• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

AMD "Navi" Features 8 Streaming Engines, Possible ROP Count Doubling?

btarunr

Editor & Senior Moderator
Staff member
Joined
Oct 9, 2007
Messages
36,956 (8.66/day)
Location
Hyderabad, India
Processor AMD Ryzen 7 2700X
Motherboard MSI B450 Gaming Pro Carbon AC
Cooling AMD Wraith Prism
Memory 2x 16GB Corsair Vengeance LPX DDR4-3000
Video Card(s) Colorful iGame GTX 1070 Ti Vulcan X
Storage Western Digital Black NVMe 512GB
Display(s) Samsung U28D590 28-inch 4K UHD
Case Corsair Carbide 100R
Audio Device(s) Creative Sound Blaster Recon3D PCIe
Power Supply Antec EarthWatts Pro Gold 750W
Mouse Razer Abyssus
Keyboard Microsoft Sidewinder X4
Software Windows 10 Pro
AMD's 7 nm "Navi 10" silicon may finally address two architectural shortcomings of its performance-segment GPUs, memory bandwidth, and render-backends (deficiency thereof). The GPU almost certainly features a 256-bit GDDR6 memory interface, bringing about a 50-75 percent increase in memory bandwidth over "Polaris 30." According to a sketch of the GPU's SIMD schematic put out by KOMACHI Ensaka, Navi's main number crunching machinery is spread across eight shader engines, each with five compute units (CUs).

Five CUs spread across eight shader engines, assuming each CU continues to pack 64 stream processors, works out to 2,560 stream processors on the silicon. This arrangement is in stark contrast to the "Hawaii" silicon from 2013, which crammed 10 CUs per shader engine across four shader engines to achieve the same 2,560 SP count on the Radeon R9 290. The "Fiji" silicon that followed "Hawaii" stuck to the 4-shader engine arrangement. Interestingly, both these chips featured four render-backends per shader engine, working out to 64 ROPs. AMD's decision to go with 8 shader engines raises hopes for the company doubling ROP counts over "Polaris," to 64, by packing two render backends per shader engine. AMD unveils Navi in its May 27 Computex keynote, followed by a possible early-July launch.



View at TechPowerUp Main Site
 
Joined
Oct 1, 2006
Messages
4,321 (0.93/day)
Location
Hong Kong
Processor Ryzen Threadripper 1950X
Motherboard X399 Fatal1ty Professional Gaming
Cooling Custom loop water, 3x 420 Rad
Memory Team T-FORCE XCALIBUR RGB 4000 @3466 CL14
Video Card(s) Zotac RTX 2080 Ti
Storage Samsung 960 Evo 512GB 3-way Raid 0
Display(s) Samsung C27HG70
Case Thermaltake Level 20 XT
Audio Device(s) Xonar Essence STX II
Power Supply FSP Aurum PT 1200W
Software Windows 10 Pro 64-bit
If this is true, there might be some hope for RTG.
Assuming Raja actually did his job before he left for Intel, which was to make GCN more scale-able.
The Geometry limit was the Achilles heel of GCN in terms of gaming performance.

In Pixel Fill-rate the Radeon VII actually beats the 2080, but when you look at Geometry it is far behind.
The ROP count might not be the actual issue.


 
Last edited:
Joined
Mar 23, 2016
Messages
2,144 (1.82/day)
Processor Ryzen 5 2600X
Motherboard MSI B350M Gaming Pro
Cooling Scythe Kotetsu with AM4 bracket
Memory PNY Anarchy-X XLR8 Red DDR4 3200 MHz C15-17-17-17-35
Video Card(s) MSI GeForce RTX 2060 GAMING Z 6G
Storage Samsung 970 EVO NVMe M.2 500 GB, SanDisk Ultra II 480 GB
Display(s) Samsung SyncMaster C27H711 OC refresh rate 110Hz
Case Phantek Eclipse P400S (PH-EC416PS)
Audio Device(s) Creative Labs Sound Blaster Audigy Rx
Power Supply EVGA 850 BQ
Mouse SteelSeries Rival 310
Keyboard Logitech G G413 Silver
Software Windows 10 Professional 64-bit v1903
CU continues to pack 64 stream processors
The other change was the CU has supposedly been broken up into 2x32 unlike Fuji/Vega (8x5x2x32 = 2560)
 
Joined
Jan 13, 2015
Messages
46 (0.03/day)
Raja did *something* in AMD for sure. Head of Research or Engineering Team or whatever, he couldn't *just* reiterate GCN for endless generations. I would go that far to say that he probably developed at least a base for Next architecture (so say many others). In Intel, his influence also won't be felt for few years... To consumers, that is.
 
