titans and run everything on the highest settings but you can't really have that. So balancing their budget, it's not a horrible decision to suggest, 2GB, ok even 3GB that is 'enough for most purposes'.
PC
(9 items)
 
  
CPUMotherboardGraphicsRAM
4790K 4.6G 1.24v/1.74v, 5.1G 1.35v validation. GA-Z97X-Gaming 7 Tri-X R9 290 1100/1350 +0.012v G.Skill 2400 c10 
CoolingMonitorPowerCase
Noctua NH-D15 24EA53 IPS 76Hz OC EVGA 1000 G2 Phanteks Enthoo Pro 
Audio
SoundMagic E10 
  hide details  
Reply
PC
(9 items)
 
  
CPUMotherboardGraphicsRAM
4790K 4.6G 1.24v/1.74v, 5.1G 1.35v validation. GA-Z97X-Gaming 7 Tri-X R9 290 1100/1350 +0.012v G.Skill 2400 c10 
CoolingMonitorPowerCase
Noctua NH-D15 24EA53 IPS 76Hz OC EVGA 1000 G2 Phanteks Enthoo Pro 
Audio
SoundMagic E10 
  hide details  
Reply
post #134 of 234
Since when is this news? I thought this was common enthusiast knowledge
Enkidu
(20 items)
 
  
CPUMotherboardGraphicsGraphics
Intel Core i5 3570K @ 4.7GHZ Stable MAXIMUS V FORMULA GTX 780 GTX 780 
RAMHard DriveOptical DriveCooling
Corsair Vengeance 16GB Red 2133MHZ Kingston Hyper X 3K 120GB SSD Lite On 24X DVD/CD Drive XSPC Raystorm Waterblock with Red LEDs 
CoolingCoolingCoolingCooling
Swiftech MCP655-B Pump  Black Ice 240mm Long 42mm Thick Radiator  Corsair Air Series SP120 High Performance Editi... Primochill Red Tubing 
OSMonitorKeyboardPower
Windows 7 Professional 64bit Overlord X270OC 2560x1440p Monitor Overclocked ... Razer Blackwidow Ultimate OCZ Z Series Gold OCZZ1000M 1000W Power Supply 
CaseMouseMouse PadAudio
Thermaltake Level 10 GT Black Razer Imperator 2012 Expert Ergonomic Gaming Mouse Steelseries 4HD Random 5.1 Speaker System  
  hide details  
Reply
Enkidu
(20 items)
 
  
CPUMotherboardGraphicsGraphics
Intel Core i5 3570K @ 4.7GHZ Stable MAXIMUS V FORMULA GTX 780 GTX 780 
RAMHard DriveOptical DriveCooling
Corsair Vengeance 16GB Red 2133MHZ Kingston Hyper X 3K 120GB SSD Lite On 24X DVD/CD Drive XSPC Raystorm Waterblock with Red LEDs 
CoolingCoolingCoolingCooling
Swiftech MCP655-B Pump  Black Ice 240mm Long 42mm Thick Radiator  Corsair Air Series SP120 High Performance Editi... Primochill Red Tubing 
OSMonitorKeyboardPower
Windows 7 Professional 64bit Overlord X270OC 2560x1440p Monitor Overclocked ... Razer Blackwidow Ultimate OCZ Z Series Gold OCZZ1000M 1000W Power Supply 
CaseMouseMouse PadAudio
Thermaltake Level 10 GT Black Razer Imperator 2012 Expert Ergonomic Gaming Mouse Steelseries 4HD Random 5.1 Speaker System  
  hide details  
Reply
post #135 of 234
Quote:
Originally Posted by Sunreeper View Post

Since when is this news? I thought this was common enthusiast knowledge
That's what I said the first time I replied. biggrin.gif I guess people just use the thread for other stuff since the main one in uninteresting redface.gif
PC
(9 items)
 
  
CPUMotherboardGraphicsRAM
4790K 4.6G 1.24v/1.74v, 5.1G 1.35v validation. GA-Z97X-Gaming 7 Tri-X R9 290 1100/1350 +0.012v G.Skill 2400 c10 
CoolingMonitorPowerCase
Noctua NH-D15 24EA53 IPS 76Hz OC EVGA 1000 G2 Phanteks Enthoo Pro 
Audio
SoundMagic E10 
  hide details  
Reply
PC
(9 items)
 
  
CPUMotherboardGraphicsRAM
4790K 4.6G 1.24v/1.74v, 5.1G 1.35v validation. GA-Z97X-Gaming 7 Tri-X R9 290 1100/1350 +0.012v G.Skill 2400 c10 
CoolingMonitorPowerCase
Noctua NH-D15 24EA53 IPS 76Hz OC EVGA 1000 G2 Phanteks Enthoo Pro 
Audio
SoundMagic E10 
  hide details  
Reply
post #136 of 234
Quote:
Originally Posted by specopsFI View Post

So the problem is that you don't understand what those percentile graphs are telling us, not the graphs themselves.

