Overclock.net › Forums › Industry News › Hardware News › [Various] Futuremark Releases 3DMark Time Spy DirectX 12 Benchmark
New Posts  All Forums:Forum Nav:

[Various] Futuremark Releases 3DMark Time Spy DirectX 12 Benchmark - Page 73

post #721 of 772
im still scratching my head as why we need a benchmark that pretends to be a game in certain conditions.

we have a games that can do that on their own.

here i was thinking benchmarks were to show tech available and to push hardware to the max to see what copes best in which situations. I read Futuremark's report as a giant cop-out. A DX12 feature level 11 bench that doesnt attempt to push hardware, rather just emulates what they think will be the average game over next 2 years is not what i thought i was paying for...

one for the crapware pile.
haswell-e build
(19 items)
 
  
CPUMotherboardGraphicsGraphics
i7 5820k GA-X99-UD4 r9 290x r9 290x  
GraphicsRAMHard DriveHard Drive
r9 290x  crucial DDR4 samsung evo  WD caviar blck 
CoolingOSMonitorMonitor
Custom loop Win 8.1  dell 2311 dell 2311 
MonitorKeyboardPowerCase
dell 2311  Logitech G110 1xcorsair hx1050 HAF stacker 945 
AudioOtherOther
x-fi fatal1ty pro corsair tx750 haf stacker 915r 
  hide details  
Reply
haswell-e build
(19 items)
 
  
CPUMotherboardGraphicsGraphics
i7 5820k GA-X99-UD4 r9 290x r9 290x  
GraphicsRAMHard DriveHard Drive
r9 290x  crucial DDR4 samsung evo  WD caviar blck 
CoolingOSMonitorMonitor
Custom loop Win 8.1  dell 2311 dell 2311 
MonitorKeyboardPowerCase
dell 2311  Logitech G110 1xcorsair hx1050 HAF stacker 945 
AudioOtherOther
x-fi fatal1ty pro corsair tx750 haf stacker 915r 
  hide details  
Reply
post #722 of 772
Quote:
Originally Posted by JackCY View Post

You mean the PcPer interview on YT yesterday?

I didn't even know PcPer had an interview with Futuremark. Looks like Futuremark has went into a full blown PR crisis mode.
    
CPUMotherboardGraphicsRAM
Xeon 5660 @ 4.8Ghz [Highest OC 5.4Ghz] ASUS Sabertooth X58 AMD Fury X 24GB - 1600Mhz Triple Channel 
Hard DriveHard DriveHard DriveHard Drive
Seagate Barracuda 7200 1TB RAID 0 - B Seagate Barracuda 7200 1TB RAID 0 - B Seagate Barracuda 7200 1TB RAID 0 - C Seagate Barracuda 7200 1TB RAID 0 - C 
Hard DriveHard DriveHard DriveCooling
SSD 128GB RAID - A SSD 128GB RAID - A SSD 256GB  Antec Kuhler H2O 620 [Pull] 
OSMonitorPowerOther
Windows 10 Professional  Dual 24-inch Monitors EVGA SuperNOVA G2 1300W x2 Delta FFB1212EH-F00 Fan 4,000rpm  
Other
x4 Scythe Gentle Typhoon D1225C12BBAP-31 Fan 54... 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
Xeon 5660 @ 4.8Ghz [Highest OC 5.4Ghz] ASUS Sabertooth X58 AMD Fury X 24GB - 1600Mhz Triple Channel 
Hard DriveHard DriveHard DriveHard Drive
Seagate Barracuda 7200 1TB RAID 0 - B Seagate Barracuda 7200 1TB RAID 0 - B Seagate Barracuda 7200 1TB RAID 0 - C Seagate Barracuda 7200 1TB RAID 0 - C 
Hard DriveHard DriveHard DriveCooling
SSD 128GB RAID - A SSD 128GB RAID - A SSD 256GB  Antec Kuhler H2O 620 [Pull] 
OSMonitorPowerOther
Windows 10 Professional  Dual 24-inch Monitors EVGA SuperNOVA G2 1300W x2 Delta FFB1212EH-F00 Fan 4,000rpm  
Other
x4 Scythe Gentle Typhoon D1225C12BBAP-31 Fan 54... 
  hide details  
Reply
post #723 of 772
Quote:
Originally Posted by Kana-Maru View Post

