Overclock.net › Forums › Benchmarks › Benchmarking Software and Discussion › [Various] Futuremark's Time Spy DirectX 12 "Benchmark" Compromised. Less Compute/Parallelism than Doom/Aots. Also...
New Posts  All Forums:Forum Nav:

[Various] Futuremark's Time Spy DirectX 12 "Benchmark" Compromised. Less Compute/Parallelism than Doom/Aots. Also... - Page 16

post #151 of 253
Quote:
Originally Posted by Majin SSJ Eric View Post

Warning: Spoiler! (Click to show)
Seems to me that this is all just a failure of DX12 itself to adhere to one standard regardless of vendor. Features like Async should just be part of the API whether or not certain hardware can take advantage of it. It shouldn't be left to vendors or developers to pick and choose which feature set they want to implement, it should all be standard. Now of course I know very little about API's and how they work so maybe this comes off as sounding stupid to the guys that really understand this stuff but it seems to me (an admitted layman) that DX12's implementation is far more fractured and sloppy than DX11's was, which is a shame considering I thought the point of it was to be low level and simple to utilize. It just doesn't make sense to offer all kinds of differing features that may or may not ever be supported and just leave it up to everyone else to pick and choose which ones they will utilize a la carte.

TL;DR - There should just be one unified feature set of DX12 that includes everything and either hardware will be compatible or it won't. It makes much more sense why Nvidia didn't even bother to include certain features on a hardware level now since they knew they could simply muscle everybody to code for them first anyway though...
DX12 FL11_0 fits most Hardware from 2012 until now

Maybe Because AMD was who pushed the first API was the (prime) Sample for other APIs like Vulkan And DX12 and (one) of the features mantle allowed was Asynchronous Shaders(Compute+Graphics+Copy) http://www.pcper.com/reviews/Editorial/Hot-Topic-Asynchronous-Shaders it is clearly taking advantage of GCN´s Parallelism, but When AMD released mantle, Nvidia was pushing DX11 "lower overhead" on kepler and probably tried to improve Kepler architecture while using some Features for DX12 but then still focusing their Maxwell GPUs for DX11
Edited by PontiacGTX - 7/19/16 at 7:13pm
Wanted: [WTB] GPU upgrade
$210 (USD) or best offer
  
Reply
Wanted: [WTB] GPU upgrade
$210 (USD) or best offer
  
Reply
post #152 of 253
I would like to know how the D3D Runtime takes parallel code and serializes it so efficiently. Is D3D Runtime so efficient that any card unable to process parallel queues doesn't stand to lose performance when that code is serialized by the D3D Runtime? For example, the 970 scores the same in Time Spy, using the default(async on), and with identical custom settings(w/ async off). Should we expect the 970 to maintain performance in other games that require the D3D runtime to serialize parallel queues?
Edited by Doothe - 7/19/16 at 8:27pm
post #153 of 253
Isn't it funny tho, when GCN is penalized by dx11 thats fine, but now that the hardware can be property used under the new api we must remain neutral and use paths that completions supports.
Bartholomew I
(15 items)
 
  
CPUMotherboardGraphicsRAM
Intel Core i7-4770K Gigabyte GA-Z87X-UD3H STRIX-R9390-DC3OC-8GD5-GAMING Corsair Vengeance Pro  
Hard DriveHard DriveOptical DriveCooling
Kingston SSDNow V300 Western Digital WD Green WD10EADS Lite-On Super AllWrite IHAS124-04  Cooler Master: Hyper 212 EVO 
OSMonitorKeyboardPower
Microsoft Windows 10 Pro 64-bit  Dell UltraSharp U2312HM  Genius GX Imperator I  Corsair AX760  
CaseMouseOther
Cooler Master: HAF 932 Advanced Bloody Multi-Core Gun 3 V8 Thrustmaster Ferrari F430 Force Feedback Racing... 
  hide details  
Reply
Bartholomew I
(15 items)
 
