Overclock.net › Forums › Graphics Cards › AMD/ATI › [Official] Polaris Owners Club
New Posts  All Forums:Forum Nav:

[Official] Polaris Owners Club - Page 101

post #1001 of 4140
Quote:
Originally Posted by Echoa View Post

Probably just one has higher clock speed. The Accelero might need some slight mortification but should fit but isn't necessary to keep it cool and is way overkill.

Thanks for the Answer on the Accelero.
My current RX480 Reference is quite loud under full load. I was thinking exchanging the Cooler or plain simply exchange the whole card.
Corsair Air 240
(16 items)
 
Oldie
(7 items)
 
 
CPUMotherboardGraphicsRAM
Intel Core i5-6600K 3,5 GHz (Skylake) Sockel 11... ASUS Maximus VIII GENE, Intel Z170 Mainboard - ... XFX Radeon RX 480 GTR Black Edition, 8192 MB G... Corsair Vengeance LED Series, red LED, DDR4-300... 
Hard DriveHard DriveHard DriveCooling
Samsung SM951-NVMe SSD, PCIe M.2 Typ 2280-D3-M ... Crucial MX200 2,5 Zoll SSD, SATA 6G - 500 GB Toshiba SATA 6G, DT01ACA300, 3,5 Zoll - 3 TB Corsair Cooling Hydro Series H100i V2 
CoolingPowerCaseOther
Corsair SP120 PWM Quiet Edition Twin Pack Corsair HX750i High Performance Corsair Carbide Air 240 Micro-ATX - white CableMod C-Series AXi, HXi, TX/CX/CS-M & RM Cab... 
CPUMotherboardGraphicsRAM
AMD Phenom II X4 955 SABERTOOTH 990FX ATI Radeon HD 5770 Corsair  
RAMRAMRAM
Corsair  Corsair  Corsair  
  hide details  
Reply
Corsair Air 240
(16 items)
 
Oldie
(7 items)
 
 
CPUMotherboardGraphicsRAM
Intel Core i5-6600K 3,5 GHz (Skylake) Sockel 11... ASUS Maximus VIII GENE, Intel Z170 Mainboard - ... XFX Radeon RX 480 GTR Black Edition, 8192 MB G... Corsair Vengeance LED Series, red LED, DDR4-300... 
Hard DriveHard DriveHard DriveCooling
Samsung SM951-NVMe SSD, PCIe M.2 Typ 2280-D3-M ... Crucial MX200 2,5 Zoll SSD, SATA 6G - 500 GB Toshiba SATA 6G, DT01ACA300, 3,5 Zoll - 3 TB Corsair Cooling Hydro Series H100i V2 
CoolingPowerCaseOther
Corsair SP120 PWM Quiet Edition Twin Pack Corsair HX750i High Performance Corsair Carbide Air 240 Micro-ATX - white CableMod C-Series AXi, HXi, TX/CX/CS-M & RM Cab... 
CPUMotherboardGraphicsRAM
AMD Phenom II X4 955 SABERTOOTH 990FX ATI Radeon HD 5770 Corsair  
RAMRAMRAM
Corsair  Corsair  Corsair  
  hide details  
Reply
post #1002 of 4140
Quote:
Originally Posted by blue198444 View Post

have someone the Solution? / tried it?

The BIOS has a signature that must be updated for every change to the BIOS. The driver checks for this signature and will not work if the signature doesn't match the BIOS.

AFAICT no one has found a way around this yet.
post #1003 of 4140
Quote:
Originally Posted by Gladi View Post

Hey.
Whats the difference between these two?

XFX Radeon RX 480 GTR Black Edition, 8192 MB GDDR5
XFX Radeon RX 480 GTR DD Edition, 8192 MB GDDR5

The store description shows no difference but the price (20 Euro difference)

edit:

Arctic Accelero Xtreme IV will that fit the RX480 Reference?

