Overclock.net › Forums › Industry News › Hardware News › [AdoredTV] Pascal vs Maxwell at same clockspeeds, same FLOPS (1080 vs 980 Ti)
New Posts  All Forums:Forum Nav:

[AdoredTV] Pascal vs Maxwell at same clockspeeds, same FLOPS (1080 vs 980 Ti) - Page 29

post #281 of 305
Quote:
Originally Posted by Echoa View Post

AFAIK most AMD gpu haven't been bandwidth starved for years. They've been choked by weak front end, geometry processing, ROP and cache amounts. Fiji's biggest issue was weak front end, ROPs , and limited cache, hopefully Vega resolves those issues. GCN shaders themselves have the horsepower but AMD pairs them with a front end that can't keep them well fed even with boat loads of bandwidth then squeezes everything out through a thin 64rop straw.

GCN hasn't needed much change on the shader end, it needs a front end that can keep up which it still doesn't have
polaris will face the same issue as well. weak 32 Rop @ low 1266MHz clocks @ gtx970/980 performance or 390/390x performance with both of them have 64Rop.
Gaming Rig
(10 items)
 
Work/Web Rig
(11 items)
 
Web Rig
(8 items)
 
CPUMotherboardGraphicsRAM
i5-2500K @4GHz stock volt OC Asus P8Z77-V LK Palit GTX 750Ti StormX Dual Corsair Vengence 1600MHz_CL8 4GBx4 
Hard DriveHard DriveCoolingOS
Seagate 80GB 7200rpm Maxtor 250GB 7200rpm Cooler Master Hyper 212+ Windows 10 Pro 
MonitorCase
BenQ XL2720z @ 1920x1080 144Hz Silverstone Ps06 
CPUMotherboardGraphicsRAM
Core 2 Quad Q9650 Gigabyte EP41-UD3L Asus 1GB Radeon 7790 DirectCU OC Corsair Gaming Ram 2GBx2 DDR2-800 
Hard DriveHard DriveCoolingOS
Samsung 80GB 7200rpm Western Digital 200GB 7200rpm Cooler Master Hyper TX3 Windows 7 pro 
MonitorPowerCase
Samsung 226bw 1680x1050 Acbel iPower 510w (450w PSU) LianLi PS05 
CPUMotherboardGraphicsRAM
AMD E350 MSI E350IA-E45 AMD Radeon 6310 4GB Kingston DDR3 1333 
Hard DriveOSMonitorPower
Western Digital 80GB Linux Slackware 14.1 Samsung 22" 1680x1050 LCD Acbel 300w PSU 
  hide details  
Reply
Gaming Rig
(10 items)
 
Work/Web Rig
(11 items)
 
Web Rig
(8 items)
 
CPUMotherboardGraphicsRAM
i5-2500K @4GHz stock volt OC Asus P8Z77-V LK Palit GTX 750Ti StormX Dual Corsair Vengence 1600MHz_CL8 4GBx4 
Hard DriveHard DriveCoolingOS
Seagate 80GB 7200rpm Maxtor 250GB 7200rpm Cooler Master Hyper 212+ Windows 10 Pro 
MonitorCase
BenQ XL2720z @ 1920x1080 144Hz Silverstone Ps06 
CPUMotherboardGraphicsRAM
Core 2 Quad Q9650 Gigabyte EP41-UD3L Asus 1GB Radeon 7790 DirectCU OC Corsair Gaming Ram 2GBx2 DDR2-800 
Hard DriveHard DriveCoolingOS
Samsung 80GB 7200rpm Western Digital 200GB 7200rpm Cooler Master Hyper TX3 Windows 7 pro 
MonitorPowerCase
Samsung 226bw 1680x1050 Acbel iPower 510w (450w PSU) LianLi PS05 
CPUMotherboardGraphicsRAM
AMD E350 MSI E350IA-E45 AMD Radeon 6310 4GB Kingston DDR3 1333 
Hard DriveOSMonitorPower
Western Digital 80GB Linux Slackware 14.1 Samsung 22" 1680x1050 LCD Acbel 300w PSU 
  hide details  
Reply
post #282 of 305
Quote:
Originally Posted by Kinaesthetic View Post

GCN 1.1 (aka generation 2, Hawaii) vs GCN 1.2 (aka generation 3, Fiji). Delta color compression would be a significant reason why. It was primarily introduced in hardware with GCN 1.2. Also explains why the RX 480 isn't really bandwidth starved at all either.