  
CPUMotherboardGraphicsRAM
Intel Core i7-4770K Gigabyte GA-Z87X-UD3H STRIX-R9390-DC3OC-8GD5-GAMING Corsair Vengeance Pro  
Hard DriveHard DriveOptical DriveCooling
Kingston SSDNow V300 Western Digital WD Green WD10EADS Lite-On Super AllWrite IHAS124-04  Cooler Master: Hyper 212 EVO 
OSMonitorKeyboardPower
Microsoft Windows 10 Pro 64-bit  Dell UltraSharp U2312HM  Genius GX Imperator I  Corsair AX760  
CaseMouseOther
Cooler Master: HAF 932 Advanced Bloody Multi-Core Gun 3 V8 Thrustmaster Ferrari F430 Force Feedback Racing... 
  hide details  
Reply
post #154 of 253
Quote:
Originally Posted by xboxshqip View Post

Isn't it funny tho, when GCN is penalized by dx11 thats fine, but now that the hardware can be property used under the new api we must remain neutral and use paths that completions supports.
Yep, because we don't want to see amd going ahead. instead we wait for nvidia to catch up and meanwhile hold amd back.

along with fake tears of sympathy towards amd and how we want them to compete and how we want them to be on top. but the truth is we don't really want that to happen. tongue.gif
post #155 of 253
Well a competitive AMD means even those who wear green underpants will be able to buy more green with less green, so anyone who wishes for AMD to fail is anti-consumer really.
post #156 of 253
This is how I view it. A "benchmark" should be a program that is able to utilize the latest and greatest and all of an API's features. Even features that aren't really seen in game yet. So that users get an idea of what you can come to expect. Direct X 12 benchmark should be a program that makes use of FL_12.0 and 12.1. Not something less because "Oh we only see the market going this way in a years time".

How I see this situation. And this is purely dumbed down due to my limited understanding of API's and someone can correct me if I am wrong.

However let's use Heaven Bench for a minute. Showing us all these wonderful features of what DX11 is capable of.

Now let's have three graphics cards.

GTX260 216 does not fully support DX11. It is 10.1. But it can still run Heaven benchmark. The card is being utilized to 100% of its potential. And it scores let's say 100.

Here comes Cypress XT with its full DX11 Support it can run heaven with all the latest and greatest turned on cause it has the hardware to do it. It is being utilized to 100% of its ability and scores at 150 points.

Here comes Fermi. It has better Tesselation than Cypress and also fully supports DX11. Being utilized 100% and scores 250.

Each piece of hardware regardless of its capabilities is being used to 100% of that particular pieces potential. The benchmark is fair.

Now enter time spy.

GTX680 comes along. With only FL_11.3 or DX12 fl_11.0 support or whatever. Time spy runs, the card gets used to 100% of its potential and spits out a score of say 100.

GTX1080 comes along with preemption and time spy runs and the card is utilized to 100% of its potential. And scores let's say 200.

Then a GCN based card comes along. But has full DX12 support with true asynchronous support. But time spy cannot USE full asynchronous compute as it only has a single render pipeline so a GCN card runs time spy is being utilize to say 65% of its potential and spits out a score of 150.

This is not a fair benchmark. As not all the cards are being used to thir full potential and niether is time spy showing what the API's or the hardware that supports it is truly capable of. It therefor isn't really a benchmark then.

Joe Gamer comes along and sees card X is 5% behind card Y in time spy, yet costs 15% less. So Joe decides to buy card X because he can't see the value in spending 15% more for only 5% performance. But then Joe gamer runs FL_12.0 games or Vulcan games that fully utilize the capability of Card Y and suddenly card Y ends up being 15% faster than X. Which was more than he was lead to believe when he looked at the time spy scores. Who does he blame?

Yes you and I here on OCN know lots about hardware and benching and how tests are done and know not to put a whole Lotta stock into synthetic tests. But average Joe Gamer doesn't know this. So now who does Joe Gamer blame when his card does not perform like Time spy lead him to belive? Or his card performs far worse than the competing card yet the deficit was only shown to be 5%. What then?