Found the difference.
Its the core clock -
1338 MHz
1288 MHz
Corsair Air 240
(16 items)
 
Oldie
(7 items)
 
 
CPUMotherboardGraphicsRAM
Intel Core i5-6600K 3,5 GHz (Skylake) Sockel 11... ASUS Maximus VIII GENE, Intel Z170 Mainboard - ... XFX Radeon RX 480 GTR Black Edition, 8192 MB G... Corsair Vengeance LED Series, red LED, DDR4-300... 
Hard DriveHard DriveHard DriveCooling
Samsung SM951-NVMe SSD, PCIe M.2 Typ 2280-D3-M ... Crucial MX200 2,5 Zoll SSD, SATA 6G - 500 GB Toshiba SATA 6G, DT01ACA300, 3,5 Zoll - 3 TB Corsair Cooling Hydro Series H100i V2 
CoolingPowerCaseOther
Corsair SP120 PWM Quiet Edition Twin Pack Corsair HX750i High Performance Corsair Carbide Air 240 Micro-ATX - white CableMod C-Series AXi, HXi, TX/CX/CS-M & RM Cab... 
CPUMotherboardGraphicsRAM
AMD Phenom II X4 955 SABERTOOTH 990FX ATI Radeon HD 5770 Corsair  
RAMRAMRAM
Corsair  Corsair  Corsair  
  hide details  
Reply
Corsair Air 240
(16 items)
 
Oldie
(7 items)
 
 
CPUMotherboardGraphicsRAM
Intel Core i5-6600K 3,5 GHz (Skylake) Sockel 11... ASUS Maximus VIII GENE, Intel Z170 Mainboard - ... XFX Radeon RX 480 GTR Black Edition, 8192 MB G... Corsair Vengeance LED Series, red LED, DDR4-300... 
Hard DriveHard DriveHard DriveCooling
Samsung SM951-NVMe SSD, PCIe M.2 Typ 2280-D3-M ... Crucial MX200 2,5 Zoll SSD, SATA 6G - 500 GB Toshiba SATA 6G, DT01ACA300, 3,5 Zoll - 3 TB Corsair Cooling Hydro Series H100i V2 
CoolingPowerCaseOther
Corsair SP120 PWM Quiet Edition Twin Pack Corsair HX750i High Performance Corsair Carbide Air 240 Micro-ATX - white CableMod C-Series AXi, HXi, TX/CX/CS-M & RM Cab... 
CPUMotherboardGraphicsRAM
AMD Phenom II X4 955 SABERTOOTH 990FX ATI Radeon HD 5770 Corsair  
RAMRAMRAM
Corsair  Corsair  Corsair  
  hide details  
Reply
post #1004 of 4140
Quote:
Originally Posted by looncraz View Post

I will be doing a voltage scaling in regards to power consumption analysis very shortly - it's pretty insane how quickly power usage increases, even at idle, with increased voltage on these cards.
Yep. I had the same on Tahiti. Very voltage sensitive for consumption.
Quote:
nVidia has been running their hardware much closer to their peak clocks and advertising lower frequencies so the overclocked frequencies look more impressive... but the reality is that the GTX 1060 and RX 480 overclock nearly the same, with a slight edge to the RX 480 (an overclock of 2.05Ghz on GTX 1060 is the same 1340MHz on RX 480).
RX 480 scales nearly perfectly linearly with memory frequency. It is ABSURDLY bound by memory performance:



Anyone who says otherwise is just experiencing driver problems and may not even know it.

I've put hours of work into examining Polaris 10 in both RX 470 and RX 480 clothes. I have even created full curves of memory sensitivity in 200MHz increments from 600Mhz:



You can see that scaling only really stops with the 750MHz GPU and 1600MHz RAM. That would be 2.667GHz RAM clocks for 1250MHz... or 340GB/s. Meaning Polaris 10 needs 33% more bandwidth than it has for optimal performance. I haven't done the math, but I think it would be about 25% faster with that much bandwidth, with no increase in GPU clocks... and almost no change in power usage.