Joined
Nov 11, 2004
Messages
3,068 (0.58/day)
Location
Formosa
System Name Overlord Mk MVII
Processor AMD Ryzen 7 1700 @ 3.85GHz
Motherboard Asus Prime X370-Pro
Cooling Corsair H110i
Memory 16GB Corsair Vengeance LPX 3200
Video Card(s) Galax GeForce GTX 1080 EXOC-SNPR
Storage 1TB WD Black NVMe (2018), 500GB Crucial MX200, 1TB WD Blue 3D NAND
Display(s) Asus PG27AQ
Case Corsair Carbide 400Q
Power Supply Corsair AX860i
Mouse Logitech G500s
Keyboard Wooting Two
Software Windows 10
Benchmark Scores https://valid.x86.fr/h01zlr
I'm not planning on getting a new graphics card this year anyhow, but I do hope AMD can come up with something competitive at least, as it's much needed.
Monopolies aren't good for anyone, as it normally means higher prices, slower innovation and poor selection.
Nvidia might not be a monopoly, but with their performance lead on the higher end of the market, they might as well be.
Here's also fingers crossed that Intel will bring out something competitive when they launch their GPUs.
I long for the days when there were half a dozen competitive GPU makers, but that was a very long time ago and before they were called GPUs...
 
Joined
Oct 1, 2006
Messages
4,321 (0.93/day)
Location
Hong Kong
Processor Ryzen Threadripper 1950X
Motherboard X399 Fatal1ty Professional Gaming
Cooling Custom loop water, 3x 420 Rad
Memory Team T-FORCE XCALIBUR RGB 4000 @3466 CL14
Video Card(s) Zotac RTX 2080 Ti
Storage Samsung 960 Evo 512GB 3-way Raid 0
Display(s) Samsung C27HG70
Case Thermaltake Level 20 XT
Audio Device(s) Xonar Essence STX II
Power Supply FSP Aurum PT 1200W
Software Windows 10 Pro 64-bit
8 SEs ? unbelievable !
Yeah it would require substantial change to the front end.
Remember the 128ROP BS that pop up around Vega 20? This can easily be another round of BS before launch.
 
Joined
Mar 23, 2016
Messages
2,144 (1.82/day)
Processor Ryzen 5 2600X
Motherboard MSI B350M Gaming Pro
Cooling Scythe Kotetsu with AM4 bracket
Memory PNY Anarchy-X XLR8 Red DDR4 3200 MHz C15-17-17-17-35
Video Card(s) MSI GeForce RTX 2060 GAMING Z 6G
Storage Samsung 970 EVO NVMe M.2 500 GB, SanDisk Ultra II 480 GB
Display(s) Samsung SyncMaster C27H711 OC refresh rate 110Hz
Case Phantek Eclipse P400S (PH-EC416PS)
Audio Device(s) Creative Labs Sound Blaster Audigy Rx
Power Supply EVGA 850 BQ
Mouse SteelSeries Rival 310
Keyboard Logitech G G413 Silver
Software Windows 10 Professional 64-bit v1903
Here's a better rendition of the 8 Streaming Engines.
 