This is why I think Time spy should either be patched. Or removed completely as a benchmark. Or Time Spy should be renamed to reflect that is only a DX12 FL_11.0 benchmark and not a full FL_12 bench.
Edited by Alastair - 7/20/16 at 4:08am
GHOST rev 3.1
(28 items)
 
 
Rura Penthe
(12 items)
 
CPUMotherboardGraphicsRAM
AMD FX-8370 @ 4.95GHz 1.5V 2700MHz NB/3000MHz HTT ASUS M5A99FX PRO R2.0 2x Sapphire R9-Fury 3840 Corsair Vengeance 2133 4x4GB @ 2000 9-9-10-27 1CR 
Hard DriveHard DriveHard DriveHard Drive
Samsung 850 PRO 1TB WD Blue 500GB WD Blue 1 TB Samsung Spinpoint HD502HI 
CoolingCoolingCoolingCooling
XSPC D5 Bay Res 3x CoolerMaster Storm Force 200's 2x EK-FC Fury X fullcovers EK-FC Terminal Dual Parallel 
CoolingCoolingCoolingCooling
2x EK-FC Fury X Backplates XSPC Raystorm CPU Block EKWB Coolstream CE280 EKWB Coolstream PE360 
CoolingCoolingCoolingOS
3x CoolerMaster Jetflo's 120mm 6 Corsair ML140's  Laing D5 Vario Windows 10 Pro 64-Bit 
MonitorKeyboardPowerCase
LG E2341V Roccat Ryos MK Pro Antec High Current Pro Platinum 1300W NZXT Phantom 820 Black 
MouseMouse PadAudioAudio
Mionix Naos 7000 Mionix Alioth GX Gaming SW-G2.1 3000 Kingston Hyper X Cloud Core 
CPUMotherboardGraphicsGraphics
AMD Athlon 860K Asus A88X-Pro MSI R6850 Cyclone Power Edition MSI R6850 Cyclone Power Edition 
RAMHard DriveOptical DriveCooling
G-Skill Ares 2133 CL10 8GB (2x4GB) Seagate ST1000VX003 Samsung HD502HI Arctic Cooling Freezer A30 
OSKeyboardPowerCase
Windows 10 Home Edition 64bit Razer Lycrosa Seagate M12II 620W Corsair Carbide SPEC-02 
CPUMotherboardGraphicsGraphics
Intel Celeron G1840 MSI Z97-Guard Pro Gigabyte RX580 Gaming 4G Gigabyte RX570 Gaming 4G 
GraphicsGraphicsGraphicsRAM
XFX RX480 reference edition 8G Gigabyte RX570 Gaming 4g Gigabyte RX570 Gaming 4G G-Skill Ares F3-1600C9D 
Hard DriveCoolingOSPower
Samsung 750 Evo 120GB Stock  Windows 10 Pro  Gigabyte XP1200M 
  hide details  
Reply
GHOST rev 3.1
(28 items)
 
 
Rura Penthe
(12 items)
 