Ok, finally found it again:



He uses a golden RX 480 card with 1465MHz OC though so beware, that's +100MHz from the usual max OC.
Try games not synthetic benchmarks, I don't play Heaven, Valley, Furmark or Firestrike do you? biggrin.gif
I would say the P10 is ROP limited and doesn't scale as well as we would like on core clocks with the voltages it uses, meaning it runs hot and slow especially by default. Of course I do prefer wider bus but then it would be even more power hungry and 6GB version on a 384bit? Nah, they scaled down to 256bit long time ago from the Tahiti 384bit, using memory compressions etc. helps, faster VRAM too, it's not the best but in games bandwidth doesn't seem to be a big deal for GCN, sure sometimes 280x would be faster than 380x but those can also be flukes.
1060 seems to be mainly bandwidth limited as it has 192bit bus, which is half of what Tahiti had lol, yet it almost matches 980 256bit in performance.

Yeah the Paxwell is Maxwell stretched as far as it will go probably, but it's fast nonetheless and AMD has problem keeping up with anything NV releases frown.gif
AMD is getting killed by the bad pricing and availability on 4xx, at least for me. It was the same with 280x, 290s, 3xx series, always priced higher than comparable NV cards and in short supply. Yet never selling more than NV cards and gaining market share either. They are not going to get market share in gaming if their cards can't be bought by gamers especially when they do not offer better performance/price either anymore.

---

Yes AC AX IV fits RX 480. Any cooler that doesn't obstruct the VRMs toward I/O ports will work without modding.
Edited by JackCY - 8/31/16 at 7:24am
post #1005 of 4140
Quote:
Originally Posted by greytoad View Post

I changed to lower my offset and get the same voltage. (.5 mv different.)

My overclock still puts idle at 887.5.mv. I'll still go back to stock for netflix sessions I guess. Unless someone tells me it's perfectly fine to run 87mv higher than stock for only 300mh p0 speeds.

I would assume that streaming or anything else (but pure idling) at least sets your memory to its performance state. So your core voltage would go up anyway (it is limited from below by the "memory" voltage).
Quote:
Originally Posted by looncraz View Post

The BIOS has a signature that must be updated for every change to the BIOS. The driver checks for this signature and will not work if the signature doesn't match the BIOS.

AFAICT no one has found a way around this yet.

That's all done already. You need a modified driver which bypasses the signature check. This driver is not signed, so you need testmode and disable the driver (not bios) signature check.
i5 3470 + RX 480
(13 items)
 
  
CPUMotherboardGraphicsRAM
Intel Core i5-3470 Gigabyte H77-DS3H AMD Radeon RX 480 2 x 4GB Corsair Vengeance LP DDR3-1600 
Hard DriveHard DriveCoolingOS
128GB Samsung 830 SSD 240GB OCZ Trion 150 SSD Thermalright HR-02 Macho Rev. A (BW) Windows 10 Pro 
MonitorKeyboardPowerCase
Dell U2713HM generic Thermaltake Munic 430W Antec Three Hundred 
Mouse
Roccat Kone Pure Military 
  hide details  
Reply
i5 3470 + RX 480
(13 items)
 
  
CPUMotherboardGraphicsRAM
Intel Core i5-3470 Gigabyte H77-DS3H AMD Radeon RX 480 2 x 4GB Corsair Vengeance LP DDR3-1600 
Hard DriveHard DriveCoolingOS
128GB Samsung 830 SSD 240GB OCZ Trion 150 SSD Thermalright HR-02 Macho Rev. A (BW) Windows 10 Pro 
MonitorKeyboardPowerCase
Dell U2713HM generic Thermaltake Munic 430W Antec Three Hundred 
Mouse
Roccat Kone Pure Military 
  hide details  
Reply
post #1006 of 4140
Thread Starter 
Quote:
Originally Posted by Gladi View Post

Thanks for the Answer on the Accelero.
My current RX480 Reference is quite loud under full load. I was thinking exchanging the Cooler or plain simply exchange the whole card.

yea just mortify it tongue.gif

tell it it dresses awful, make fun of it, etc. you know....eventually itll just fit


lol at typos
Toxic-DT
(19 items)
 
Toxic-Mini
(6 items)
 
Toxic-SV
(10 items)
 
CPUMotherboardGraphicsRAM
i7 4770k Gigabyte Z87X-UD4H Sapphire Nitro + RX 470 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
Rk3288 Firefly  Mali t760 mp4 Ddr3 1600 
Hard DriveOS
Internal EMMC Android 
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-Mini
(6 items)
 
Toxic-SV
(10 items)
 
CPUMotherboardGraphicsRAM
i7 4770k Gigabyte Z87X-UD4H Sapphire Nitro + RX 470 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
Rk3288 Firefly  Mali t760 mp4 Ddr3 1600 
Hard DriveOS
Internal EMMC Android 
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 #1007 of 4140
Quote:
Originally Posted by JackCY View Post

Warning: Spoiler! (Click to show)

That's to a good way to compare them, that's for sure.

