Overclock.net › Forums › Industry News › Hardware News › [Computerbase] - 3gb vs 4gb vs 6gb vs 8gb GDDR5 VRAM Frametime Testing
New Posts  All Forums:Forum Nav:

[Computerbase] - 3gb vs 4gb vs 6gb vs 8gb GDDR5 VRAM Frametime Testing - Page 15

post #141 of 175
Quote:
Originally Posted by Piraal View Post

It all started when someone said there is evidence because of HBM's wide interface, it has much lower latencies,which gave the card more VRAM headroom.(paraphrasing) I asked if it was true(not really), half expecting to see said evidence, but it turned out there isn't any evidence at all.

A wide interface doesn't imply lower latency. In general, higher frequency means lower latency. HBM is low latency, relatively speaking, because of architectural and timing changes, as well as the memory's physical proximity to the GPU. However, neither latency, nor bandwidth, have much of anything to do with how VRAM capacity can be utilized.

Primary assets loaded into VRAM are largely static; they are not removed, cleared, or freed up, until they are no longer needed and they are needed at least until no frames being rendered use them. If they don't fit in local VRAM, they have to be fetched, repeatedly, from much slower subsystems and it's those subsystems that quickly become a bottleneck.

Imagine someone painting a picture. They are the GPU and their pallet is their VRAM. Neither reducing the time it takes to move the brush from the pallet to the canvas (latency), nor the size of the brush (width), will help them if the pallet they are holding doesn't have the paint they need. If that pallet is too small to hold every color that needs to be mixed, part of it will need to be cleared off then other paints will need to be found and added to the pallet...then, if the colors originally replaced need to be used again, the same process must be repeated. With even a few missing colors, work that doesn't involve moving mixing colors or moving the brush from pallet to canvas (all that VRAM can help with) quickly becomes the majority of time needed to paint the picture. The number of pictures you can paint in a given period of time tanks, no matter how fast you are at actually painting, because you spend most of your time running down the hall to dig out new paints (asking the CPU, across PCI-E, to deliver assets in system memory, or on a drive, to you so you can put them in your VRAM) or to find ones you were forced to put back before you were done with them (evicted assets).
Primary
(15 items)
 
Secondary
(13 items)
 
In progress
(10 items)
 
CPUMotherboardGraphicsRAM
5820K @ 4.2/3.5GHz core/uncore, 1.175/1.15v Gigabyte X99 SOC Champion (F22n) Gigabyte AORUS GTX 1080 Ti (F3P) @ 2025/1485, 1... 4x4GiB Crucial @ 2667, 12-12-12-28-T1, 1.34v 
Hard DriveHard DriveHard DriveCooling
Plextor M6e 128GB (fw 1.06) M.2 (PCI-E 2.0 2x) 2x Crucial M4 256GB 4x WD Scorpio Black 500GB Noctua NH-D15 
OSMonitorKeyboardPower
Windows 7 Professional x64 SP1 BenQ BL3200PT Filco Majestouch Tenkeyless (MX Brown) Corsair RM1000x 
CaseMouseAudio
Fractal Design Define R4 Logitech G402 Realtek ALC1150 + M-Audio AV40 
CPUMotherboardGraphicsRAM
X5670 @ 4.4/3.2GHz core/uncore, 1.36 vcore, 1.2... Gigabyte X58A-UD5 r2.0 w/FF3mod10 BIOS Sapphire Fury Nitro OC+ @ 1053/500, 1.225vGPU/1... 2x Samsung MV-3V4G3D/US @ 2000, 10-11-11-30-T1,... 
RAMHard DriveHard DriveHard Drive
1x Crucial BLT4G3D1608ET3LX0 @ 2000, 10-11-11-3... OCZ (Toshiba) Trion 150 120GB Hyundai Sapphire 120GB 3x Hitachi Deskstar 7k1000.C 1TB 
CoolingOSPowerCase
Noctua NH-D14 Windows 7 Pro x64 SP1 Antec TP-750 Fractal Design R5 
Audio
ASUS Xonar DS 
CPUMotherboardGraphicsRAM
i7-6800K @ 4.3/3.5GHz core/uncore, 1.36/1.2v ASRock X99 OC Formula (P3.10) GTX 780 (temporary) 4x4GiB Crucial DDR4-2400 @ 11-13-12-28-T2, 1.33v 
Hard DriveHard DriveCoolingOS
Intel 600p 256GB NVMe 2x HGST Travelstar 7k1000 1TB Corsair H55 (temporary) Windows Server 2016 Datacenter 
PowerCase
Seasonic SS-860XP2 Corsair Carbide Air 540 
  hide details  
Reply
Primary
(15 items)
 