I didn't even know PcPer had an interview with Futuremark. Looks like Futuremark has went into a full blown PR crisis mode.
Someone just posted it here on OP. But it's the same all over again they are just doing the PR/politically correct info is given and explanations of APIs. They don't want to delve into HW and specifics even in email communication, they simply explain stuff and how they made the "benchmark". I'm trying more technical questions but will see if they reply again. I was actually surprised a little they did reply, I thought their inbox must be full lol.

My last question for them is in regard to Fury X if TimeSpy is able to fully utilize it or it has unnecessary idle time, and if the code was optimized better the idle time could be reduced.
As far as all the queues and compute engines go, DX12 doesn't give access to that, it's up to the driver it seems to split the work around the HW. Of course if you feed the GPU like a moron it will work suboptimal and the driver can't do a thing about it. Aka the usual universal approach frown.gif
Quote:
Published: Tuesday, 19 July 2016 22:13
Written by Kana Maru

Hey didn't you write that article about the Fury X being under utilized in TimeSpy? biggrin.gif
If so you should talk to Futuremark thumb.gif
Or did they already reply to your question about that?
Edited by JackCY - 7/21/16 at 9:55am
post #724 of 772
Fact...Time Spy is only optimized for a 2 path paralell in GFX cards.

Fact...NV GPUs are only capable of doing thier versionnof async on 2 pipelines.

Fact...AMD has cards that range from 2 pipelines of parallelism all the way up to 8 depending on the card.

Fact...FM stated in both print and the PCper interview that they made TS in such a way that it will only take advanatbe of Hardware that ALL cards are capable of.

That being the case these results make perfect sense. In a serial pipeline like D11 NV cards beat AMD head to head on their raw "horsepower".

AMD knew their limitations in mgf and RnD wouldnt allow them to win head to head so they changed their desgin to feature massive amounts of parallelism instead. An idea that theoretically would work better if codes for properly. Enter Mantle...Vulcan...DX12.

All these APIs have one major thing in common over previous generations. The ability to let developers utilize more of the idle portions of a GPU through parallelism. It is the one thing that changes the game entirely. Raw horsepower is great. But spliting up tasks andnrunning them simultaneously when possible can also be just as great.

This TS benchmark is biased in that it doesnt allow AMD to use then level of paralellism itnis capable of. Paralellism is not an advanced feature or an independent piece of tech like gameworks or phyx. It is fundamentally THE point of the change in APIs.

All this bench shows is that with current levels of tech...AMD and NV display perform in presictable ways.
When limiting the number of parallel tasks to one or two, NV horses run a bit faster.

The catch is AMD has as many as 6 horses it is being forced not to hitch to the chariot. There is already ample evidence to show that in this generation of tech more horses when given the chance to do work out shines faster ones. When they are left on the sideline results are what we expect. Faster horses win in an equal race.

The real question is will FM comment rung true about the future. Specifically will developers only code for 2 paths since 70% of the market will see a benifit but then negelct the other 30% not beceasue it isnt capablenof more, but because they just dont have the time or resources or care to bother.
post #725 of 772
Quote:
Originally Posted by gapottberg View Post