RX 480: 8.5% more bandwidth should result in ~.8.5% more performance. And it does pretty close to that.

GTX 1060 has 17.5% higher memory clocks, and gains about 15% more performance.

This suggests both cards a very bandwidth limited, but the RX 480 more so (not surprising since the last AMD used cards of this performance level they gave it 320GB/s of bandwidth .. and it still wasn't quite enough).

Not to mention just how poor 2175Mhz is for an RX 480 memory overclock. I can run 2250MHz memory clocks all day long, but I have to increase volts on the GPU (even without a GPU overclock). People think the memory voltage entry is some other voltage bus - it isn't, it's just a setting that sets a minimum GPU voltage while the memory is running at that frequency. AFAICT, the memory controllers work at GPU voltage.
Quote:
Originally Posted by JackCY View Post

He uses a golden RX 480 card with 1465MHz OC though so beware, that's +100MHz from the usual max OC.

But not unusual for water-cooled RX 480s. 14nm LPP is very temperature sensitive. I can run 1.4Ghz if I keep the temperatures under ~65C... which is not easy on air.
Quote:
Originally Posted by JackCY View Post

Try games not synthetic benchmarks, I don't play Heaven, Valley, Furmark or Firestrike do you? biggrin.gif

I have, here's BF4 scaling nearly perfectly with memory frequency:



Quote:
Originally Posted by JackCY View Post

I would say the P10 is ROP limited

ROP limitations most easily show with higher resolutions. GTX 1060 scales worse than Polaris at higher resolutions... albeit not by much. We are seeing that the ROPs are being constricted by memory bandwidth, which is why scaling is so linear.
Quote:
Originally Posted by JackCY View Post

and doesn't scale as well as we would like on core clocks with the voltages it uses, meaning it runs hot and slow especially by default.

You and I have very different definitions of "hot" and "slow." The RX 480 is very similar to the GTX 1060 across the board. Its default configuration uses TOO MUCH power... and it is just 30W more than the GTX 1060 in that configuration. Some AIB cards are even worse (Sapphire Nitro+, I'm looking at you!), but others have it under control (XFX RX 480 GTR).

Take a look (full system usage above idle, not specifically card power):




RX 480 uses 34W less at 1275Mhz with just proper voltages. That's should make it roughly on par with GTX 1060.
Quote:
Originally Posted by JackCY View Post

in games bandwidth doesn't seem to be a big deal for GCN

Depends on the specific floor plan. Earlier GCN was less sensitive than new GCN. The new ROPs are more hungry for bandwidth as they are nearly twice as fast as the old ROPs. 32 GCN4 ROPs needs similar bandwidth as 64 GCN 1.1 ROPs.

And, at the same time, memory compression has helped alleviate some of the bandwidth concerns. nVidia is still well ahead here, but AMD will catch up as maximum entropy comes into play (already very close)... compression can only get you so far.
Quote:
Originally Posted by JackCY View Post

1060 seems to be mainly bandwidth limited as it has 192bit bus, which is half of what Tahiti had lol, yet it almost matches 980 256bit in performance.

It has a more narrow bus, but it has superior memory compression. This does mean that it will be limited, particularly with incompressible data in play... which makes the impact very dependent on the workload.
Quote:
Originally Posted by JackCY View Post

Yeah the Paxwell is Maxwell stretched as far as it will go probably, but it's fast nonetheless and AMD has problem keeping up with anything NV releases frown.gif

You seem to think that AMD knows the performance level to target when they design a card... they don't. They have to guess nVidia's moves and hope they are right... with the full knowledge that nVidia probably has spies in AMD. What else would explain how these companies stay so close when they release products that take years to develop nearly at the same time and with such similar performance?