Doesn't really have a single thing to do with HBM.

So you say, but again you can't prove that.
post #283 of 305
Quote:
Originally Posted by Majin SSJ Eric View Post


So you say, but again you can't prove that.

 

Arguing with you is like talking to a brick wall. Dude, go do some research on your own to find the answer to your questions. Every time I get into a discussion with you, you do jack all of your own research, and consistently put down well researched arguments or arguments from people who just flat out know more than you do. Yes, I've snapped because this gets ridiculous on this forum.

 

For the last time I'll ever do this (any amount of easily Google'able searches) for you:

 

http://gpuopen.com/dcc-overview/

 

https://db.thegpu.guru/architecture/GCN/1.2

 

 

Pitifully easy Google search.

 

Also, it should be complete common sense by now if you have even remotely been paying attention to the pitifully small bus widths on Maxwell and Pascal GPUs, yet they still keep up with significantly higher bus width  (similarly clocked VRAM modules) GPUs from either the previous generations or AMD.

post #284 of 305
Quote:
Originally Posted by Clocknut View Post

polaris will face the same issue as well. weak 32 Rop @ low 1266MHz clocks @ gtx970/980 performance or 390/390x performance with both of them have 64Rop.

They actually improved the ROP and front end with Polaris but we have yet to see how this will translate on the high end. Polaris manages to hit 500 shaders, 256bit bus, and 32rops above its weight class which is a good improvement so far.
Toxic-DT
(19 items)
 
Toxic-LT
(12 items)
 
Toxic-SV
(10 items)
 
CPUMotherboardGraphicsRAM
i7 4770k Gigabyte Z87X-UD4H Gigabyte GTX 1060 Gskill Ares 
Hard DriveHard DriveHard DriveOptical Drive
Mushkin Eco3 Kingdian  Corsair LE  Lite On 
CoolingOSMonitorKeyboard
Cryorig H5 Windows 10 pro Dell S2415h Azio MK Retro 
PowerCaseMouseAudio
Seasonic G Series 650w Define R5  Logitech G500s Asus Xonar DSX 
AudioAudioAudio
Monoprice Retro Heaphones Monoprice Vocal Dynamic Mic Edifier r980t  
CPUMotherboardGraphicsRAM
I7 920xm Asus g73jh Radeon HD 6990m 8gb Hynix DDR3 1333 
Hard DriveHard DriveCoolingOS
Seagate Laptop Thin 500gb Corsair Force LE  Stock G73jh Windows 10 
MonitorKeyboardPowerMouse
1600x900 Glossy G73jh Screen G73jh Backlit Keyboard Delta 150w PSU Verbatim Laptop mouse 
CPUMotherboardGraphicsRAM
Xeon e5649 HP z400 FirePro v4800 12gb Gkill Ripjaw 
Hard DriveHard DriveCoolingOS
Hitachi Ultrastar Western Digital 320gb HP High Performance Cooler Ubuntu Gnome 16.10 
PowerCase
HP z400 HP z400 
  hide details  
Reply
Toxic-DT
(19 items)
 
Toxic-LT
(12 items)
 
Toxic-SV
(10 items)
 