Fact...AMD has cards that range from 2 pipelines of parallelism all the way up to 8 depending on the card.
I couldn't get a proper response for that from FM. They wouldn't say how well they send work to the GPUs for the driver to be able to split it on as many compute engines as possible. Actually Nvidia has 1 not 2, they simply have to switch and Pascal improves the switching with preemption.
Quote:
Fact...FM stated in both print and the PCper interview that they made TS in such a way that it will only take advanatbe of Hardware that ALL cards are capable of.
Yes.
Quote:
This TS benchmark is biased in that it doesnt allow AMD to use then level of paralellism itnis capable of. Paralellism is not an advanced feature or an independent piece of tech like gameworks or phyx. It is fundamentally THE point of the change in APIs.
I agree. And their reply to that is that they cooperate with HW vendors/GPU makers. Tada, no further comment on how well or how poorly TS uses paralellism. They protect themselves with "we estimate how game development WILL look like and how much time devs will invest into optimizations".
Quote:
When limiting the number of parallel tasks to one or two, NV horses run a bit faster.
They have worse IPC but they clock faster and are optimized to run the way many devs write apps because of Nvidia supporting devs more IMHO + larger market share in recent years.
It would be interesting to see performance per transistor count when you max out the GPU with optimized application, but that's hard to get info. I don't remember the current standing without the optimizations but that is easy to do. I remember projecting that 1060 will need about 4.4mil transistors to match 480 when doing the math from existing 1080/70 and 480 numbers.
Quote:
The real question is will FM comment rung true about the future. Specifically will developers only code for 2 paths since 70% of the market will see a benifit but then negelct the other 30% not beceasue it isnt capablenof more, but because they just dont have the time or resources or care to bother.
Their comment is that devs and their management often doesn't want to bother with optimizations, which I get they are risky and time consuming but in this case the change to utilize the HW better should not be as difficult as trying to optimize some last 0-5% of performance on Nvidia.
Still looking at recent DX11 titles you can see the state of development, it's difficult and gets worse with younger generations at the wheels. Software development in general is getting more high level where as to extract performance you need to be as low level as possible.
Edited by JackCY - 7/21/16 at 10:27am
post #726 of 772
Quote:
Originally Posted by JackCY View Post

Someone just posted it here on OP. But it's the same all over again they are just doing the PR/politically correct info is given and explanations of APIs. They don't want to delve into HW and specifics even in email communication, they simply explain stuff and how they made the "benchmark". I'm trying more technical questions but will see if they reply again. I was actually surprised a little they did reply, I thought their inbox must be full lol.

My last question for them is in regard to Fury X if TimeSpy is able to fully utilize it or it has unnecessary idle time, and if the code was optimized better the idle time could be reduced.
As far as all the queues and compute engines go, DX12 doesn't give access to that, it's up to the driver it seems to split the work around the HW. Of course if you feed the GPU like a moron it will work suboptimal and the driver can't do a thing about it. Aka the usual universal approach frown.gif

Hey didn't you write that article about the Fury X being under utilized in TimeSpy? biggrin.gif
If so you should talk to Futuremark thumb.gif
Or did they already reply to your question about that?

Yeah I did write it. Time Spy has some interesting behavior for sure. I don't think Futuremark wants to talk to me lol. I think they would rather debunk anything I have to say and move on. Good luck trying to get answers out of Futuremark though.
    
CPUMotherboardGraphicsRAM
Xeon 5660 @ 4.8Ghz [Highest OC 5.4Ghz] ASUS Sabertooth X58 AMD Fury X 24GB - 1600Mhz Triple Channel 
Hard DriveHard DriveHard DriveHard Drive
Seagate Barracuda 7200 1TB RAID 0 - B Seagate Barracuda 7200 1TB RAID 0 - B Seagate Barracuda 7200 1TB RAID 0 - C Seagate Barracuda 7200 1TB RAID 0 - C 
Hard DriveHard DriveHard DriveCooling
SSD 128GB RAID - A SSD 128GB RAID - A SSD 256GB  Antec Kuhler H2O 620 [Pull] 
OSMonitorPowerOther
Windows 10 Professional  Dual 24-inch Monitors EVGA SuperNOVA G2 1300W x2 Delta FFB1212EH-F00 Fan 4,000rpm  
Other
x4 Scythe Gentle Typhoon D1225C12BBAP-31 Fan 54... 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
Xeon 5660 @ 4.8Ghz [Highest OC 5.4Ghz] ASUS Sabertooth X58 AMD Fury X 24GB - 1600Mhz Triple Channel 
Hard DriveHard DriveHard DriveHard Drive
Seagate Barracuda 7200 1TB RAID 0 - B Seagate Barracuda 7200 1TB RAID 0 - B Seagate Barracuda 7200 1TB RAID 0 - C Seagate Barracuda 7200 1TB RAID 0 - C 
Hard DriveHard DriveHard DriveCooling
SSD 128GB RAID - A SSD 128GB RAID - A SSD 256GB  Antec Kuhler H2O 620 [Pull] 
OSMonitorPowerOther
Windows 10 Professional  Dual 24-inch Monitors EVGA SuperNOVA G2 1300W x2 Delta FFB1212EH-F00 Fan 4,000rpm  
Other
x4 Scythe Gentle Typhoon D1225C12BBAP-31 Fan 54... 
  hide details  
Reply
post #727 of 772
I think this is a neurotic defence mechanism. Before Dx12, intrinsic "hardware specific" code was out of the question, as high level application programming interfaces ruled the roost. Now, it is up to the developer to implement higher levels of code complexity, but hey! Let's kill the messenger and pretend this transaction never happened.
Edited by mtcn77 - 7/21/16 at 11:35am
The Machine
(14 items)
 