On a technological basis, AMD is actually well ahead in many areas. Not just async compute, either. AMD has better ROPs, scheduler, internal bandwidth, caches, and probably more I know nothing about. They are well behind in only a few areas (memory compression and frequency, primarily). Most of AMD's technical advantages are easily eroded by higher frequency. How to predict nVidia would go speed-demon with Pascal? You couldn't. Particularly not up to 2Ghz.

AMD's real internal problem is their relatively poor utilization of their execution resources. nVidia designed their GPUs strictly for DX11 games. AMD built a world-class generic compute unit that can do gaming. One way is more efficient than the other, but will fall behind in non specialized workloads. For me, everything my AMD cards can do is very valuable. My software suites rely on GPU performance for calculations (mainly my video editing software). My HD 5770 was several times faster than my GTX 560 for editing videos... and that before nVidia really hit the chopping block.
Quote:
Originally Posted by JackCY View Post

AMD is getting killed by the bad pricing and availability on 4xx, at least for me.

The market will bear the price at this time, so the prices will only come down slowly. Availability is not an issue like it was, even if it is where you are... that's your own suppliers' problem, not AMD's. Prices will drop when demand drops relative to supply. It's the mining effect all over again - this time, though, supplies are actually better, which has helped keep prices somewhat tolerable. Hawaii prices reached $600... and some people paid $800 for them, just because they could make it back with mining.
Quote:
Originally Posted by JackCY View Post

It was the same with 280x, 290s, 3xx series, always priced higher than comparable NV cards and in short supply.

Same reason. Problem is that AMD can't rely on the miners to buy the cards predictably - so they don't order a large first batch. nVidia knows they will be able to sale their GPUs, and they have cash to burn, so they buy tons of wafers at low yields to provide volume. This time around, they even have a die-size advantage due to high clocks... something I suspect AMD will copy.
Quote:
Originally Posted by JackCY View Post

Yet never selling more than NV cards and gaining market share either.

AMD's market share has been increasing. They have stolen market share from nVidia for the last three or four quarters, IIRC. They have done this because their generation over generation improvements have actually been more impressive than nVidia's, without cutting features, and their older cards have shown remarkable quality of life compared to nVidia cards. Someone who bought a 7970Ghz edition can still play any game on the market using the latest APIs. That card competed with the GTX 780. Today, the 780 is a joke.
Quote:
Originally Posted by JackCY View Post

They are not going to get market share in gaming if their cards can't be bought by gamers especially when they do not offer better performance/price either anymore.
[//quote]

It doesn't matter where the cards are used, only that they are purchased. They could be thrown in the trash... it would still help AMD.
post #1008 of 4140
Quote:
Originally Posted by looncraz View Post

That's to a good way to compare them, that's for sure.

RX 480: 8.5% more bandwidth should result in ~.8.5% more performance. And it does pretty close to that.

GTX 1060 has 17.5% higher memory clocks, and gains about 15% more performance.

This suggests both cards a very bandwidth limited, but the RX 480 more so (not surprising since the last AMD used cards of this performance level they gave it 320GB/s of bandwidth .. and it still wasn't quite enough).

Not to mention just how poor 2175Mhz is for an RX 480 memory overclock. I can run 2250MHz memory clocks all day long, but I have to increase volts on the GPU (even without a GPU overclock). People think the memory voltage entry is some other voltage bus - it isn't, it's just a setting that sets a minimum GPU voltage while the memory is running at that frequency. AFAICT, the memory controllers work at GPU voltage.
But not unusual for water-cooled RX 480s. 14nm LPP is very temperature sensitive. I can run 1.4Ghz if I keep the temperatures under ~65C... which is not easy on air.
I have, here's BF4 scaling nearly perfectly with memory frequency:

The GPU is like a CPU, the cache/memory controller operate on a different voltage than the core. Power throttling even kicks in earlier with that voltage raised.
μRyzen
(12 items)
 
Mini Box
(4 items)
 