CPUMotherboardGraphicsRAM
i7 4770k Gigabyte Z87X-UD4H Gigabyte GTX 1060 Gskill Ares 
Hard DriveHard DriveHard DriveOptical Drive
Mushkin Eco3 Kingdian  Corsair LE  Lite On 
CoolingOSMonitorKeyboard
Cryorig H5 Windows 10 pro Dell S2415h Azio MK Retro 
PowerCaseMouseAudio
Seasonic G Series 650w Define R5  Logitech G500s Asus Xonar DSX 
AudioAudioAudio
Monoprice Retro Heaphones Monoprice Vocal Dynamic Mic Edifier r980t  
CPUMotherboardGraphicsRAM
I7 920xm Asus g73jh Radeon HD 6990m 8gb Hynix DDR3 1333 
Hard DriveHard DriveCoolingOS
Seagate Laptop Thin 500gb Corsair Force LE  Stock G73jh Windows 10 
MonitorKeyboardPowerMouse
1600x900 Glossy G73jh Screen G73jh Backlit Keyboard Delta 150w PSU Verbatim Laptop mouse 
CPUMotherboardGraphicsRAM
Xeon e5649 HP z400 FirePro v4800 12gb Gkill Ripjaw 
Hard DriveHard DriveCoolingOS
Hitachi Ultrastar Western Digital 320gb HP High Performance Cooler Ubuntu Gnome 16.10 
PowerCase
HP z400 HP z400 
  hide details  
Reply
post #285 of 305
Quote:
Originally Posted by Kinaesthetic View Post


Pitifully easy Google search.

Also, it should be complete common sense by now if you have even remotely been paying attention to the pitifully small bus widths on Maxwell and Pascal GPUs, yet they still keep up with significantly higher bus width  (similarly clocked VRAM modules) GPUs from either the previous generations or AMD.

Funny thing is that Maxwell and Pascal are actually bandwidth starved and can't fully utilize the ROPs and only manage around 80-90% their theoretical filtrate (they still keep up with gcn which has the opposite issue). Pascal would likely perform better with more bandwidth

Pic doesn't include Pascal bit illustrates what I mean



With adequate bandwidth they should be hitting 85/95/103 on the 980/ti/titan respectively but don't. HBM would be a huge boon for nvidia right now
Edited by Echoa - 7/30/16 at 11:36pm
Toxic-DT
(19 items)
 
Toxic-LT
(12 items)
 
Toxic-SV
(10 items)
 
CPUMotherboardGraphicsRAM
i7 4770k Gigabyte Z87X-UD4H Gigabyte GTX 1060 Gskill Ares 
Hard DriveHard DriveHard DriveOptical Drive
Mushkin Eco3 Kingdian  Corsair LE  Lite On 
CoolingOSMonitorKeyboard
Cryorig H5 Windows 10 pro Dell S2415h Azio MK Retro 
PowerCaseMouseAudio
Seasonic G Series 650w Define R5  Logitech G500s Asus Xonar DSX 
AudioAudioAudio
Monoprice Retro Heaphones Monoprice Vocal Dynamic Mic Edifier r980t  
CPUMotherboardGraphicsRAM
I7 920xm Asus g73jh Radeon HD 6990m 8gb Hynix DDR3 1333 
Hard DriveHard DriveCoolingOS
Seagate Laptop Thin 500gb Corsair Force LE  Stock G73jh Windows 10 
MonitorKeyboardPowerMouse
1600x900 Glossy G73jh Screen G73jh Backlit Keyboard Delta 150w PSU Verbatim Laptop mouse 
CPUMotherboardGraphicsRAM
Xeon e5649 HP z400 FirePro v4800 12gb Gkill Ripjaw 
Hard DriveHard DriveCoolingOS
Hitachi Ultrastar Western Digital 320gb HP High Performance Cooler Ubuntu Gnome 16.10 
PowerCase
HP z400 HP z400 
  hide details  
Reply
Toxic-DT
(19 items)
 
Toxic-LT
(12 items)
 
Toxic-SV
(10 items)
 