Nexus 7 2013
(11 items)
 
 
CPUMotherboardGraphicsRAM
A10 6800K Asus F2A85-V MSI 6870 Hawx, VTX3D 5770, AMD HD6950(RIP), Sap... G.skill Ripjaws PC12800 6-8-6-24 
Hard DriveOptical DriveOSMonitor
Seagate 7200.5 1TB NEC 3540 Dvd-Rom Windows 7 x32 Ultimate Samsung P2350 23" 1080p 
PowerCaseMouseAudio
Seasonic s12-600w CoolerMaster Centurion 5 Logitech G600 Auzen X-Fi Raider 
CPUMotherboardGraphicsRAM
Quad Krait 300 at 1.5Ghz Qualcomm APQ8064-1AA SOC Adreno 320 at 400mhz 2GB DDR3L-1600 
Hard DriveOSMonitorKeyboard
32GB Internal NAND Android 5.0 7" 1920X1200 103% sRGB & 572 cd/m2 LTPS IPS Microsoft Wedge Mobile Keyboard 
PowerAudio
3950mAh/15.01mAh Battery Stereo Speakers 
  hide details  
Reply
The Machine
(14 items)
 
Nexus 7 2013
(11 items)
 
 
CPUMotherboardGraphicsRAM
A10 6800K Asus F2A85-V MSI 6870 Hawx, VTX3D 5770, AMD HD6950(RIP), Sap... G.skill Ripjaws PC12800 6-8-6-24 
Hard DriveOptical DriveOSMonitor
Seagate 7200.5 1TB NEC 3540 Dvd-Rom Windows 7 x32 Ultimate Samsung P2350 23" 1080p 
PowerCaseMouseAudio
Seasonic s12-600w CoolerMaster Centurion 5 Logitech G600 Auzen X-Fi Raider 
CPUMotherboardGraphicsRAM
Quad Krait 300 at 1.5Ghz Qualcomm APQ8064-1AA SOC Adreno 320 at 400mhz 2GB DDR3L-1600 
Hard DriveOSMonitorKeyboard
32GB Internal NAND Android 5.0 7" 1920X1200 103% sRGB & 572 cd/m2 LTPS IPS Microsoft Wedge Mobile Keyboard 
PowerAudio
3950mAh/15.01mAh Battery Stereo Speakers 
  hide details  
Reply
post #728 of 772
Quote:
Originally Posted by Kana-Maru View Post

Yeah I did write it. Time Spy has some interesting behavior for sure. I don't think Futuremark wants to talk to me lol. I think they would rather debunk anything I have to say and move on. Good luck trying to get answers out of Futuremark though.