 
CPUMotherboardGraphicsRAM
Ryzen R5 1400 MSI B350M Gaming Pro Zotac GTX 670 4GB G.SKILL FORTIS Series 8GB (2 x 4GB) 
Hard DriveCoolingOSOS
WD Green 3tb Wraith Stealth Windows 10 Debian 8.7 
MonitorKeyboardPowerMouse
ViewSonic VX-2257-8 Chinese backlit mechanical Kingwin 850w Chinese laser optical 
CPUMotherboardGraphicsRAM
Athlon 5350 Asus AM1I-A EVGA GTX 750 Ti SC 2x4GB DDR 3 1333 
  hide details  
Reply
μRyzen
(12 items)
 
Mini Box
(4 items)
 
 
CPUMotherboardGraphicsRAM
Ryzen R5 1400 MSI B350M Gaming Pro Zotac GTX 670 4GB G.SKILL FORTIS Series 8GB (2 x 4GB) 
Hard DriveCoolingOSOS
WD Green 3tb Wraith Stealth Windows 10 Debian 8.7 
MonitorKeyboardPowerMouse
ViewSonic VX-2257-8 Chinese backlit mechanical Kingwin 850w Chinese laser optical 
CPUMotherboardGraphicsRAM
Athlon 5350 Asus AM1I-A EVGA GTX 750 Ti SC 2x4GB DDR 3 1333 
  hide details  
Reply
post #1009 of 4140
Quote:
Originally Posted by looncraz View Post

Warning: Spoiler! (Click to show)
That's to a good way to compare them, that's for sure.

RX 480: 8.5% more bandwidth should result in ~.8.5% more performance. And it does pretty close to that.

GTX 1060 has 17.5% higher memory clocks, and gains about 15% more performance.

This suggests both cards a very bandwidth limited, but the RX 480 more so (not surprising since the last AMD used cards of this performance level they gave it 320GB/s of bandwidth .. and it still wasn't quite enough).

Not to mention just how poor 2175Mhz is for an RX 480 memory overclock. I can run 2250MHz memory clocks all day long, but I have to increase volts on the GPU (even without a GPU overclock). People think the memory voltage entry is some other voltage bus - it isn't, it's just a setting that sets a minimum GPU voltage while the memory is running at that frequency. AFAICT, the memory controllers work at GPU voltage.
But not unusual for water-cooled RX 480s. 14nm LPP is very temperature sensitive. I can run 1.4Ghz if I keep the temperatures under ~65C... which is not easy on air.
I have, here's BF4 scaling nearly perfectly with memory frequency:


ROP limitations most easily show with higher resolutions. GTX 1060 scales worse than Polaris at higher resolutions... albeit not by much. We are seeing that the ROPs are being constricted by memory bandwidth, which is why scaling is so linear.
You and I have very different definitions of "hot" and "slow." The RX 480 is very similar to the GTX 1060 across the board. Its default configuration uses TOO MUCH power... and it is just 30W more than the GTX 1060 in that configuration. Some AIB cards are even worse (Sapphire Nitro+, I'm looking at you!), but others have it under control (XFX RX 480 GTR).

Take a look (full system usage above idle, not specifically card power):




RX 480 uses 34W less at 1275Mhz with just proper voltages. That's should make it roughly on par with GTX 1060.
Depends on the specific floor plan. Earlier GCN was less sensitive than new GCN. The new ROPs are more hungry for bandwidth as they are nearly twice as fast as the old ROPs. 32 GCN4 ROPs needs similar bandwidth as 64 GCN 1.1 ROPs.

And, at the same time, memory compression has helped alleviate some of the bandwidth concerns. nVidia is still well ahead here, but AMD will catch up as maximum entropy comes into play (already very close)... compression can only get you so far.
It has a more narrow bus, but it has superior memory compression. This does mean that it will be limited, particularly with incompressible data in play... which makes the impact very dependent on the workload.
You seem to think that AMD knows the performance level to target when they design a card... they don't. They have to guess nVidia's moves and hope they are right... with the full knowledge that nVidia probably has spies in AMD. What else would explain how these companies stay so close when they release products that take years to develop nearly at the same time and with such similar performance?

On a technological basis, AMD is actually well ahead in many areas. Not just async compute, either. AMD has better ROPs, scheduler, internal bandwidth, caches, and probably more I know nothing about. They are well behind in only a few areas (memory compression and frequency, primarily). Most of AMD's technical advantages are easily eroded by higher frequency. How to predict nVidia would go speed-demon with Pascal? You couldn't. Particularly not up to 2Ghz.