Secondary
(13 items)
 
In progress
(10 items)
 
CPUMotherboardGraphicsRAM
5820K @ 4.2/3.5GHz core/uncore, 1.175/1.15v Gigabyte X99 SOC Champion (F22n) Gigabyte AORUS GTX 1080 Ti (F3P) @ 2025/1485, 1... 4x4GiB Crucial @ 2667, 12-12-12-28-T1, 1.34v 
Hard DriveHard DriveHard DriveCooling
Plextor M6e 128GB (fw 1.06) M.2 (PCI-E 2.0 2x) 2x Crucial M4 256GB 4x WD Scorpio Black 500GB Noctua NH-D15 
OSMonitorKeyboardPower
Windows 7 Professional x64 SP1 BenQ BL3200PT Filco Majestouch Tenkeyless (MX Brown) Corsair RM1000x 
CaseMouseAudio
Fractal Design Define R4 Logitech G402 Realtek ALC1150 + M-Audio AV40 
CPUMotherboardGraphicsRAM
X5670 @ 4.4/3.2GHz core/uncore, 1.36 vcore, 1.2... Gigabyte X58A-UD5 r2.0 w/FF3mod10 BIOS Sapphire Fury Nitro OC+ @ 1053/500, 1.225vGPU/1... 2x Samsung MV-3V4G3D/US @ 2000, 10-11-11-30-T1,... 
RAMHard DriveHard DriveHard Drive
1x Crucial BLT4G3D1608ET3LX0 @ 2000, 10-11-11-3... OCZ (Toshiba) Trion 150 120GB Hyundai Sapphire 120GB 3x Hitachi Deskstar 7k1000.C 1TB 
CoolingOSPowerCase
Noctua NH-D14 Windows 7 Pro x64 SP1 Antec TP-750 Fractal Design R5 
Audio
ASUS Xonar DS 
CPUMotherboardGraphicsRAM
i7-6800K @ 4.3/3.5GHz core/uncore, 1.36/1.2v ASRock X99 OC Formula (P3.10) GTX 780 (temporary) 4x4GiB Crucial DDR4-2400 @ 11-13-12-28-T2, 1.33v 
Hard DriveHard DriveCoolingOS
Intel 600p 256GB NVMe 2x HGST Travelstar 7k1000 1TB Corsair H55 (temporary) Windows Server 2016 Datacenter 
PowerCase
Seasonic SS-860XP2 Corsair Carbide Air 540 
  hide details  
Reply
post #142 of 175
Quote:
Originally Posted by Piraal View Post

It all started when someone said there is evidence because of HBM's wide interface, it has much lower latencies,which gave the card more VRAM headroom.(paraphrasing) I asked if it was true(not really), half expecting to see said evidence, but it turned out there isn't any evidence at all.