Joined
Jun 19, 2010
Messages
137 (0.04/day)
Location
North-Rhine-Westphalia
Processor IvyBridge Xeon E3-1230v2 8 Threads
Motherboard B75
Cooling Arctic FreezerXtreme
Memory 2x 8GB DDR3-1600
Video Card(s) GTX 960 2GB
Storage 256GB SSD
Display(s) 1080p 32" TV
Audio Device(s) USB Headset
Power Supply beQuiet F1 600W 80+Gold
Mouse yes
Keyboard yes
Software Win10
1. AMD has as far as i know nearly no pain when using ROP-Blending, so at that side it has nearly no Problems with bandwidth, opposing to that Nvidia looses a good bunch when using ROP-Blending.
2. AMD is very flexible in the wired amount of ROP, so they could´ve used 128 ROP with Hawaii if they wanted, they´ve seen no need for that by now, even for the Radeon VII, or MI60 they didn´t.
3. Even Nvidia is shy for using 8 Geo-Engines because the wirering will owerwhelm the chip with nearly no efficiency-gain
4. Navi is GCN and more than 4 Shader-Arrays are forbidden in GCN.

The picture on the bottom is changeable to avec (with) blending. The Vegas will be over 100 for the upper 4 numbers, wich is no bad at all.
It´s the newest one Mr Triolet made before his departure to AMD and later to Intels Graphic division.
 
Joined
Apr 10, 2019
Messages
1 (0.01/day)
Processor Intel i7 4790K delid + lapping [4.8@1.31V core/4.4@1.20V uncore]
Motherboard MSI Z97M Gaming
Cooling NZXT Kraken X41 + 2x Everflow 2600RPM
Memory 4x4GB G.Skill TridentX 2400 CL10
Video Card(s) Sapphire Vega 64 mod LC BIOS + 142% PowerPlay table + Kraken X61 + 2x FHP141 [1712/1150@1.25V]
Storage Samsung 840EVO 250GB | HGST 4TB
Display(s) LG IPS 24" 60Hz > 74Hz overclock
Case Bitfenix Aegis red
Audio Device(s) Logitech Z906 5.1 THX
Power Supply XFX XTR 750 Gold
Mouse Roccat Kone Pure red
Keyboard Gamdias Ares
Software Windows 10 64bits
Joined
Jun 10, 2014
Messages
1,562 (0.85/day)
AMD's 7 nm "Navi 10" silicon may finally address two architectural shortcomings of its performance-segment GPUs, memory bandwidth, and render-backends (deficiency thereof).
Neither memory bandwidth nor render backends are shortcomings of GCN. First of all, if ROPs were a bottleneck, they would have easily added more. Secondly, GCN cards have plenty of memory bandwidth vs. their Nvidia counterparts;
Radeon VII (1 TB/s) vs. RTX 2080 (448 GB/s)
Vega 64 (484 GB/s) vs. GTX 1080 (320 GB/s)
RTX 580 (256 GB/s) vs. GTX 1060 (192 GB/s)
 
  • Like
Reactions: M2B

M2B

Joined
Jun 2, 2017
Messages
201 (0.27/day)
Location
Iran
Processor Intel Core i5-8600K @4.9GHz
Motherboard MSI Z370 Gaming Pro Carbon
Cooling Cooler Master MasterLiquid ML240L RGB
Memory XPG 8GBx2 - 3200MHz CL16
Video Card(s) Asus Strix GTX 1080 OC Edition 8G 11Gbps
Storage 2x Samsung 850 EVO 1TB
Display(s) BenQ PD3200U
Case Thermaltake View 71 Tempered Glass RGB Edition
Power Supply EVGA 650 P2
Neither memory bandwidth nor render backends are shortcomings of GCN. First of all, if ROPs were a bottleneck, they would have easily added more. Secondly, GCN cards have plenty of memory bandwidth vs. their Nvidia counterparts;
Radeon VII (1 TB/s) vs. RTX 2080 (448 GB/s)
Vega 64 (484 GB/s) vs. GTX 1080 (320 GB/s)
RTX 580 (256 GB/s) vs. GTX 1060 (192 GB/s)
It doesn't even need that much knowledge to understand these simple facts.
Just looking at benchmarks and comparing AMD cards to their Nvidia rivals in different resolutions is enough.
AMD generally puts more bandwidth on their cards simply because they can't compete head to head and need more raw resourses to do so.
 
