Overclock.net › Forums › Industry News › Rumors and Unconfirmed Articles › [VideoCardz] AMD Radeon RX 490X Specifications
New Posts  All Forums:Forum Nav:

[VideoCardz] AMD Radeon RX 490X Specifications - Page 33

post #321 of 374
Quote:
Originally Posted by epic1337 View Post

its actually the opposite, they've made a card that can excel on GDDR5 instead of resorting to using HBM, thats a good architectural feat right there, transmute old rust to gold.

whats a slap on the face is using something drastically superior and still falling short of performing better.
in this case HBM's lack luster performance on Fiji cards,

Again Fiji XT and Pro are bottlenecked in its architecture, Raster(RBes) Rasterizers ,Geometry processors.
Quote:
Originally Posted by lolerk52 View Post

HBM actually held back Fury X, rather than being an advantage
Fiji XT doesnt use properly bandwidth due to vram amount and maybe its architecture.
Quote:
It has tons of bandwidth, but due to the low clock speeds, the latency is far higher than GDDR5.
HBM latency is lower than GDDR5.
Edited by PontiacGTX - 6/10/16 at 7:30am
post #322 of 374
Quote:
Originally Posted by epic1337 View Post

its actually the opposite, they've made a card that can excel on GDDR5 instead of resorting to using HBM, thats a good architectural feat right there, transmute old rust to gold.

whats a slap on the face is using something drastically superior and still falling short of performing better.
in this case HBM's lack luster performance on Fiji cards, either HBM was all hype and worth nothing, or Fiji was crap architecture that can't capitalize on using HBM.

I'm glad that someone sees the point I was making.
post #323 of 374
Quote:
Originally Posted by PontiacGTX View Post

HBM latency is lower than GDDR5.
Memory latency is a combination of how many cycles it takes to respond, and the clock speed.

http://i1013.photobucket.com/albums/af254/Jaquith/CAS_Frequncy_Chart.png

HBM takes less cycles to respond than GDDR5, but since it has a substantially lower clock speed, the end result is higher latency. As I said, won't be an issue with HBM2, but it held back Fury X.



That's not to say Fury X wasn't bottlenecked in other areas, but LN2 overclocking of HBM reveals a MAJOR bottleneck was HBM latency.
Edited by lolerk52 - 6/10/16 at 7:51am
post #324 of 374
Please be under $300

Please be under $300
Zen
(17 items)
 
  
CPUMotherboardGraphicsRAM
i5-6600k Gigabyte Z170XP-SLI RX 480 4GB 24 GB DDR4 2133MHz 
Hard DriveHard DriveHard DriveHard Drive
Samsung 850 evo OCZ Vertex 4 Crucial MX 500 256GB WD Black 3 TB 
Hard DriveCoolingOSMonitor
Seagate 1 TB H60 Windows 10 Asus mx27a 
MonitorKeyboardPowerMouse
hp 2009 m Corsair MR Brown Antec Earthwatts 650 Razer Naga 
Audio
Soundblaster Omni 
  hide details  
Reply
Zen
(17 items)
 
  
CPUMotherboardGraphicsRAM
i5-6600k Gigabyte Z170XP-SLI RX 480 4GB 24 GB DDR4 2133MHz 
Hard DriveHard DriveHard DriveHard Drive
Samsung 850 evo OCZ Vertex 4 Crucial MX 500 256GB WD Black 3 TB 
Hard DriveCoolingOSMonitor
Seagate 1 TB H60 Windows 10 Asus mx27a 
MonitorKeyboardPowerMouse
hp 2009 m Corsair MR Brown Antec Earthwatts 650 Razer Naga 
Audio
Soundblaster Omni 
  hide details  
Reply
post #325 of 374
I was hoping AMD would have a new flagship out comparable to the gtx 1080 in performance before my rift shipped... my rift just shipped yesterday. I may have to snag as gtx 1080 for VR horsepower
post #326 of 374
Quote:
Originally Posted by Gabkicks View Post

I was hoping AMD would have a new flagship out comparable to the gtx 1080 in performance before my rift shipped... my rift just shipped yesterday. I may have to snag as gtx 1080 for VR horsepower

Vr is a completely different animal afaik, i would wait on benchmarks for that kind of thing. A couple 480's may end up being amazing for it.
Sorry AMD
(13 items)
 
   
CPUMotherboardGraphicsRAM
i5-8400 ROG STRIX Z370-I GAMING ZOTAC GTX 1080 Ti Mini 16gb Gskill Ripjaw V 3600mhz CL16 
Hard DriveCoolingOSMonitor
256gb m.2 Crucial noctua nh-l9i Windows 10 Pro AOC G2460PF 
KeyboardPowerCaseMouse
Mechanical Eagle Z-77 (Cheap) Corsair SF600 Fractal Design Node 202 Logitech G400 
Audio
Sennheiser PC 350 (hero mod) 
CPUMotherboardGraphicsRAM
AMD Ryzen™ 1700 Asrock Fatal1ty x370 Gaming K4 Zotac mini GTX 1080ti TridentZ 16GB (2x8GB) DDR4 
Hard DriveHard DriveOSMonitor
256GB M.2 SSD 1TB HDD Windows 10 Pro Acer XG270HU 
KeyboardPowerCaseMouse
Logitech G15 750W Thermaltake Toughpower Gold Rated Thermaltake View 27 w/ riser Logitech G40 
CPUMotherboardGraphicsRAM
fx-8350 @5ghz 990fxa-ud3 rev 1.1 PowerColor 290x @1150/1500mhz Gskill 8gig 1600mhz (2x4) @2033 mhz 
Hard DriveCoolingOSPower
1TB WD Caviar Black rs360 raystorm kit Windows 8 Pro 700 Watt Corsair 
CaseMouse
HAF 912  Logitech G400 
  hide details  
Reply
Sorry AMD
(13 items)
 