You dont see evidence because you dont look.

https://youtu.be/xp5lhCL4CFg

https://youtu.be/MrsZFJ6Ivy0

https://youtu.be/qH5_gBf4MUw

https://youtu.be/qN550loG-hY

https://youtu.be/KK9PuF9-Qoo

https://youtu.be/yfd9sNoLDfs

Big difference between hbm And ddr5, observe the vram usage. Also can pull up direct 4gb ddr5 vs 4gb hbm vram numbers
Edited by Tgrove - 9/26/16 at 6:06pm
Yokai Onmyoji
(18 items)
 
  
CPUMotherboardGraphicsRAM
I7-4930k @ 4.5GHz ASUS P9X79 Deluxe  Sapphire RX Vega 64 Liquid  G. Skill Ripjaws Z 16GB (4x4) 2400 Mhz  
Hard DriveHard DriveOptical DriveCooling
Samsung 840 Pro 256GB (OS) Samsung 840 Pro 512GB (Games) Lite-on Corsair H100 
OSMonitorKeyboardPower
Windows 7 Ultimate Wasabi Mango UHD490 IPS 49" Freesync 33-60hz Corsair K90 Corsair Ax1200i 
CaseMouseMouse PadAudio
Corsair 500r Razer Naga Dx Racer Arm wrest  Asus Xonar DX 
AudioOther
Altec Lansing 5.1 Razer Nostromo 
  hide details  
Reply
Yokai Onmyoji
(18 items)
 
  
CPUMotherboardGraphicsRAM
I7-4930k @ 4.5GHz ASUS P9X79 Deluxe  Sapphire RX Vega 64 Liquid  G. Skill Ripjaws Z 16GB (4x4) 2400 Mhz  
Hard DriveHard DriveOptical DriveCooling
Samsung 840 Pro 256GB (OS) Samsung 840 Pro 512GB (Games) Lite-on Corsair H100 
OSMonitorKeyboardPower
Windows 7 Ultimate Wasabi Mango UHD490 IPS 49" Freesync 33-60hz Corsair K90 Corsair Ax1200i 
CaseMouseMouse PadAudio
Corsair 500r Razer Naga Dx Racer Arm wrest  Asus Xonar DX 
AudioOther
Altec Lansing 5.1 Razer Nostromo 
  hide details  
Reply
post #143 of 175
Quote:
Originally Posted by Tgrove View Post

You dont see evidence because you dont look.

https://youtu.be/xp5lhCL4CFg

https://youtu.be/MrsZFJ6Ivy0

https://youtu.be/qH5_gBf4MUw

https://youtu.be/qN550loG-hY

https://youtu.be/KK9PuF9-Qoo

https://youtu.be/yfd9sNoLDfs

Big difference between hbm And ddr5, observe the vram usage.

None of that is remotely proof of what you're claiming. It's a fallacy to assume that memory type is the cause of the VRAM usage differences here.

It stands to reason that the part with less VRAM would be more aggressive in evicting assets.
Quote:
Originally Posted by Tgrove View Post

Also can pull up direct 4gb ddr5 vs 4gb hbm vram numbers

And I can show different uses in the same task among two GDDR5 cards, even if they have the same amount of physical VRAM, and even if they are of the same brand. Hell, I can show a difference in many tasks from one run to another on the same system, hardware, and software environment. Figures for allocated VRAM don't show what that VRAM is being used for, nor if it's strictly necessary to have that much allocated.

Only a performance comparison in a known VRAM limited scenario where a part with HBM is seeing a substantially lower relative performance impact than a GDDR5 part with the same or greater amount of VRAM would imply HBM is somehow more efficient at VRAM utilization. To the best of my knowledge, no one has ever demonstrated any such thing, and probably for good reason (as the idea of HBM providing an intrinsic advantage in utilization and not simply performance, flies in the face of logic).
Primary
(15 items)
 
Secondary
(13 items)
 
In progress
(10 items)
 