CPUMotherboardGraphicsRAM
AMD FX-8370 @ 4.95GHz 1.5V 2700MHz NB/3000MHz HTT ASUS M5A99FX PRO R2.0 2x Sapphire R9-Fury 3840 Corsair Vengeance 2133 4x4GB @ 2000 9-9-10-27 1CR 
Hard DriveHard DriveHard DriveHard Drive
Samsung 850 PRO 1TB WD Blue 500GB WD Blue 1 TB Samsung Spinpoint HD502HI 
CoolingCoolingCoolingCooling
XSPC D5 Bay Res 3x CoolerMaster Storm Force 200's 2x EK-FC Fury X fullcovers EK-FC Terminal Dual Parallel 
CoolingCoolingCoolingCooling
2x EK-FC Fury X Backplates XSPC Raystorm CPU Block EKWB Coolstream CE280 EKWB Coolstream PE360 
CoolingCoolingCoolingOS
3x CoolerMaster Jetflo's 120mm 6 Corsair ML140's  Laing D5 Vario Windows 10 Pro 64-Bit 
MonitorKeyboardPowerCase
LG E2341V Roccat Ryos MK Pro Antec High Current Pro Platinum 1300W NZXT Phantom 820 Black 
MouseMouse PadAudioAudio
Mionix Naos 7000 Mionix Alioth GX Gaming SW-G2.1 3000 Kingston Hyper X Cloud Core 
CPUMotherboardGraphicsGraphics
AMD Athlon 860K Asus A88X-Pro MSI R6850 Cyclone Power Edition MSI R6850 Cyclone Power Edition 
RAMHard DriveOptical DriveCooling
G-Skill Ares 2133 CL10 8GB (2x4GB) Seagate ST1000VX003 Samsung HD502HI Arctic Cooling Freezer A30 
OSKeyboardPowerCase
Windows 10 Home Edition 64bit Razer Lycrosa Seagate M12II 620W Corsair Carbide SPEC-02 
CPUMotherboardGraphicsGraphics
Intel Celeron G1840 MSI Z97-Guard Pro Gigabyte RX580 Gaming 4G Gigabyte RX570 Gaming 4G 
GraphicsGraphicsGraphicsRAM
XFX RX480 reference edition 8G Gigabyte RX570 Gaming 4g Gigabyte RX570 Gaming 4G G-Skill Ares F3-1600C9D 
Hard DriveCoolingOSPower
Samsung 750 Evo 120GB Stock  Windows 10 Pro  Gigabyte XP1200M 
  hide details  
Reply
post #157 of 253
Quote:
Originally Posted by Alastair View Post

This is why I think Time spy should either be patched. Or removed completely as a benchmark. Or Time Spy should be renamed to reflect that is only a DX12 FL_11.3 benchmark and not a full FL_12 bench.

What's FL_11.3? Do you mean DX11.3? There's no such thing as FL_11.3

Time Spy is a DX12 FL 11_0 benchmark, not a DX11.3 benchmark. There's a difference: DX11.3 is old DX11 with newer DX12 rendering features, no changes to submission model. DX12 FL 11_0 is DX11 rendering features + multi-queue and multi-thread submission. Totally different beasts.

Time Spy never claims to be a FL 12_0 bench.

And why stop at FL 12_0? Why not FL 12_1? You would be okay if Maxwell/Pascal blows all other cards out of the water with conservative rasterization and ROVs, since all cards will have it in the future, right?
post #158 of 253
Why is everyone in uproar? 3D mark is known over many benchmarks they've put put out in the past to be heavily Nvidia biased. This behaviors of synchronous compute is not surprising.

However, it is a ethics issue that they called it in the benchmark "asynchronous compute" when in reality what the benchmark does is synchronous compute. Two very different things.
What
(12 items)
 
  
CPUMotherboardGraphicsRAM
i7-6950X X99 Deluxe Vega 64 32GB @ 3200 
Hard DriveOSMonitorKeyboard
Samsung 960 Pro Windows 10 Pro x64 Dell UltraSharp U3014 Microsoft SideWinder X4 
PowerCaseMouseAudio
Corsair AX860 NZXT Switch 810 Mionix Castor Asus Xonar STX + 2x LME49860NA OPs + HD650s 
  hide details  
Reply
What
(12 items)
 
  
CPUMotherboardGraphicsRAM
i7-6950X X99 Deluxe Vega 64 32GB @ 3200 
Hard DriveOSMonitorKeyboard
Samsung 960 Pro Windows 10 Pro x64 Dell UltraSharp U3014 Microsoft SideWinder X4 
PowerCaseMouseAudio
Corsair AX860 NZXT Switch 810 Mionix Castor Asus Xonar STX + 2x LME49860NA OPs + HD650s 
  hide details  
Reply
post #159 of 253
they do a little bit of async but its like 10% compared to real games..
also if they show that 390X is 10% slower than GTX1070 under DX12 their main money flow from NVIDIA will maybe not flow anymore
FX
(7 items)
 
  
CPUMotherboardGraphicsRAM
FX-8320@4.4Ghz M5A99FX PRO R2.0 AMD Radeon R9 290X Patriot Memory  
Hard DriveCoolingMonitor
Samsung 840 Raijintek Ereboss iiyama X4071UHSU (4K) 
  hide details  
Reply
FX
(7 items)
 