Joined
Feb 22, 2009
Messages
244 (0.06/day)
Location
Grand Prairie Texas
System Name Little Girl
Processor Intel Q9650 @ 3.6GHz
Motherboard Gigabyte x48 DQ6
Cooling liquid cooling
Memory 4gb (2x2) OCZ DDR2 PC2-9200
Video Card(s) Gigabyte HD6950 unlock to Asus 6970 specs
Storage Crucial CT128M225 128gb SSD
Display(s) Acer 27" LCD @ 2048x1152
Case DIY (spit & glue, ducktape, cardboard)
Audio Device(s) On-board HD Audio
Power Supply ABS Tagan 850w
Software Win7 64bit
AMD still struggles to compete with last years' card. It's over for AMD. It's over. I'm an Nvidia believer now.
 
Joined
Jun 3, 2010
Messages
445 (0.13/day)
In Pixel Fill-rate the Radeon VII actually beats the 2080, but when you look at Geometry it is far behind.
The ROP count might not be the actual issue.
Rop count is indeed the issue, otherwise color compression scores would be higher. The shaders as a total cannot be pipelined more than the transmitted data packets. It is a simple modem with all the bells and whistles that make it tick. If you select 4-byte packets, the router is overloaded. 8&16-byte packing must be the basic unit count for maximum effect.
Neither memory bandwidth nor render backends are shortcomings of GCN. First of all, if ROPs were a bottleneck, they would have easily added more. Secondly, GCN cards have plenty of memory bandwidth vs. their Nvidia counterparts;
Radeon VII (1 TB/s) vs. RTX 2080 (448 GB/s)
Vega 64 (484 GB/s) vs. GTX 1080 (320 GB/s)
RTX 580 (256 GB/s) vs. GTX 1060 (192 GB/s)
Yes, but for latency reasons all cannot be utilised in short shaders - that all changes when shader packing api is integrated into the Vulkan pipeline. Timothy Lottes did much on that end.
1. AMD has as far as i know nearly no pain when using ROP-Blending, so at that side it has nearly no Problems with bandwidth, opposing to that Nvidia looses a good bunch when using ROP-Blending.
2. AMD is very flexible in the wired amount of ROP, so they could´ve used 128 ROP with Hawaii if they wanted, they´ve seen no need for that by now, even for the Radeon VII, or MI60 they didn´t.
3. Even Nvidia is shy for using 8 Geo-Engines because the wirering will owerwhelm the chip with nearly no efficiency-gain
4. Navi is GCN and more than 4 Shader-Arrays are forbidden in GCN.

The picture on the bottom is changeable to avec (with) blending. The Vegas will be over 100 for the upper 4 numbers, wich is no bad at all.
It´s the newest one Mr Triolet made before his departure to AMD and later to Intels Graphic division.
  1. Yes, but that happened due to rop backends having their own seperate caches, with Vega AMD reverts to the same in-cache rop bandwidth amplification - that is more bandwidth for cacheable operations, but due to common cache architecture, buffer overflows lead to cataclysmic performance loss. It is like reverting from dual cores to single core hyperthreading.
  2. AMD is improving upon bitpacking. They already had the most integrated rop pipeline since Cayman 6900's. The depreciation of rop functions made that obsolete, now everything is done in shaders and common buffers. That lead to a general high-cache low-shader design, the same as Nvidia. This is not much to say about efficiency since there are better alternatives to pushing simple pixels; however as it stands quality pixels aren't as equitable as shader effects. You just wouldn't base a pick upon say, rapid packed math - although that is just the thing to fit 4K into a 150w console form factor. Initiating writes take up memory interface latency, so two writes for the instance of one is fundamentally efficient.
 
