[RM][Fudzilla]GTX 1660 Ti, GTX 1660 and the GTX 1650 - Page 4 - Overclock.net - An Overclocking Community

Forum Jump: 

[RM][Fudzilla]GTX 1660 Ti, GTX 1660 and the GTX 1650

Reply
 
Thread Tools
post #31 of 81 (permalink) Old 02-09-2019, 07:22 PM
New to Overclock.net
 
tajoh111's Avatar
 
Join Date: Aug 2013
Posts: 1,341
Rep: 212
Quote: Originally Posted by epic1337 View Post
just take note that 590 is overpriced compared to either 580 or 570, not to mention the performance gap between GTX1060 and RX590 is small.
if you're implying that GTX1660Ti would only be 15%~20% faster than GTX1060 then i doubt a price of $280 would make it look good.







lets do some mathematical comparison between Turing and Pascal:
RTX2060 = 1920:120:48 @ 1680Mhz | 192bit GDDR6 @ 336GB/s | relative performance 116.9% (based on TPU)
GTX1070 = 1920:120:64 @ 1683Mhz | 256bit GDDR5 @ 256GB/s | relative performance 100% (based on TPU)
this implies that Turing is approximately 17% faster than Pascal.

now lets apply it to 1660Ti:
GTX1660Ti = 1536:96:48 @ 1770Mhz? | 192bit GDDR6 @ 288GB/s?
GTX1060 = 1280:80:48 @ 1708Mhz | 192bit GDD5 @ 192GB/s

lets ignore the clock speed increase
1536/1280 = +20%
Turing IPC = +17%
1.2 * 1.17 = 1.40 or +40%

then if we apply 40% on top of GTX1060 in that image, we'd see it sit right beside RX Vega 56.
but then, if it really does perform close to Vega 56... the price may as well be $300, no wonder they priced the RTX2060 FE at $350.


in any case, when i posted this, i very much already did the mathematical estimation.
This is around where I have it too using a very similar methodology. AdoredTV calculations are not terribly correct.

The IPC gain when bandwidth is not an issue is substantial with turing.

Considering how similar a 1920 Turing part is to a 2560 Pascal part, there was certainly an increase in IPC.

With the ROP and bandwidth Ratio particularly high with the GTX 1660 ti, it's performance per tflop should be the highest in the product stock if it indeed gets GDDR6.

A 27% percent gain would basically be terrible. That would imply the 20% gain in shaders + 4% increase is where all the gains are from. That would imply no IPC gain.

This card being having so much more bandwidth than a GTX 1060 should open up a significant amount of performance.

https://www.techpowerup.com/reviews/...GDDR5X/34.html

[img/https://tpucdn.com/reviews/KFA2/GTX_1060_6_GB_GDDR5X/images/overclocked-performance.png[/img]

Worst overclock of all the GTX 1060s, but the greatest gain in terms of overclock by 9% which is all due to memory overclocking. 20% more shaders + 4 percent higher clocks + GDDR 6 = 30% gain at the very least.

With the 5% faster clocks on top of the better tflop/bandwidth ratio, there should be only about a 15-17% difference in performance between the RTX 2060 and gtx 1660 ti, if the only thing they do in the transition is remove the ray tracing hardware.
tajoh111 is offline  
Sponsored Links
Advertisement
 
post #32 of 81 (permalink) Old 02-09-2019, 07:53 PM
New to Overclock.net
 
matthew87's Avatar
 
Join Date: May 2010
Posts: 177
Rep: 4
Quote: Originally Posted by epic1337 View Post
just take note that 590 is overpriced compared to either 580 or 570, not to mention the performance gap between GTX1060 and RX590 is small.
if you're implying that GTX1660Ti would only be 15%~20% faster than GTX1060 then i doubt a price of $280 would make it look good.







lets do some mathematical comparison between Turing and Pascal:
RTX2060 = 1920:120:48 @ 1680Mhz | 192bit GDDR6 @ 336GB/s | relative performance 116.9% (based on TPU)
GTX1070 = 1920:120:64 @ 1683Mhz | 256bit GDDR5 @ 256GB/s | relative performance 100% (based on TPU)
this implies that Turing is approximately 17% faster than Pascal.

now lets apply it to 1660Ti:
GTX1660Ti = 1536:96:48 @ 1770Mhz? | 192bit GDDR6 @ 288GB/s?
GTX1060 = 1280:80:48 @ 1708Mhz | 192bit GDD5 @ 192GB/s

lets ignore the clock speed increase
1536/1280 = +20%
Turing IPC = +17%
1.2 * 1.17 = 1.40 or +40%

then if we apply 40% on top of GTX1060 in that image, we'd see it sit right beside RX Vega 56.
but then, if it really does perform close to Vega 56... the price may as well be $300, no wonder they priced the RTX2060 FE at $350.


in any case, when i posted this, i very much already did the mathematical estimation.
This is NVidia we're talking about.