  
CPUMotherboardGraphicsRAM
FX-8320@4.4Ghz M5A99FX PRO R2.0 AMD Radeon R9 290X Patriot Memory  
Hard DriveCoolingMonitor
Samsung 840 Raijintek Ereboss iiyama X4071UHSU (4K) 
  hide details  
Reply
post #160 of 253
Quote:
Originally Posted by Exilon View Post

Quote:
Originally Posted by Alastair View Post

This is why I think Time spy should either be patched. Or removed completely as a benchmark. Or Time Spy should be renamed to reflect that is only a DX12 FL_11.3 benchmark and not a full FL_12 bench.

What's FL_11.3? Do you mean DX11.3? There's no such thing as FL_11.3

Time Spy is a DX12 FL 11_0 benchmark, not a DX11.3 benchmark. There's a difference: DX11.3 is old DX11 with newer DX12 rendering features, no changes to submission model. DX12 FL 11_0 is DX11 rendering features + multi-queue and multi-thread submission. Totally different beasts.

Time Spy never claims to be a FL 12_0 bench.

And why stop at FL 12_0? Why not FL 12_1? You would be okay if Maxwell/Pascal blows all other cards out of the water with conservative rasterization and ROVs, since all cards will have it in the future, right?
yes I did mean 11.0 I simply got Co fused between the two.

However my point still stands if the Hardware and the API support it show it off. If you read my post I mentioned both Fl12. 0 and 12.1.

My view point if the hardware supports it use it. Show it off. Show off the features the end user has purchased. Or show them what various parts are capable of so they users may make informed buying decisions. Don't hide features various products are capable of and then go touting it as a fair benchmark. When it no longer is

The fact is FM claimed time spy as a DX12 benchmark. So what they really mean is "we only kinda support DX12 at a bare minimum level. As we don't actually have proper pipelines for the really headline features such as Asynchronous compute. But we won't tell you that cause we will let you all assume it's fully complaint DX12 bench" This is NOT the right attitude.

And yes. If Maxwell and Pascal have fancy features too then by all means. Show them off. I have no issues with X card being better than Y card if X card is ACTUALLY better than Y card.

But I am not happy with Xcard being shown to be better than Y card by a benchmark when this may not be true because the benchmark is picking and choosing what features to use. And so thereby pulling the wool over consumers eyes.
Edited by Alastair - 7/20/16 at 4:33am
GHOST rev 3.1
(28 items)
 
 
Rura Penthe
(12 items)
 
CPUMotherboardGraphicsRAM
AMD FX-8370 @ 4.95GHz 1.5V 2700MHz NB/3000MHz HTT ASUS M5A99FX PRO R2.0 2x Sapphire R9-Fury 3840 Corsair Vengeance 2133 4x4GB @ 2000 9-9-10-27 1CR 
Hard DriveHard DriveHard DriveHard Drive
Samsung 850 PRO 1TB WD Blue 500GB WD Blue 1 TB Samsung Spinpoint HD502HI 
CoolingCoolingCoolingCooling
XSPC D5 Bay Res 3x CoolerMaster Storm Force 200's 2x EK-FC Fury X fullcovers EK-FC Terminal Dual Parallel 
CoolingCoolingCoolingCooling
2x EK-FC Fury X Backplates XSPC Raystorm CPU Block EKWB Coolstream CE280 EKWB Coolstream PE360 
CoolingCoolingCoolingOS
3x CoolerMaster Jetflo's 120mm 6 Corsair ML140's  Laing D5 Vario Windows 10 Pro 64-Bit 
MonitorKeyboardPowerCase
LG E2341V Roccat Ryos MK Pro Antec High Current Pro Platinum 1300W NZXT Phantom 820 Black 
MouseMouse PadAudioAudio
Mionix Naos 7000 Mionix Alioth GX Gaming SW-G2.1 3000 Kingston Hyper X Cloud Core 
CPUMotherboardGraphicsGraphics
AMD Athlon 860K Asus A88X-Pro MSI R6850 Cyclone Power Edition MSI R6850 Cyclone Power Edition 
RAMHard DriveOptical DriveCooling
G-Skill Ares 2133 CL10 8GB (2x4GB) Seagate ST1000VX003 Samsung HD502HI Arctic Cooling Freezer A30 
OSKeyboardPowerCase
Windows 10 Home Edition 64bit Razer Lycrosa Seagate M12II 620W Corsair Carbide SPEC-02 
CPUMotherboardGraphicsGraphics
Intel Celeron G1840 MSI Z97-Guard Pro Gigabyte RX580 Gaming 4G Gigabyte RX570 Gaming 4G 
GraphicsGraphicsGraphicsRAM
XFX RX480 reference edition 8G Gigabyte RX570 Gaming 4g Gigabyte RX570 Gaming 4G G-Skill Ares F3-1600C9D 
Hard DriveCoolingOSPower
Samsung 750 Evo 120GB Stock  Windows 10 Pro  Gigabyte XP1200M 
  hide details  
Reply
GHOST rev 3.1
(28 items)
 