CPUMotherboardGraphicsRAM
5820K @ 4.2/3.5GHz core/uncore, 1.175/1.15v Gigabyte X99 SOC Champion (F22n) Gigabyte AORUS GTX 1080 Ti (F3P) @ 2025/1485, 1... 4x4GiB Crucial @ 2667, 12-12-12-28-T1, 1.34v 
Hard DriveHard DriveHard DriveCooling
Plextor M6e 128GB (fw 1.06) M.2 (PCI-E 2.0 2x) 2x Crucial M4 256GB 4x WD Scorpio Black 500GB Noctua NH-D15 
OSMonitorKeyboardPower
Windows 7 Professional x64 SP1 BenQ BL3200PT Filco Majestouch Tenkeyless (MX Brown) Corsair RM1000x 
CaseMouseAudio
Fractal Design Define R4 Logitech G402 Realtek ALC1150 + M-Audio AV40 
CPUMotherboardGraphicsRAM
X5670 @ 4.4/3.2GHz core/uncore, 1.36 vcore, 1.2... Gigabyte X58A-UD5 r2.0 w/FF3mod10 BIOS Sapphire Fury Nitro OC+ @ 1053/500, 1.225vGPU/1... 2x Samsung MV-3V4G3D/US @ 2000, 10-11-11-30-T1,... 
RAMHard DriveHard DriveHard Drive
1x Crucial BLT4G3D1608ET3LX0 @ 2000, 10-11-11-3... OCZ (Toshiba) Trion 150 120GB Hyundai Sapphire 120GB 3x Hitachi Deskstar 7k1000.C 1TB 
CoolingOSPowerCase
Noctua NH-D14 Windows 7 Pro x64 SP1 Antec TP-750 Fractal Design R5 
Audio
ASUS Xonar DS 
CPUMotherboardGraphicsRAM
i7-6800K @ 4.3/3.5GHz core/uncore, 1.36/1.2v ASRock X99 OC Formula (P3.10) GTX 780 (temporary) 4x4GiB Crucial DDR4-2400 @ 11-13-12-28-T2, 1.33v 
Hard DriveHard DriveCoolingOS
Intel 600p 256GB NVMe 2x HGST Travelstar 7k1000 1TB Corsair H55 (temporary) Windows Server 2016 Datacenter 
PowerCase
Seasonic SS-860XP2 Corsair Carbide Air 540 
  hide details  
Reply
Primary
(15 items)
 
Secondary
(13 items)
 
In progress
(10 items)
 
CPUMotherboardGraphicsRAM
5820K @ 4.2/3.5GHz core/uncore, 1.175/1.15v Gigabyte X99 SOC Champion (F22n) Gigabyte AORUS GTX 1080 Ti (F3P) @ 2025/1485, 1... 4x4GiB Crucial @ 2667, 12-12-12-28-T1, 1.34v 
Hard DriveHard DriveHard DriveCooling
Plextor M6e 128GB (fw 1.06) M.2 (PCI-E 2.0 2x) 2x Crucial M4 256GB 4x WD Scorpio Black 500GB Noctua NH-D15 
OSMonitorKeyboardPower
Windows 7 Professional x64 SP1 BenQ BL3200PT Filco Majestouch Tenkeyless (MX Brown) Corsair RM1000x 
CaseMouseAudio
Fractal Design Define R4 Logitech G402 Realtek ALC1150 + M-Audio AV40 
CPUMotherboardGraphicsRAM
X5670 @ 4.4/3.2GHz core/uncore, 1.36 vcore, 1.2... Gigabyte X58A-UD5 r2.0 w/FF3mod10 BIOS Sapphire Fury Nitro OC+ @ 1053/500, 1.225vGPU/1... 2x Samsung MV-3V4G3D/US @ 2000, 10-11-11-30-T1,... 
RAMHard DriveHard DriveHard Drive
1x Crucial BLT4G3D1608ET3LX0 @ 2000, 10-11-11-3... OCZ (Toshiba) Trion 150 120GB Hyundai Sapphire 120GB 3x Hitachi Deskstar 7k1000.C 1TB 
CoolingOSPowerCase
Noctua NH-D14 Windows 7 Pro x64 SP1 Antec TP-750 Fractal Design R5 
Audio
ASUS Xonar DS 
CPUMotherboardGraphicsRAM
i7-6800K @ 4.3/3.5GHz core/uncore, 1.36/1.2v ASRock X99 OC Formula (P3.10) GTX 780 (temporary) 4x4GiB Crucial DDR4-2400 @ 11-13-12-28-T2, 1.33v 
Hard DriveHard DriveCoolingOS
Intel 600p 256GB NVMe 2x HGST Travelstar 7k1000 1TB Corsair H55 (temporary) Windows Server 2016 Datacenter 
PowerCase
Seasonic SS-860XP2 Corsair Carbide Air 540 
  hide details  
Reply
post #144 of 175
The problem is that people are laying all the VRAM capacity changes at HBM's feet. If VRAM is being dynamically allocated by anything other than the game engine or graphics driver then its going to take dedicated hardware on the graphics chip to tag and sort all allocated resources, which is nothing really new and it could very easily be tested by running a benchmark which utilizes a repeated texture.
post #145 of 175
Have you ever even used a card with hbm?
Yokai Onmyoji
(18 items)
 