Joined
Mar 23, 2016
Messages
2,144 (1.82/day)
Processor Ryzen 5 2600X
Motherboard MSI B350M Gaming Pro
Cooling Scythe Kotetsu with AM4 bracket
Memory PNY Anarchy-X XLR8 Red DDR4 3200 MHz C15-17-17-17-35
Video Card(s) MSI GeForce RTX 2060 GAMING Z 6G
Storage Samsung 970 EVO NVMe M.2 500 GB, SanDisk Ultra II 480 GB
Display(s) Samsung SyncMaster C27H711 OC refresh rate 110Hz
Case Phantek Eclipse P400S (PH-EC416PS)
Audio Device(s) Creative Labs Sound Blaster Audigy Rx
Power Supply EVGA 850 BQ
Mouse SteelSeries Rival 310
Keyboard Logitech G G413 Silver
Software Windows 10 Professional 64-bit v1903

btarunr

Editor & Senior Moderator
Staff member
Joined
Oct 9, 2007
Messages
36,956 (8.66/day)
Location
Hyderabad, India
Processor AMD Ryzen 7 2700X
Motherboard MSI B450 Gaming Pro Carbon AC
Cooling AMD Wraith Prism
Memory 2x 16GB Corsair Vengeance LPX DDR4-3000
Video Card(s) Colorful iGame GTX 1070 Ti Vulcan X
Storage Western Digital Black NVMe 512GB
Display(s) Samsung U28D590 28-inch 4K UHD
Case Corsair Carbide 100R
Audio Device(s) Creative Sound Blaster Recon3D PCIe
Power Supply Antec EarthWatts Pro Gold 750W
Mouse Razer Abyssus
Keyboard Microsoft Sidewinder X4
Software Windows 10 Pro
Joined
Nov 3, 2011
Messages
272 (0.10/day)
System Name Fractal Define R5 |Fractal Define R6 | HP X360 Envy 15in
Processor Intel Core i7-7820X 4.5 Ghz | Intel Core i7-9900K 5 Ghz | AMD Ryzen 5 2500U
Motherboard ASUS ROG Strix X299-E Gaming | MSI Z390 Gaming Pro Carbon AC | HP X360 Envy 15in
Memory 32GB 3200 Mhz | 32GB 3600Mhz | 16 GB 2400Mhz
Video Card(s) MSI Geforce GTX 1080 Ti FE| MSI Geforce RTX 2080 Ti GX TRIO| AMD RX Vega 8
Display(s) 3X Samsung 23in LED | LG 32in 4K HDR FreeSync| built-in 15 inch LED touch screen.
Case Fractal R5 tempered glass | Fractal R6 tempered glass| HP X360 Envy 15in
Audio Device(s) ASUS Xonar D2X | Creative Sound Blaster Z| built-in
Power Supply Seasonic 750 watts| Seasonic 1000 watts | 65 watts
Software All machines has MS Windows 10 Pro 190x version.
Neither memory bandwidth nor render backends are shortcomings of GCN. First of all, if ROPs were a bottleneck, they would have easily added more. Secondly, GCN cards have plenty of memory bandwidth vs. their Nvidia counterparts;
Radeon VII (1 TB/s) vs. RTX 2080 (448 GB/s)
Vega 64 (484 GB/s) vs. GTX 1080 (320 GB/s)
RTX 580 (256 GB/s) vs. GTX 1060 (192 GB/s)
NVIDIA's GPUs has superior memory compression.
 
Joined
Sep 17, 2014
Messages
9,038 (5.21/day)
Location
Too Long to fit in a single line here.
Processor i7 8700k 4.7Ghz @ 1.26v
Motherboard AsRock Fatal1ty K6 Z370
Cooling beQuiet! Dark Rock Pro 3
Memory 16GB Corsair Vengeance LPX 3200/C16
Video Card(s) MSI GTX 1080 Gaming X @ 2100/5500
Storage Samsung 850 EVO 1TB + Samsung 830 256GB + Crucial BX100 250GB + Toshiba 1TB HDD
Display(s) Eizo Foris FG2421
Case Fractal Design Define C TG
Power Supply EVGA G2 750w
Mouse Logitech G502 Protheus Spectrum
Keyboard Sharkoon MK80 (Brown)
Software W10 x64
AMD still struggles to compete with last years' card. It's over for AMD. It's over. I'm an Nvidia believer now.
It was over when they announced they were going to 'focus on midrange' with Polaris. The hidden message there is 'we can't keep up', and every high end release after that simply confirmed it.

They're stuck, and they have been stuck since Hawaii, its what I've been seeing and saying ever since. Fury X was not competitive and HBM for the gaming segment was a stopgap measure to keep GCN in the game, not something you do if you like a healthy profit margin. Vega simply didn't perform as it should have (or should have been ready for launch when Nvidia launched GP104), and VII is saved by the 7nm node; 'sorta'.