CPUMotherboardGraphicsRAM
i7 4770k Gigabyte Z87X-UD4H Gigabyte GTX 1060 Gskill Ares 
Hard DriveHard DriveHard DriveOptical Drive
Mushkin Eco3 Kingdian  Corsair LE  Lite On 
CoolingOSMonitorKeyboard
Cryorig H5 Windows 10 pro Dell S2415h Azio MK Retro 
PowerCaseMouseAudio
Seasonic G Series 650w Define R5  Logitech G500s Asus Xonar DSX 
AudioAudioAudio
Monoprice Retro Heaphones Monoprice Vocal Dynamic Mic Edifier r980t  
CPUMotherboardGraphicsRAM
I7 920xm Asus g73jh Radeon HD 6990m 8gb Hynix DDR3 1333 
Hard DriveHard DriveCoolingOS
Seagate Laptop Thin 500gb Corsair Force LE  Stock G73jh Windows 10 
MonitorKeyboardPowerMouse
1600x900 Glossy G73jh Screen G73jh Backlit Keyboard Delta 150w PSU Verbatim Laptop mouse 
CPUMotherboardGraphicsRAM
Xeon e5649 HP z400 FirePro v4800 12gb Gkill Ripjaw 
Hard DriveHard DriveCoolingOS
Hitachi Ultrastar Western Digital 320gb HP High Performance Cooler Ubuntu Gnome 16.10 
PowerCase
HP z400 HP z400 
  hide details  
Reply
post #286 of 305
Quote:
Originally Posted by Echoa View Post


Funny thing is that Maxwell and Pascal are actually bandwidth starved and can't fully utilize the ROPs and only manage around 80-90% their theoretical filtrate (they still keep up with gcn which has the opposite issue). Pascal would likely perform better with more bandwidth

 

I never said they weren't bandwidth starved. I'm just trying to point out that despite exceptionally low bandwidth, they are still competitive due to DCC amongst other compression techniques in hardware. I entirely agree with you though that they are bandwidth starved to begin with.

post #287 of 305
When you guys say "bandwidth", do you mean "memory bandwidth"? Because Pascal (or GP104 at least) barely responds to memory bandwidth improvements:



http://www.anandtech.com/show/10325/the-nvidia-geforce-gtx-1080-and-1070-founders-edition-review/31
post #288 of 305
Quote:
Originally Posted by Kinaesthetic View Post

I never said they weren't bandwidth starved. I'm just trying to point out that despite exceptionally low bandwidth, they are still competitive due to DCC amongst other compression techniques in hardware. I entirely agree with you though that they are bandwidth starved to begin with.

I know you didn't I was just adding to your talk of bus width and mine of AMD not being bandwidth starved

The humor in competition having opposing issues
Toxic-DT
(19 items)
 
Toxic-LT
(12 items)
 
Toxic-SV
(10 items)
 
CPUMotherboardGraphicsRAM
i7 4770k Gigabyte Z87X-UD4H Gigabyte GTX 1060 Gskill Ares 
Hard DriveHard DriveHard DriveOptical Drive
Mushkin Eco3 Kingdian  Corsair LE  Lite On 
CoolingOSMonitorKeyboard
Cryorig H5 Windows 10 pro Dell S2415h Azio MK Retro 
PowerCaseMouseAudio
Seasonic G Series 650w Define R5  Logitech G500s Asus Xonar DSX 
AudioAudioAudio
Monoprice Retro Heaphones Monoprice Vocal Dynamic Mic Edifier r980t  
CPUMotherboardGraphicsRAM
I7 920xm Asus g73jh Radeon HD 6990m 8gb Hynix DDR3 1333 
Hard DriveHard DriveCoolingOS
Seagate Laptop Thin 500gb Corsair Force LE  Stock G73jh Windows 10 
MonitorKeyboardPowerMouse
1600x900 Glossy G73jh Screen G73jh Backlit Keyboard Delta 150w PSU Verbatim Laptop mouse 
CPUMotherboardGraphicsRAM
Xeon e5649 HP z400 FirePro v4800 12gb Gkill Ripjaw 
Hard DriveHard DriveCoolingOS
Hitachi Ultrastar Western Digital 320gb HP High Performance Cooler Ubuntu Gnome 16.10 
PowerCase
HP z400 HP z400 
  hide details  
Reply
Toxic-DT
(19 items)
 