   
CPUMotherboardGraphicsRAM
i5-8400 ROG STRIX Z370-I GAMING ZOTAC GTX 1080 Ti Mini 16gb Gskill Ripjaw V 3600mhz CL16 
Hard DriveCoolingOSMonitor
256gb m.2 Crucial noctua nh-l9i Windows 10 Pro AOC G2460PF 
KeyboardPowerCaseMouse
Mechanical Eagle Z-77 (Cheap) Corsair SF600 Fractal Design Node 202 Logitech G400 
Audio
Sennheiser PC 350 (hero mod) 
CPUMotherboardGraphicsRAM
AMD Ryzen™ 1700 Asrock Fatal1ty x370 Gaming K4 Zotac mini GTX 1080ti TridentZ 16GB (2x8GB) DDR4 
Hard DriveHard DriveOSMonitor
256GB M.2 SSD 1TB HDD Windows 10 Pro Acer XG270HU 
KeyboardPowerCaseMouse
Logitech G15 750W Thermaltake Toughpower Gold Rated Thermaltake View 27 w/ riser Logitech G40 
CPUMotherboardGraphicsRAM
fx-8350 @5ghz 990fxa-ud3 rev 1.1 PowerColor 290x @1150/1500mhz Gskill 8gig 1600mhz (2x4) @2033 mhz 
Hard DriveCoolingOSPower
1TB WD Caviar Black rs360 raystorm kit Windows 8 Pro 700 Watt Corsair 
CaseMouse
HAF 912  Logitech G400 
  hide details  
Reply
post #327 of 374
Quote:
Originally Posted by lolerk52 View Post

Memory latency is a combination of how many cycles it takes to respond, and the clock speed.

http://i1013.photobucket.com/albums/af254/Jaquith/CAS_Frequncy_Chart.png

HBM takes less cycles to respond than GDDR5, but since it has a substantially lower clock speed, the end result is higher latency. As I said, won't be an issue with HBM2, but it held back Fury X.



That's not to say Fury X wasn't bottlenecked in other areas, but LN2 overclocking of HBM reveals a MAJOR bottleneck was HBM latency.
We do not know the CAS latency of HBM. Saying that since OCing HBM yields a massive performance increase, it must be a latency issue, is a huge stretch. GPUs are very complicated things, and the GPU may not be efficiently using the available memory bandwidth, and/or it may have other components inside synchronized to the memory clock. For example, the ROPs are usually tied very closely to the memory system, so if the ROPs are synchronized to the memory clock, you will see huge performance increase there, since Fiji is ROP starved compared to other competitive designs.
post #328 of 374
Quote:
Originally Posted by AmericanLoco View Post

We do not know the CAS latency of HBM. Saying that since OCing HBM yields a massive performance increase, it must be a latency issue, is a huge stretch. GPUs are very complicated things, and the GPU may not be efficiently using the available memory bandwidth, and/or it may have other components inside synchronized to the memory clock. For example, the ROPs are usually tied very closely to the memory system, so if the ROPs are synchronized to the memory clock, you will see huge performance increase there, since Fiji is ROP starved compared to other competitive designs.

CAS does not matter. If you do not change latencies when overclocking RAM by 100% the latency becomes 1/2. Latency is time for a request to be returned, and with a very low bus clock latency will inherently be high compared to a bus that is running 2-4x faster.

Memory timings are measured in cycles mostly, and CAS 5 @ 500mhz is slower than CAS 18 @ 2000mhz as far as response time is concerned.

ROPs are not tied to memory frequency, and have never been tied to memory frequency.
Edited by KarathKasun - 6/10/16 at 6:24pm
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 #329 of 374
Quote:
Originally Posted by KarathKasun View Post

CAS does not matter. If you do not change latencies when overclocking RAM by 100% the latency becomes 1/2. Latency is time for a request to be returned, and with a very low bus clock latency will inherently be high compared to a bus that is running 2-4x faster.

Memory timings are measured in cycles mostly, and CAS 5 @ 500mhz is slower than CAS 18 @ 2000mhz as far as response time is concerned.

ROPs are not tied to memory frequency, and have never been tied to memory frequency.
I'm well aware of what latency is, thanks.

How does CAS not mater? All the timings matter. The latency is not "inherently high" at low clocks, since low clocked memory often has an extremely low number of cycles needed for a response. For example, DDR1 400MHz @ CAS 2, it has the same latency as DDR3-2400 @ CAS 12. If HBM's timings are sufficiently fast, it can easily have the same, or better latency as GDDR5. In fact, Hynix advertises that it has similar or faster overall latency (in terms of ns) as high-clocked GDDR5.
post #330 of 374
HBM has a second memory controller in the line of access to the DRAM. So you get GPU -> stack control logic -> DRAM. Its very similar to FB-DIMM in that sense. So regardless of the actual DRAM CAS you have at minimum a 1-2 cycle latency on top of that, and at 500mhz that is large hit as a percentage of the DRAM latency. A 4-5 cycle latency becomes 5-7 cycles.
Edited by KarathKasun - 6/10/16 at 7:23pm
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
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Rumors and Unconfirmed Articles
Overclock.net › Forums › Industry News › Rumors and Unconfirmed Articles › [VideoCardz] AMD Radeon RX 490X Specifications