Beyond that, there is nothing to give. At the same time, they haven't got the technology/performance lead that provides the necessary time to complete revamp GCN from the ground up. Ironically, its rather similar to Intel's current CPU roadmap. Perhaps that is part of the rationale for Intel to focus on GPU as well; perhaps they've seen you can't be leading all the time without creating new risk (stagnation).
 
Last edited:
Joined
Jun 10, 2014
Messages
1,562 (0.85/day)
NVIDIA's GPUs has superior memory compression.
Memory compression really only helps for sparse texture/buffer data, and while Nvidia employs more advanced compression than AMD, it doesn't account for 30-50% more effective bandwidth.
 
Joined
Jun 3, 2010
Messages
445 (0.13/day)
Memory compression really only helps for sparse texture/buffer data, and while Nvidia employs more advanced compression than AMD, it doesn't account for 30-50% more effective bandwidth.
AMD is band-limited for texture sampling by the pixel clock rate. They have an advantage, but not in the pixel shader pathway. Anisotropic filtering is also cache-limited as per every 4th clock cycle, so to start every pixel from the pixel shader - it is a significant difference having 2x rops like Nvidia, or not.
The other method is the compute shader: it does not throttle tmus, but the tmu cache is still quarter rate per 16x af and it does not work like the pixel shader. One benefit of the pixel shader is, it is fully pipelined: you don't go full netburst-prescott disaster; it is pipelined, every data is memory mapped and you get the usual benefits. The gain of compute shader is that it is the non-native version of this pipeline - whether the developer can benefit from his own custom pipeline is his doing. While the pixel shader is streamlined per memory accesses(reads) for less latency by default, the compute shader has the benefit of write streamlining, since caches are a faster storage medium than memory. It is just a coincidence which fits the attempted end result - using caches instead of vram has the added benefit of cutting out the middleman oem manufacturers at setting the memory timing parameters in their premium gpu lines; caches are a more uniform solution than custom gddr dies.
 
Joined
Nov 3, 2011
Messages
272 (0.10/day)
System Name Fractal Define R5 |Fractal Define R6 | HP X360 Envy 15in
Processor Intel Core i7-7820X 4.5 Ghz | Intel Core i7-9900K 5 Ghz | AMD Ryzen 5 2500U
Motherboard ASUS ROG Strix X299-E Gaming | MSI Z390 Gaming Pro Carbon AC | HP X360 Envy 15in
Memory 32GB 3200 Mhz | 32GB 3600Mhz | 16 GB 2400Mhz
Video Card(s) MSI Geforce GTX 1080 Ti FE| MSI Geforce RTX 2080 Ti GX TRIO| AMD RX Vega 8
Display(s) 3X Samsung 23in LED | LG 32in 4K HDR FreeSync| built-in 15 inch LED touch screen.
Case Fractal R5 tempered glass | Fractal R6 tempered glass| HP X360 Envy 15in
Audio Device(s) ASUS Xonar D2X | Creative Sound Blaster Z| built-in
Power Supply Seasonic 750 watts| Seasonic 1000 watts | 65 watts
Software All machines has MS Windows 10 Pro 190x version.
Memory compression really only helps for sparse texture/buffer data, and while Nvidia employs more advanced compression than AMD, it doesn't account for 30-50% more effective bandwidth.
Nvidia has robust immediate mode tile cache render since Maxwell.