Toxic-LT
(12 items)
 
Toxic-SV
(10 items)
 
CPUMotherboardGraphicsRAM
i7 4770k Gigabyte Z87X-UD4H Gigabyte GTX 1060 Gskill Ares 
Hard DriveHard DriveHard DriveOptical Drive
Mushkin Eco3 Kingdian  Corsair LE  Lite On 
CoolingOSMonitorKeyboard
Cryorig H5 Windows 10 pro Dell S2415h Azio MK Retro 
PowerCaseMouseAudio
Seasonic G Series 650w Define R5  Logitech G500s Asus Xonar DSX 
AudioAudioAudio
Monoprice Retro Heaphones Monoprice Vocal Dynamic Mic Edifier r980t  
CPUMotherboardGraphicsRAM
I7 920xm Asus g73jh Radeon HD 6990m 8gb Hynix DDR3 1333 
Hard DriveHard DriveCoolingOS
Seagate Laptop Thin 500gb Corsair Force LE  Stock G73jh Windows 10 
MonitorKeyboardPowerMouse
1600x900 Glossy G73jh Screen G73jh Backlit Keyboard Delta 150w PSU Verbatim Laptop mouse 
CPUMotherboardGraphicsRAM
Xeon e5649 HP z400 FirePro v4800 12gb Gkill Ripjaw 
Hard DriveHard DriveCoolingOS
Hitachi Ultrastar Western Digital 320gb HP High Performance Cooler Ubuntu Gnome 16.10 
PowerCase
HP z400 HP z400 
  hide details  
Reply
post #289 of 305
Quote:
Originally Posted by magnek View Post

When you guys say "bandwidth", do you mean "memory bandwidth"? Because Pascal (or GP104 at least) barely responds to memory bandwidth improvements:



http://www.anandtech.com/show/10325/the-nvidia-geforce-gtx-1080-and-1070-founders-edition-review/31

It's not just from memory, it's the memory controller and bandwidth throughout the entire arch including cache distribution, etc.

Nvidia has some real brute power but isn't getting data around it fast enough, and there's periods where Pascal is actually doing nothing because they don't keep the pipeline filled, it's just so fast that it doesn't matter vs AMDs issue until you hit loads that GCN can better show is muscle (async for example)

GCN kinda works like hyperthreading by filling in otherwise idle time with other stuff. Problem is you need a beefy front end to do it fast enough and enough cache to handle it right and is why even though GCN and more specfifically Fiji has more bandwidth it ends up useless and and only utilizes maybe 60-70% the available bandwidth it has. AMDs improved that some in Polaris, but Pascal ends up waiting for things to do more frequently than Fiji for example but Fiji ends up waiting when the front end is trying to cram enough data down it to keep all the shaders occupied and then the data cant be pushed to your screen fast enough either. Fiji wouldve been far better on a 14nm node because it ran into a die size restriction and they had to make sacrifices. Pascal and Maxwell have good front end and ROP power but simple cant fill it because of bandwidth restrictions but it has so much brute force it still out paces GCN in work that GCN doesnt excell at (i.e dx11)
Edited by Echoa - 7/31/16 at 12:11am
Toxic-DT
(19 items)
 
Toxic-LT
(12 items)
 
Toxic-SV
(10 items)
 