Frame rate and frame time go hand in hand.They are just different ways to say the same thing. What you want is a steady frame rate, that is to say a frame rate that has a flat and linear percentile graph. You can demonstrate frame time issues perfectly well with frame rate numbers, just not average fps numbers. What we have in those percentile graphs by Guru3D is the frame rate information from all the frames, whereas an average fps number only tells the average frame rate.

That is why a frame rate percentile graph (based on data gathered by FCAT, not FRAPS) is a perfectly objective measure. Average fps is not, because it filters the data to only tell a part of the whole story. What PCPer is doing by rating the frames "good" or "bad" (a.k.a "runt" as they like to call them) is also a much more subjective thing than the percentile graphs by Guru3D. PCPer judges the data and filters it according to their judgement, Guru3D shows us the raw data and lets us do the judgement. I'm not saying that PCPer's judgement is flawed, I'm just saying that there is no such level of subjectivity in those Guru3D's percentile graphs.

NO and NO

average frame rate is the amount of frames rendered divided by the time it took to render those frames.
frame time is the amount of time it took to render a frame.
percentile tells how many frames vary from the shortest amount to the longest amount of time. (imo the only percentile that matter is >95%)

all three metrics need to be looked at as no one is more important than the three combined.
please read (or reread as it may) the first description of the process:
Inside the second: A new look at game benchmarking.
followed by:
Inside the second with Nvidia's frame capture tools
and also:
AMD Comments on GPU Stuttering, Offers Driver Roadmap & Perspective on Benchmarking
Frame Rating Dissected: Full Details on Capture-based Graphics Performance Testing


cheers.
loon 3.2
(18 items)
 
  
CPUMotherboardGraphicsRAM
i7-3770K Asus P8Z77-V Pro EVGA 980TI SC+ 16Gb PNY ddr3 1866 
Hard DriveHard DriveHard DriveOptical Drive
PNY 1311 240Gb 1 TB Seagate 3 TB WD Blue DVD DVDRW+/- 
CoolingCoolingOSMonitor
EKWB P280 kit EK-VGA supremacy Win X LG 24MC57HQ-P 
KeyboardPowerCaseMouse
Ducky Zero [blues] EVGA SuperNova 750 G2 Stryker M [hammered and drilled] corsair M65 
AudioAudio
SB Recon3D Klipsch ProMedia 2.1  
  hide details  
Reply
loon 3.2
(18 items)
 
  
CPUMotherboardGraphicsRAM
i7-3770K Asus P8Z77-V Pro EVGA 980TI SC+ 16Gb PNY ddr3 1866 
Hard DriveHard DriveHard DriveOptical Drive
PNY 1311 240Gb 1 TB Seagate 3 TB WD Blue DVD DVDRW+/- 
CoolingCoolingOSMonitor
EKWB P280 kit EK-VGA supremacy Win X LG 24MC57HQ-P 
KeyboardPowerCaseMouse
Ducky Zero [blues] EVGA SuperNova 750 G2 Stryker M [hammered and drilled] corsair M65 
AudioAudio
SB Recon3D Klipsch ProMedia 2.1  
  hide details  
Reply
post #137 of 234
Quote:
Originally Posted by looniam View Post

NO and NO

average frame rate is the amount of frames rendered divided by the time it took to render those frames.
frame time is the amount of time it took to render a frame.
percentile tells how many frames vary from the shortest amount to the longest amount of time. (imo the only percentile that matter is >95%)

all three metrics need to be looked at as no one is more important than the three combined.
please read (or reread as it may) the first description of the process:
Inside the second: A new look at game benchmarking.
followed by:
Inside the second with Nvidia's frame capture tools
and also:
AMD Comments on GPU Stuttering, Offers Driver Roadmap & Perspective on Benchmarking
Frame Rating Dissected: Full Details on Capture-based Graphics Performance Testing


cheers.

You're not exactly wrong in any of what you said (although the part I put in bolded and italic is pretty cryptic...) but you didn't understand what I said. Also, you don't seem to understand what those percentile graphs consist of.

Average frame rate is just that, average. Frame count by total time. Frame rate on the other can be used for frame-by-frame analysis, since it is the reciprocal of frame time: they tell the exact same thing in different terms. For distinction, I'm going to call this "frame-by-frame" frame rate the effective frame rate: that is, what the fps would be if all the frames (in a second, or in the test scene) were to render at the rate of the frame in question.