  
CPUMotherboardGraphicsRAM
I7-4930k @ 4.5GHz ASUS P9X79 Deluxe  Sapphire RX Vega 64 Liquid  G. Skill Ripjaws Z 16GB (4x4) 2400 Mhz  
Hard DriveHard DriveOptical DriveCooling
Samsung 840 Pro 256GB (OS) Samsung 840 Pro 512GB (Games) Lite-on Corsair H100 
OSMonitorKeyboardPower
Windows 7 Ultimate Wasabi Mango UHD490 IPS 49" Freesync 33-60hz Corsair K90 Corsair Ax1200i 
CaseMouseMouse PadAudio
Corsair 500r Razer Naga Dx Racer Arm wrest  Asus Xonar DX 
AudioOther
Altec Lansing 5.1 Razer Nostromo 
  hide details  
Reply
Yokai Onmyoji
(18 items)
 
  
CPUMotherboardGraphicsRAM
I7-4930k @ 4.5GHz ASUS P9X79 Deluxe  Sapphire RX Vega 64 Liquid  G. Skill Ripjaws Z 16GB (4x4) 2400 Mhz  
Hard DriveHard DriveOptical DriveCooling
Samsung 840 Pro 256GB (OS) Samsung 840 Pro 512GB (Games) Lite-on Corsair H100 
OSMonitorKeyboardPower
Windows 7 Ultimate Wasabi Mango UHD490 IPS 49" Freesync 33-60hz Corsair K90 Corsair Ax1200i 
CaseMouseMouse PadAudio
Corsair 500r Razer Naga Dx Racer Arm wrest  Asus Xonar DX 
AudioOther
Altec Lansing 5.1 Razer Nostromo 
  hide details  
Reply
post #146 of 175
Quote:
Originally Posted by Blameless View Post

None of that is remotely proof of what you're claiming. It's a fallacy to assume that memory type is the cause of the VRAM usage differences here.

It stands to reason that the part with less VRAM would be more aggressive in evicting assets.
And I can show different uses in the same task among two GDDR5 cards, even if they have the same amount of physical VRAM, and even if they are of the same brand. Hell, I can show a difference in many tasks from one run to another on the same system, hardware, and software environment. Figures for allocated VRAM don't show what that VRAM is being used for, nor if it's strictly necessary to have that much allocated.

Only a performance comparison in a known VRAM limited scenario where a part with HBM is seeing a substantially lower relative performance impact than a GDDR5 part with the same or greater amount of VRAM would imply HBM is somehow more efficient at VRAM utilization. To the best of my knowledge, no one has ever demonstrated any such thing, and probably for good reason (as the idea of HBM providing an intrinsic advantage in utilization and not simply performance, flies in the face of logic).