I wouldn't at all be surprised if the 1160Ti or whatever they call it has a locked BIOS with very low power limits to artificially restrict performance to a hardset threshold including over clocking. So too that they limit the GDDR6 speeds, which we know Turing benefits from substantially, to ensure there's a noticeable gulf between it and the RTX cards and ensure there's a very clear and distinct performance gap.

Nvidia would be shooting themselves in the foot to release a card - 1160Ti - that with some overclocking and decent custom designs from card vendors could match the RTX 2060 in games. Nvidia don't just want the RTX to be about ray tracing etc, they want them to be their 'premium' brand and cards. That no GTX, or non RTX card will match them across the board, it's about maintaining the product segmentation.

And measuring GPU performance in terms of 'IPC' is very flawed given their parallelism and number of independent functions like ROPs, rasterizers , cores.
matthew87 is offline  
post #33 of 81 (permalink) Old 02-09-2019, 08:36 PM
Otherworlder
 
epic1337's Avatar
 
Join Date: Feb 2011
Posts: 7,065
Rep: 213
lets look at it from a different perspective, Pascal had a healthy performance gap between major card divisions due to it's wide difference between die configuration.
GTX1080Ti = 3584:224:88 = 144% performance
GTX1080 = 2560:160:64 = 120% performance
GTX1070 = 1920:120:64 = 100% performance
GTX1060 = 1280:80:48 = 75% performance

but if we look at Turing, you'd pretty much see Nvidia shot themselves on the foot.
RTX2080 = 2944:184:64 = 134% performance
RTX2070 = 2304:144:64 = 115% performance
RTX2060 = 1920:120:48 = 100% performance
GTX1660Ti = 1536:96:48 ≈ 89% performance

lets not talk about the RTX gimmick, just look at those die configurations.
2560 ► 3584 «VS» 2304 ► 2944 = lets take a look at these cards, you can see tha 2944 is significantly smaller than 3584, yet 2304 is just a bit smaller than 2560.
1920 ► 2560 «VS» 1920 ► 2304 = then theres these cards, its pretty obvious that 2080 and 2070 would have a really small performance gap.
1280 ► 1920 «VS» 1536 ► 1920 = even worse yet is this gap, 1536 is much higher than 1280, the gap between 1660Ti and 2060 would be really really small.

Quote: Originally Posted by matthew87 View Post
And measuring GPU performance in terms of 'IPC' is very flawed given their parallelism and number of independent functions like ROPs, rasterizers , cores.
oh? is "image per compute cycle" flawed now?

trolling an adult is very dangerous, don't try it at home nor at work. you don't want to play tag with a rabid man.

Last edited by epic1337; 02-09-2019 at 08:52 PM.
epic1337 is offline  
Sponsored Links
Advertisement
 
post #34 of 81 (permalink) Old 02-09-2019, 08:48 PM
⤷ αC
 
AlphaC's Avatar
 
Join Date: Sep 2012
Posts: 9,934
Rep: 803
Quote: Originally Posted by matthew87 View Post
This is NVidia we're talking about.

I wouldn't at all be surprised if the 1160Ti or whatever they call it has a locked BIOS with very low power limits to artificially restrict performance to a hardset threshold including over clocking. So too that they limit the GDDR6 speeds, which we know Turing benefits from substantially, to ensure there's a noticeable gulf between it and the RTX cards and ensure there's a very clear and distinct performance gap.

Nvidia would be shooting themselves in the foot to release a card - 1160Ti - that with some overclocking and decent custom designs from card vendors could match the RTX 2060 in games. Nvidia don't just want the RTX to be about ray tracing etc, they want them to be their 'premium' brand and cards. That no GTX, or non RTX card will match them across the board, it's about maintaining the product segmentation.

And measuring GPU performance in terms of 'IPC' is very flawed given their parallelism and number of independent functions like ROPs, rasterizers , cores.

It's very likely to be 80% of a RTX 2060 or better since it's the same memory bus and 6GB GDDR6.


RTX 2060 has ray tracing and FP16 performance as a selling point as well as VR-Link. Nvidia putting out a GTX 1660 Ti just allows them to EOL the GTX 1070 completely , with the GTX 1660 replacing GTX 1060 6GB and GTX 1060 3GB.


I don't understand the idea of GTX 1660 TI being not much faster than a RX 590. RX 590 is 10% over RX 580 at best. This is going to be 20% over GTX 1060 6GB just on CUDA cores alone, with all the memory bus bottlenecking removed.


Out of the box clocks on the Palit GTX 1660 Ti are already matching that of RTX 2060 aftermarket boards so there's no real reason to think GPU clocks are going to be terrible.

► Recommended GPU Projects: [email protected] , [email protected] (FP64) (AMD moreso) ► Other notable GPU projects: [email protected] (Nvidia), GPUGrid (Nvidia) ► Project list



Last edited by AlphaC; 02-09-2019 at 08:56 PM.
AlphaC is online now  
post #35 of 81 (permalink) Old 02-09-2019, 08:56 PM
New to Overclock.net
 
matthew87's Avatar
 
Join Date: May 2010
Posts: 177
Rep: 4
You're ignoring memory config, it's not just about the die. Look at pascals performance scaling with nothing but memory overclocking.