CPUMotherboardGraphicsRAM
i7 4770k Gigabyte Z87X-UD4H Gigabyte GTX 1060 Gskill Ares 
Hard DriveHard DriveHard DriveOptical Drive
Mushkin Eco3 Kingdian  Corsair LE  Lite On 
CoolingOSMonitorKeyboard
Cryorig H5 Windows 10 pro Dell S2415h Azio MK Retro 
PowerCaseMouseAudio
Seasonic G Series 650w Define R5  Logitech G500s Asus Xonar DSX 
AudioAudioAudio
Monoprice Retro Heaphones Monoprice Vocal Dynamic Mic Edifier r980t  
CPUMotherboardGraphicsRAM
I7 920xm Asus g73jh Radeon HD 6990m 8gb Hynix DDR3 1333 
Hard DriveHard DriveCoolingOS
Seagate Laptop Thin 500gb Corsair Force LE  Stock G73jh Windows 10 
MonitorKeyboardPowerMouse
1600x900 Glossy G73jh Screen G73jh Backlit Keyboard Delta 150w PSU Verbatim Laptop mouse 
CPUMotherboardGraphicsRAM
Xeon e5649 HP z400 FirePro v4800 12gb Gkill Ripjaw 
Hard DriveHard DriveCoolingOS
Hitachi Ultrastar Western Digital 320gb HP High Performance Cooler Ubuntu Gnome 16.10 
PowerCase
HP z400 HP z400 
  hide details  
Reply
Toxic-DT
(19 items)
 
Toxic-LT
(12 items)
 
Toxic-SV
(10 items)
 
CPUMotherboardGraphicsRAM
i7 4770k Gigabyte Z87X-UD4H Gigabyte GTX 1060 Gskill Ares 
Hard DriveHard DriveHard DriveOptical Drive
Mushkin Eco3 Kingdian  Corsair LE  Lite On 
CoolingOSMonitorKeyboard
Cryorig H5 Windows 10 pro Dell S2415h Azio MK Retro 
PowerCaseMouseAudio
Seasonic G Series 650w Define R5  Logitech G500s Asus Xonar DSX 
AudioAudioAudio
Monoprice Retro Heaphones Monoprice Vocal Dynamic Mic Edifier r980t  
CPUMotherboardGraphicsRAM
I7 920xm Asus g73jh Radeon HD 6990m 8gb Hynix DDR3 1333 
Hard DriveHard DriveCoolingOS
Seagate Laptop Thin 500gb Corsair Force LE  Stock G73jh Windows 10 
MonitorKeyboardPowerMouse
1600x900 Glossy G73jh Screen G73jh Backlit Keyboard Delta 150w PSU Verbatim Laptop mouse 
CPUMotherboardGraphicsRAM
Xeon e5649 HP z400 FirePro v4800 12gb Gkill Ripjaw 
Hard DriveHard DriveCoolingOS
Hitachi Ultrastar Western Digital 320gb HP High Performance Cooler Ubuntu Gnome 16.10 
PowerCase
HP z400 HP z400 
  hide details  
Reply
post #290 of 305
I'm starting to learn a few things in this thread after all. I forgot that "67,200 MPixel/s" meant '67.2 Gigapixels per second'. According to that, GCN is hitting 95-100% of the theoretically maximum fill rate capacity. Bandwidth utilization to the same extent, however, is nowhere to be seen. The theoretical utilisation rate between 290X(DC II Top spec) and Fury X is dropping from 76% to 65% respectively. It could relate to rising losses from increased magnitude, however wasn't HBM expected to avert that trend somewhat?
Clearly, each gpu shows higher utilisation at the step where the bottleneck is present. So far, I think all convictions that GCN is fillrate and geometry limited while Geforce is bandwidth limited are kind of vindicated looking at the available data(which is awesome by the way).
[Source]
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
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Hardware News
Overclock.net › Forums › Industry News › Hardware News › [AdoredTV] Pascal vs Maxwell at same clockspeeds, same FLOPS (1080 vs 980 Ti)