 
Rura Penthe
(12 items)
 
CPUMotherboardGraphicsRAM
AMD FX-8370 @ 4.95GHz 1.5V 2700MHz NB/3000MHz HTT ASUS M5A99FX PRO R2.0 2x Sapphire R9-Fury 3840 Corsair Vengeance 2133 4x4GB @ 2000 9-9-10-27 1CR 
Hard DriveHard DriveHard DriveHard Drive
Samsung 850 PRO 1TB WD Blue 500GB WD Blue 1 TB Samsung Spinpoint HD502HI 
CoolingCoolingCoolingCooling
XSPC D5 Bay Res 3x CoolerMaster Storm Force 200's 2x EK-FC Fury X fullcovers EK-FC Terminal Dual Parallel 
CoolingCoolingCoolingCooling
2x EK-FC Fury X Backplates XSPC Raystorm CPU Block EKWB Coolstream CE280 EKWB Coolstream PE360 
CoolingCoolingCoolingOS
3x CoolerMaster Jetflo's 120mm 6 Corsair ML140's  Laing D5 Vario Windows 10 Pro 64-Bit 
MonitorKeyboardPowerCase
LG E2341V Roccat Ryos MK Pro Antec High Current Pro Platinum 1300W NZXT Phantom 820 Black 
MouseMouse PadAudioAudio
Mionix Naos 7000 Mionix Alioth GX Gaming SW-G2.1 3000 Kingston Hyper X Cloud Core 
CPUMotherboardGraphicsGraphics
AMD Athlon 860K Asus A88X-Pro MSI R6850 Cyclone Power Edition MSI R6850 Cyclone Power Edition 
RAMHard DriveOptical DriveCooling
G-Skill Ares 2133 CL10 8GB (2x4GB) Seagate ST1000VX003 Samsung HD502HI Arctic Cooling Freezer A30 
OSKeyboardPowerCase
Windows 10 Home Edition 64bit Razer Lycrosa Seagate M12II 620W Corsair Carbide SPEC-02 
CPUMotherboardGraphicsGraphics
Intel Celeron G1840 MSI Z97-Guard Pro Gigabyte RX580 Gaming 4G Gigabyte RX570 Gaming 4G 
GraphicsGraphicsGraphicsRAM
XFX RX480 reference edition 8G Gigabyte RX570 Gaming 4g Gigabyte RX570 Gaming 4G G-Skill Ares F3-1600C9D 
Hard DriveCoolingOSPower
Samsung 750 Evo 120GB Stock  Windows 10 Pro  Gigabyte XP1200M 
  hide details  
Reply
New Posts  All Forums:Forum Nav:
  Return Home
Overclock.net › Forums › Benchmarks › Benchmarking Software and Discussion › [Various] Futuremark's Time Spy DirectX 12 "Benchmark" Compromised. Less Compute/Parallelism than Doom/Aots. Also...