Frame rating is a really good tool to make the data easier to interpret, but all the data it needs for its analysis is included in a single percentile graph. A percentile graph shows the frame time (or the effective frame rate) across all the frames. Another way to look at a percentile graph is that it tells you the frame time (or the effective frame rate) across the whole percentile spectrum. That is why a percentile graph is so objective: it doesn't include subjective judgement calls, such as "imo the only percentile that matter is >95%".

Thanks for the links. I've read all of them a dozen times, so I can find them quite easily myself, but for some in this thread they might be useful.
The Air Tunnel
(10 items)
 
  
CPUMotherboardGraphicsRAM
i7-4820K (4500MHz@1.28V) P9X79 Sapphire R9 290 Tri-X New Edition Team Group Vulcan 4x4GB 2133MHz 
Hard DriveCoolingOSPower
2x240GB SSD @RAID 0 Noctua NH-D14 SE2011 Windows 10 Pro Corsair AX750 
CaseAudio
Rosewill Armor EVO hiFace+AudioGD+DT770Pro 
  hide details  
Reply
The Air Tunnel
(10 items)
 
  
CPUMotherboardGraphicsRAM
i7-4820K (4500MHz@1.28V) P9X79 Sapphire R9 290 Tri-X New Edition Team Group Vulcan 4x4GB 2133MHz 
Hard DriveCoolingOSPower
2x240GB SSD @RAID 0 Noctua NH-D14 SE2011 Windows 10 Pro Corsair AX750 
CaseAudio
Rosewill Armor EVO hiFace+AudioGD+DT770Pro 
  hide details  
Reply
post #138 of 234
Quote:
Originally Posted by specopsFI View Post

There are also many very well informed people saying that GK104 doesn't have the shader power needed to run settings that might require more than 2GB. IMO, for single chip they have a point. SLI is cutting it close, there are some cases where more VRAM would give a little more wiggle room for playable settings but only when you have tri-SLI GK104 will you really be bottlenecked by 2GB.

But again and unfortunately the GTX 690 owners thread came to the conclusion very quickly at release that the GTX 690 could use its 2GB of VRAM and hit the hard limit well before its processing power ran out in some games but particularly BF3 (Surround, no AA, no motion blur would hit the hard limit with around 60-70% GPU usage). If these well informed people would be as forward to argue that GK104 can't utilise 2GB of VRAM in SLI then they aren't as informed as they think they are. Sure all day anyone can say a single GK104 can't use 2GB but in SLI they most definitely can.

A single GK110 can use over 2GB of VRAM in surround with playable framerates also. People under estimate GK104 and over exhadurate the limitation of its memory bus.
    
CPUMotherboardGraphicsRAM
i7 4770k Gigabyte Z87X OC  Gigabyte GTX Titan Gskill Ripjaws 2400mhz 16GB 4x4GB  
Hard DriveHard DriveCoolingCooling
Corsair Force GS 240GB Western Digital Velociraptor 320GB EK GTX Titan Acetal Copper FC Block with Backplate EK Monarch DIMM Module heatspreaders 
CoolingCoolingCoolingOS
EK X3 Reservoir 150 EK Gigabyte Z87 X OC Acetal Copper Block EK Monarch X4 Copper Acetal Block with clean top Windows 8 64-Bit 
MonitorMonitorMonitorKeyboard
Dell U2412M Dell U2412M Dell U2412M Razer BlackWidow 
PowerCaseMouseMouse Pad
Corsair AX1200i Custome Acetal case CNC Machined from 8mm sheet  Razer Mamba Razer Ironclad 
OtherOtherOtherOther
Corsair Individually Sleeved Cables Koolance 90deg Elbows  Monsoon Compression Fittings Corsair SP120 Quiet Edition PWM Fans 
OtherOtherOtherOther
Koolance OD3 Quick Disconnect Male Compression Koolance OD3 Quick Disconnect Female G1/4 Threaded EK Acetal Clean Supremacy Block XSPC RX360 Radiator 
OtherOtherOther
XSPC Acrylic Tank Reservior/Pump Top XSPC RX360 Radiator  Koolance 450S D5 Pump 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
i7 4770k Gigabyte Z87X OC  Gigabyte GTX Titan Gskill Ripjaws 2400mhz 16GB 4x4GB  
Hard DriveHard DriveCoolingCooling
Corsair Force GS 240GB Western Digital Velociraptor 320GB EK GTX Titan Acetal Copper FC Block with Backplate EK Monarch DIMM Module heatspreaders 
CoolingCoolingCoolingOS
EK X3 Reservoir 150 EK Gigabyte Z87 X OC Acetal Copper Block EK Monarch X4 Copper Acetal Block with clean top Windows 8 64-Bit 
MonitorMonitorMonitorKeyboard
Dell U2412M Dell U2412M Dell U2412M Razer BlackWidow 
PowerCaseMouseMouse Pad
Corsair AX1200i Custome Acetal case CNC Machined from 8mm sheet  Razer Mamba Razer Ironclad 
OtherOtherOtherOther
Corsair Individually Sleeved Cables Koolance 90deg Elbows  Monsoon Compression Fittings Corsair SP120 Quiet Edition PWM Fans 
OtherOtherOtherOther
Koolance OD3 Quick Disconnect Male Compression Koolance OD3 Quick Disconnect Female G1/4 Threaded EK Acetal Clean Supremacy Block XSPC RX360 Radiator 
OtherOtherOther
XSPC Acrylic Tank Reservior/Pump Top XSPC RX360 Radiator  Koolance 450S D5 Pump 
  hide details  
Reply
post #139 of 234
Quote:
Originally Posted by thestache View Post