123572
 
Last edited:
Joined
Mar 10, 2010
Messages
6,388 (1.89/day)
Location
Manchester uk
System Name RyzenGtEvo
Processor Amd R5 2600X@4.1
Motherboard Crosshair hero7 @bios 2304
Cooling 360EK extreme rad+ 360$EK slim all push, cpu Monoblock Gpu full cover all EK
Memory Corsair Vengeance Rgb pro 3466cas16 16Gb in two sticks.
Video Card(s) Sapphire refference Rx vega 64 EK waterblocked
Storage Samsung Nvme Pg981, silicon power 1Tb samsung 840 basic as a primocache drive for, WD2Tbgrn +3Tbgrn,
Display(s) Samsung UAE28"850R 4k freesync, LG 49" 4K 60hz ,Oculus
Case Lianli p0-11 dynamic
Audio Device(s) Xfi creative 7.1 on board ,Yamaha dts av setup, corsair void pro headset
Power Supply corsair 1200Hxi
Mouse Roccat Kova
Keyboard Roccat Iksu force fx
Software Win 10 Pro
Benchmark Scores 8056 vega 3dmark timespy
If this is true, there might be some hope for RTG.
Assuming Raja actually did his job before he left for Intel, which was to make GCN more scale-able.
The Geometry limit was the Achilles heel of GCN in terms of gaming performance.

In Pixel Fill-rate the Radeon VII actually beats the 2080, but when you look at Geometry it is far behind.
The ROP count might not be the actual issue.


While I Agree on the Geometry comments i think your chart shows it could still do with more rops ,perhaps for navi 20 though eh.

Been thinking about that name Navi ,I got New Architecture for Vertical Integration, i'm thinking :);) it'll be modular obviously and obviously adaptable.

The star's a coincidence.
 
Joined
Oct 14, 2017
Messages
104 (0.17/day)
System Name Lightning
Processor 4790K
Motherboard asrock z87 extreme 3
Cooling hwlabs black ice 20 fpi radiator, cpu mosfet blocks, MCW60 cpu block, full cover on 780Ti's
Memory corsair dominator platinum 2400C10, 32 giga, DDR3
Video Card(s) 2x780Ti
Storage intel S3700 400GB, samsung 850 pro 120 GB, a cheep intel MLC 120GB, an another even cheeper 120GB
Display(s) eizo foris fg2421
Case 700D
Audio Device(s) ESI Juli@
Power Supply seasonic platinum 1000
Mouse mx518
Software Lightning v2.0a
this suckx :x only 2500 shaders ? whay ?! radeon 7 gived 4000 :x
it a joke, 780 Ti from zillion years ago have 2800 :x
 
Joined
Jun 10, 2014
Messages
1,562 (0.85/day)
A chip with 2560 SPs to match RTX 2070 would require a fairly large efficiency gain, which would be appreciated, but I remain sceptical.
 

Aquinus

Resident Wat-man
Joined
Jan 28, 2012
Messages
11,220 (4.16/day)
Location
Concord, NH
System Name Kratos
Processor Intel Core i7 3930k @ 4.5Ghz
Motherboard ASUS P9X79 Deluxe
Cooling Corsair H100i V2
Memory G.Skill DDR3-2133, 16gb (4x4gb) @ 9-11-10-28-108-1T 1.65v
Video Card(s) Sapphire AMD Radeon RX Vega 64
Storage 2x120Gb SATA3 SSD Raid-0, 4x1Tb RAID-5, 1x500GB, 1x512GB Samsung 960 Pro NVMe
Display(s) 1x LG 27UD69P (4k), 2x Dell S2340M (1080p)
Case Antec 1200
Audio Device(s) Onboard Realtek® ALC898 8-Channel High Definition Audio
Power Supply Seasonic 1000-watt 80 PLUS Platinum
Mouse Logitech G602
Keyboard Rosewill RK-9100
Software Ubuntu 18.04 (5.1.3 Mainline Kernel)
Benchmark Scores Benchmarks aren't everything.
I always felt that the way Hawaii was setup was more conducive to doing GPGPU as opposed to rendering. It really had quite a lot of power, but rarely would you really see it taken advantage of. To be honest, nVidia is more like this, where each SM didn't contain quite as many shaders as a GCN CU. If AMD has made their shaders efficient enough (and I'm feeling fairly confident that they have,) then this should help mitigate some issues people have been attributing to GCN. Honestly, GCN really isn't a bad architecture. It's just how these GPUs have been designed because raw compute power doesn't always translate to better graphics performance.

Honestly, consider for a moment that a 390 has more double-precision compute power than a 2080 Ti which would be hilarious beside the fact that it gets you absolutely nothing in games.
 
Top