Link to that article?
Big Timmah
(13 items)
 
  
CPUMotherboardGraphicsRAM
Ryzen 5 1600 Asrock x370 Killer SLI/AC Sapphire Radeon Nitro Fury CORSAIR Vengeance LPX 16GB 3200mhz 
Hard DriveCoolingOSMonitor
PNY 480GB SSD PH-TC12DX Black Windows 10 Pro LG 29inch Ultrawide 
KeyboardPowerCaseMouse
Corsair K70 Thermaltake SMART M Series 850W NZXT S340 White Steel ATX Mid Tower Case Wireless Logitech thing 
Mouse Pad
With a supple pad  
  hide details  
Reply
Big Timmah
(13 items)
 
  
CPUMotherboardGraphicsRAM
Ryzen 5 1600 Asrock x370 Killer SLI/AC Sapphire Radeon Nitro Fury CORSAIR Vengeance LPX 16GB 3200mhz 
Hard DriveCoolingOSMonitor
PNY 480GB SSD PH-TC12DX Black Windows 10 Pro LG 29inch Ultrawide 
KeyboardPowerCaseMouse
Corsair K70 Thermaltake SMART M Series 850W NZXT S340 White Steel ATX Mid Tower Case Wireless Logitech thing 
Mouse Pad
With a supple pad  
  hide details  
Reply
post #729 of 772
Quote:
Originally Posted by CriticalHit View Post

im still scratching my head as why we need a benchmark that pretends to be a game in certain conditions.

we have a games that can do that on their own.

here i was thinking benchmarks were to show tech available and to push hardware to the max to see what copes best in which situations. I read Futuremark's report as a giant cop-out. A DX12 feature level 11 bench that doesnt attempt to push hardware, rather just emulates what they think will be the average game over next 2 years is not what i thought i was paying for...

one for the crapware pile.

I will start off by saying that I would also like a version that used full DirectX 12 feature level 12_1, just to see what the effect of supporting the various hardware feature levels was. However, the anger from AMD fans if they did this would be extreme; it would make the current noise look like nothing.

A benchmark that used full 12_1 with "normal" amounts of Async would run absolutely terribly on any AMD GPU available, relative to Pascal, due to the use of Conservative Rasterization Tier 1 and Rasterizer Ordered Views. But then you can go above what is required by 12_1, e.g. with code paths designed around tier 3 Resource binding, and have it unable to run well on nVidia hardware too. If you drop Conservative Rasterization and Rasterizer Ordered Views but keep the tier 3 Resource binding or turn up Async to unreasonable levels (where it hurts GCN more than it helps too) then it would be much faster on newer GCN.

I am also not sure how Async fits into that. Implementing Async seems to be odd for developers, if you have idle hardware when running without Async moving some effects to an Async load is helpful, it can use the idle hardware, but if you can fully use all the hardware without Async it is better without Async, in that case Async simply adds overhead (only a little bit on newer GCN). It seems like if you could perfectly tune a code path and scene for a particular GPU you could get all the performance available without using Async at all, AMD or Nvidia. Of course in the real world this isn't possible so Async helps, but there is a real limit to how much Async should be used. Pushing everything you can into Async and paying no attention is the lazy way out and will result in worse performance.

A DirectX 12 benchmark that "pushes the hardware to the max" isn't reasonable or useful right now, no game uses all those features and no game ever will. We will be stuck with engines using DirectX 12 Feature level 11, as a min requirement, for a while and I do not think we know what will actually be standard after that. You gain most of the benefit of DirectX 12 already with feature level 11, multi-threaded draw calls is the most important feature of DirectX 12 isn't it? We are finally free from the restriction to single threaded performance for gaming (except for AI and similar). Would you want to release a game any time soon that could not run on Kepler, GCN 1.0, or anything Intel before Skylake?
Desktop
(19 items)
 
RAID
(17 items)
 