But again and unfortunately the GTX 690 owners thread came to the conclusion very quickly at release that the GTX 690 could use its 2GB of VRAM and hit the hard limit well before its processing power ran out in some games but particularly BF3 (Surround, no AA, no motion blur would hit the hard limit with around 60-70% GPU usage). If these well informed people would be as forward to argue that GK104 can't utilise 2GB of VRAM in SLI then they aren't as informed as they think they are. Sure all day anyone can say a single GK104 can't use 2GB but in SLI they most definitely can.

A single GK110 can use over 2GB of VRAM in surround with playable framerates also. People under estimate GK104 and over exhadurate the limitation of its memory bus.

Yes I know, I've read the thread. As I said, there are some cases where GK104 SLI might use more than 2GB, but usually that is not the case. I also would never make bottleneck estimations based on software reported GPU usage, or CPU usage for that matter. The only way to really test whether a bottleneck exist or not is to remove one potential bottleneck at a time. For GK104 VRAM bottleneck testing, that would mean taking a 2GB SLI and a 4GB SLI (preferrably using the exact same PCB and memory chips) and testing each case with both. The closest I've seen is [H]ardOCP's GTX 670 4GB SLI test, only caveats being that their cards didn't have the same memory chips (or memory timings or both) and that naturally they didn't test each and every game with each and every setting. Their conclusion? Two 670's run out of shader power before 2GB of VRAM become an issue, take three of them and then you might have something going for the 4GB version.
The Air Tunnel
(10 items)
 
  
CPUMotherboardGraphicsRAM
i7-4820K (4500MHz@1.28V) P9X79 Sapphire R9 290 Tri-X New Edition Team Group Vulcan 4x4GB 2133MHz 
Hard DriveCoolingOSPower
2x240GB SSD @RAID 0 Noctua NH-D14 SE2011 Windows 10 Pro Corsair AX750 
CaseAudio
Rosewill Armor EVO hiFace+AudioGD+DT770Pro 
  hide details  
Reply
The Air Tunnel
(10 items)
 
  
CPUMotherboardGraphicsRAM
i7-4820K (4500MHz@1.28V) P9X79 Sapphire R9 290 Tri-X New Edition Team Group Vulcan 4x4GB 2133MHz 
Hard DriveCoolingOSPower
2x240GB SSD @RAID 0 Noctua NH-D14 SE2011 Windows 10 Pro Corsair AX750 
CaseAudio
Rosewill Armor EVO hiFace+AudioGD+DT770Pro 
  hide details  
Reply
post #140 of 234
You are generalizing correlation of VRAM with GPU power very dangerously.

Do some 3D programming and you'll soon realize it's depended on what you're doing. You may be using 1MB VRAM to do insanely hard calculations or you may offload the whole of wikimedia.org on it and do nothing but show a slideshow.

So, it just depends on the game. Fine, some games may behave like that but in general.. it can't be generalized.
PC
(9 items)
 
  
CPUMotherboardGraphicsRAM
4790K 4.6G 1.24v/1.74v, 5.1G 1.35v validation. GA-Z97X-Gaming 7 Tri-X R9 290 1100/1350 +0.012v G.Skill 2400 c10 
CoolingMonitorPowerCase
Noctua NH-D15 24EA53 IPS 76Hz OC EVGA 1000 G2 Phanteks Enthoo Pro 
Audio
SoundMagic E10 
  hide details  
Reply
PC
(9 items)
 
  
CPUMotherboardGraphicsRAM
4790K 4.6G 1.24v/1.74v, 5.1G 1.35v validation. GA-Z97X-Gaming 7 Tri-X R9 290 1100/1350 +0.012v G.Skill 2400 c10 
CoolingMonitorPowerCase
Noctua NH-D15 24EA53 IPS 76Hz OC EVGA 1000 G2 Phanteks Enthoo Pro 
Audio
SoundMagic E10 
  hide details  
Reply
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Hardware News
Overclock.net › Forums › Industry News › Hardware News › [ FORBES ] Crunching The Numbers: Can Dual $269 Nvidia GTX 760s Outperform A $999 Titan?