Heres your evidence
http://m.hardocp.com/article/2016/02/29/rise_tomb_raider_graphics_features_performance/14#.V-nn1tApBnE
Yokai Onmyoji
(18 items)
 
  
CPUMotherboardGraphicsRAM
I7-4930k @ 4.5GHz ASUS P9X79 Deluxe  Sapphire RX Vega 64 Liquid  G. Skill Ripjaws Z 16GB (4x4) 2400 Mhz  
Hard DriveHard DriveOptical DriveCooling
Samsung 840 Pro 256GB (OS) Samsung 840 Pro 512GB (Games) Lite-on Corsair H100 
OSMonitorKeyboardPower
Windows 7 Ultimate Wasabi Mango UHD490 IPS 49" Freesync 33-60hz Corsair K90 Corsair Ax1200i 
CaseMouseMouse PadAudio
Corsair 500r Razer Naga Dx Racer Arm wrest  Asus Xonar DX 
AudioOther
Altec Lansing 5.1 Razer Nostromo 
  hide details  
Reply
Yokai Onmyoji
(18 items)
 
  
CPUMotherboardGraphicsRAM
I7-4930k @ 4.5GHz ASUS P9X79 Deluxe  Sapphire RX Vega 64 Liquid  G. Skill Ripjaws Z 16GB (4x4) 2400 Mhz  
Hard DriveHard DriveOptical DriveCooling
Samsung 840 Pro 256GB (OS) Samsung 840 Pro 512GB (Games) Lite-on Corsair H100 
OSMonitorKeyboardPower
Windows 7 Ultimate Wasabi Mango UHD490 IPS 49" Freesync 33-60hz Corsair K90 Corsair Ax1200i 
CaseMouseMouse PadAudio
Corsair 500r Razer Naga Dx Racer Arm wrest  Asus Xonar DX 
AudioOther
Altec Lansing 5.1 Razer Nostromo 
  hide details  
Reply
post #147 of 175

"AMD is hand-tuning games to ensure 4GB will not hold Fiji performance back". Which is exactly what Blameless and others are saying, they are managing the memory allocation through driver memory management - HBM is not somehow magically storing 6GB of data in 4GB of VRAM.

And that guy lost all credibility when he said you can't compare SSD and mechanical hard drive capabilities. Umm, what? Someone better tell all the SSD manufacturers. 4GB is 4GB, whether it is HBM, GDDR5, SSD or hard drive.

No one has ever said that AMD can't more intelligently manage VRAM to keep the 4GB limitation from being a problem, but the idea that there is something inherent in HBM that makes that possible needs to die. At launch, AMD stated that they had always taken a rather cavalier approach to memory management because they had essentially always had enough, and that there were better ways to allocate assets to reduce usage. Which is exactly what they are doing. But those efforts can be applied to GDDR as well as HBM.

Edit: and all VRAM is essentialy an embedded cache, storing on-card assets that otherwise would have to be fetched from system RAM when needed.
Edited by Forceman - 9/26/16 at 10:13pm
post #148 of 175
This proves nothing, first it says with everything max settings, and 4k on a titanx it takes 10GB, then it say how on very high the 390x shows 4.2GB's(what resolution? because I find it hard to believe that AMD cards uses less then half the VRAM as an NV card at the same resolution), then it says the furyx uses 4GB's(what resolution?, is that at very high also?), it tells you a bunch of incomplete information, and to top it off it doesn't show the most important thing of all, the actual performence on screen. That was extremely poorly written, or very vague on purpose with the intention to mislead.
Edited by Piraal - 9/26/16 at 10:33pm
post #149 of 175
also certain games just dumps lots of prefetched textures, so if you use a card which has a huge VRAM buffer it'll obviously also use up a lot of VRAM.
just because it uses a lot of VRAM on a large VRAM pool, doesn't mean it requires that much VRAM.
post #150 of 175
Iris pro and xbone has small but fast eDRAM, if being fast doesn't solve the size issue ( to some extent) than why are those 2 have pretty small eDRAM in size ? why is that when a cache is fast it's smaller, and when it is slow it's bigger ? deny all you want, but the faster speed does makes up for smaller size. yes it doesn't solve it completely but it does help. because no texture is a solid 4GB+ large in size.
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Hardware News
Overclock.net › Forums › Industry News › Hardware News › [Computerbase] - 3gb vs 4gb vs 6gb vs 8gb GDDR5 VRAM Frametime Testing