 
CPUMotherboardGraphicsRAM
Intel i9-7900X @ 4.7GHz Asus ROG Rampage VI Apex Titan X (Pascal) @ 2.05GHz 32GB DDR4 4000-17-17-17-37 CR1 
Hard DriveHard DriveCoolingOS
480GB - Intel Optane 900P 2TB - Samsung 960 Pro EK Monoblock + GPU + 560 Rad Windows 10 Pro x64 
MonitorMonitorMonitorKeyboard
Acer XB270HU bprz HTC Vive LG OLED55C7P Logitech G810 
PowerCaseMouseMouse Pad
Seasonic PRIME 1200 Platinum Old Marble Slab Logitech G900 Logitech G440 
AudioAudioOther
Sennheiser HD 600 Creative SoundBlasterX AE-5 Mellanox ConnectX-3 MCX312A-XCBT 10 GbE Adapter 
CPUMotherboardGraphicsRAM
i7-5960X @ 4.2GHz Asus Rampage 5 Extreme Nvidia GeForce GT 545 32GB DDR4 (2400-12-12-12-28-1T) 
Hard DriveHard DriveHard DriveOptical Drive
Samsung 950 Pro M.2 512GB HGST NAS 4TB x8 - 21.8TB RAID6 Western Digital Black 4TB Samsung SH-S183L 
CoolingOSMonitorKeyboard
Noctua NH-D15 Windows 10 Pro Asus VG278H WASD "CODE" Keyboard 
PowerCaseMouseOther
SeaSonic Platinum-1000 DIYPC Alpha-GT3 Logitech G700s Mellanox ConnectX-3 MCX312A-XCBT 10 GbE Adapter 
Other
Adaptec RAID 71605 
  hide details  
Reply
Desktop
(19 items)
 
RAID
(17 items)
 
 
CPUMotherboardGraphicsRAM
Intel i9-7900X @ 4.7GHz Asus ROG Rampage VI Apex Titan X (Pascal) @ 2.05GHz 32GB DDR4 4000-17-17-17-37 CR1 
Hard DriveHard DriveCoolingOS
480GB - Intel Optane 900P 2TB - Samsung 960 Pro EK Monoblock + GPU + 560 Rad Windows 10 Pro x64 
MonitorMonitorMonitorKeyboard
Acer XB270HU bprz HTC Vive LG OLED55C7P Logitech G810 
PowerCaseMouseMouse Pad
Seasonic PRIME 1200 Platinum Old Marble Slab Logitech G900 Logitech G440 
AudioAudioOther
Sennheiser HD 600 Creative SoundBlasterX AE-5 Mellanox ConnectX-3 MCX312A-XCBT 10 GbE Adapter 
CPUMotherboardGraphicsRAM
i7-5960X @ 4.2GHz Asus Rampage 5 Extreme Nvidia GeForce GT 545 32GB DDR4 (2400-12-12-12-28-1T) 
Hard DriveHard DriveHard DriveOptical Drive
Samsung 950 Pro M.2 512GB HGST NAS 4TB x8 - 21.8TB RAID6 Western Digital Black 4TB Samsung SH-S183L 
CoolingOSMonitorKeyboard
Noctua NH-D15 Windows 10 Pro Asus VG278H WASD "CODE" Keyboard 
PowerCaseMouseOther
SeaSonic Platinum-1000 DIYPC Alpha-GT3 Logitech G700s Mellanox ConnectX-3 MCX312A-XCBT 10 GbE Adapter 
Other
Adaptec RAID 71605 
  hide details  
Reply
post #730 of 772
You can emulate all the 12_1 features if you already support 12_0.

And actually the specs for the 12_1 are not the finals one. It was a pure marketing move from nvidia.

The most important are the hardware features that dx12 requires.
Workstation
(4 items)
 
  
CPUMotherboardGraphicsMonitor
Xeon E5-2690 Supermicro 2011 Nvidia GP100/ Vega FE Dell ultrasharp 4k 
  hide details  
Reply
Workstation
(4 items)
 
  
CPUMotherboardGraphicsMonitor
Xeon E5-2690 Supermicro 2011 Nvidia GP100/ Vega FE Dell ultrasharp 4k 
  hide details  
Reply
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Hardware News
Overclock.net › Forums › Industry News › Hardware News › [Various] Futuremark Releases 3DMark Time Spy DirectX 12 Benchmark