AMD's real internal problem is their relatively poor utilization of their execution resources. nVidia designed their GPUs strictly for DX11 games. AMD built a world-class generic compute unit that can do gaming. One way is more efficient than the other, but will fall behind in non specialized workloads. For me, everything my AMD cards can do is very valuable. My software suites rely on GPU performance for calculations (mainly my video editing software). My HD 5770 was several times faster than my GTX 560 for editing videos... and that before nVidia really hit the chopping block.
The market will bear the price at this time, so the prices will only come down slowly. Availability is not an issue like it was, even if it is where you are... that's your own suppliers' problem, not AMD's. Prices will drop when demand drops relative to supply. It's the mining effect all over again - this time, though, supplies are actually better, which has helped keep prices somewhat tolerable. Hawaii prices reached $600... and some people paid $800 for them, just because they could make it back with mining.
Same reason. Problem is that AMD can't rely on the miners to buy the cards predictably - so they don't order a large first batch. nVidia knows they will be able to sale their GPUs, and they have cash to burn, so they buy tons of wafers at low yields to provide volume. This time around, they even have a die-size advantage due to high clocks... something I suspect AMD will copy.
AMD's market share has been increasing. They have stolen market share from nVidia for the last three or four quarters, IIRC. They have done this because their generation over generation improvements have actually been more impressive than nVidia's, without cutting features, and their older cards have shown remarkable quality of life compared to nVidia cards. Someone who bought a 7970Ghz edition can still play any game on the market using the latest APIs. That card competed with the GTX 780. Today, the 780 is a joke.
Quote:
Originally Posted by JackCY View Post

They are not going to get market share in gaming if their cards can't be bought by gamers especially when they do not offer better performance/price either anymore.
[//quote]

It doesn't matter where the cards are used, only that they are purchased. They could be thrown in the trash... it would still help AMD.
Yeah the comparison needs to be in relative numbers not absolute, you're right they seem to be similarly restricted by memory bandwidth.

Selling cards to miners that then RMA them in bulk when mining is not profitable anymore or their warranty is about to expire is actually hurting the AMD AIBs.
The supply and sell numbers of 1060s are so much better than 480s. Maybe they would sell equally if the 480s were available.

I have a 1060 on order that I can cancel. I saw 2 pieces of 480 4GB Nitro but they were sold ASAP. I just don't like the Nitro cooler and 4GB not being Samsung chips and only clocking below 8GHz frown.gif The 8GB Nitro costs the same as equivalent 1060 but again, rarely available. I've sent an inquiry to the shop about 3 AMD cards to see if they have some reasonable response about availability. I may also just test drive the 1060 for 2 weeks and return it wink.gif
I do not like how NV is approaching the market and hardware advances, tons of marketing and BS just because they can afford it.

If there was a cheap reference 480 8GB or 480 4GB with Samsung chips I would get those ASAP but they aren't.
And I'm still waiting for my gorram refund for an R9 280x gunner.gif It wasn't cutting it for me anymore and the artifacts were annoying.
Edited by JackCY - 8/31/16 at 11:01am
post #1010 of 4140
Quote:
Originally Posted by KarathKasun View Post

The GPU is like a CPU, the cache/memory controller operate on a different voltage than the core. Power throttling even kicks in earlier with that voltage raised.

If that's indeed still the case, then AMD won't let the GPU voltage fall below the memory controller voltage... or they share the same voltage plane... as VDDC is locked in tandem.

If I set low voltages on the GPU and high voltages on the memory, the GPU will run at the same voltages as the memory.

As in, the GPU will run at 1100mv if I set 1100mv on the memory and 1000mv on the core.

I have also seen no benefit from increasing the memory controller voltage if it stays below the core voltage, only once the total voltage increases do things get better.

In fact, I can set the memory controller voltage to 800mv and watch idle voltage drop to 800mv on the memory controller with 2200Mhz RAM as long as I crank up the GPU voltage enough.
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: AMD/ATI
Overclock.net › Forums › Graphics Cards › AMD/ATI › [Official] Polaris Owners Club