1080 GTX = 256bit 10Gbps
1070 GTX = 256bit 8Gbps
1060 GTX = 192bit 8Gbps

The 1060 had 2GB less memory than the other two, and 33% smaller memory bus.

At higher resolutions the difference was more pronounced than at 1080p, as that's when that increased memory pool and bandwidth really comes into play.

So sure, perhaps at 1080p the 1160 will only be 12-15% slower then the 2060 RTX, but at 1440p and above I expect the difference will be far more pronounced. Nvidia will do what any company in their position will do, protect their product stack.
matthew87 is offline  
post #36 of 81 (permalink) Old 02-09-2019, 09:01 PM
Otherworlder
 
epic1337's Avatar
 
Join Date: Feb 2011
Posts: 7,065
Rep: 213
Quote: Originally Posted by matthew87 View Post
You're ignoring memory config, it's not just about the die. Look at pascals performance scaling with nothing but memory overclocking.

1080 GTX = 256bit 10Gbps
1070 GTX = 256bit 8Gbps
1060 GTX = 192bit 8Gbps

The 1060 had 2GB less memory than the other two, and 33% smaller memory bus.

At higher resolutions the difference was more pronounced than at 1080p, as that's when that increased memory pool and bandwidth really comes into play.

So sure, perhaps at 1080p the 1160 will only be 12-15% slower then the 2060 RTX, but at 1440p and above I expect the difference will be far more pronounced. Nvidia will do what any company in their position will do, protect their product stack.

exactly, Pascal also had large difference between memory configurations, but Turing doesn't have that.
all of Turing are equipped with GDDR6, the only difference is the lower tier cards are 192bit instead of 256bit, but with their core configuration 256bit would've been excessive.
oh and for the record, both 2060 and 1660Ti comes equipped with 192bit 6GB GDDR6, if theres gonna be any difference then maybe the IMC clock speed.


if by protecting their product stack you mean how their die configuration's gaps have shrunk, then sure i suppose they're protecting their product stack really well.

trolling an adult is very dangerous, don't try it at home nor at work. you don't want to play tag with a rabid man.

Last edited by epic1337; 02-09-2019 at 09:18 PM.
epic1337 is offline  
post #37 of 81 (permalink) Old 02-09-2019, 09:29 PM
New to Overclock.net
 
m4fox90's Avatar
 
Join Date: Sep 2017
Posts: 191
Rep: 2
NVIDIA's product stack sure is something. So many options just confuse consumers and the price points are just making them angry. No wonder the stock is tanking.

Silence in the Snow
(13 items)
CPU
Ryzen 1600X
Motherboard
ASRock X370 Taichi
GPU
GTX 1080 FTW2
RAM
G.Skill RipJaws V 16 GB @3466
Hard Drive
Samsung 960 EVO 500GB
Hard Drive
Seagate Barracuda 3TB
Hard Drive
Seagate FireCuda 2TB
Power Supply
Seasonic SnowSilent
Cooling
Silverstone TD-02E
Case
Phanteks Enthoo Pro M
Operating System
Windows 10 Home
Monitor
Predator XB271HU
Monitor
ASUS PB 258Q
▲ hide details ▲
m4fox90 is offline  
post #38 of 81 (permalink) Old 02-09-2019, 10:03 PM
⤷ αC
 
AlphaC's Avatar
 
Join Date: Sep 2012
Posts: 9,934
Rep: 803
Quote: Originally Posted by matthew87 View Post
You're ignoring memory config, it's not just about the die. Look at pascals performance scaling with nothing but memory overclocking.

1080 GTX = 256bit 10Gbps
1070 GTX = 256bit 8Gbps
1060 GTX = 192bit 8Gbps

The 1060 had 2GB less memory than the other two, and 33% smaller memory bus.

At higher resolutions the difference was more pronounced than at 1080p, as that's when that increased memory pool and bandwidth really comes into play.

So sure, perhaps at 1080p the 1160 will only be 12-15% slower then the 2060 RTX, but at 1440p and above I expect the difference will be far more pronounced. Nvidia will do what any company in their position will do, protect their product stack.

RTX 2060 has 192 bit bus GDDR6 for 336.0 GB/s. The GTX 1660 Ti has 192 bit bus GDDR6 clocked lower for 288.0 GB/s.


Regardless, the GTX 1070 Ti (256.3 GB/s) shows that anything less than GTX 1080 performance doesn't have a huge difference with that kind of bandwidth even on 256 bit GDDR5. GTX 1080's GDDR5X had 320.3 GB/s.


So unless it's over GTX 1070 Ti performance it's unlikely to matter. I doubt that's the case since RTX 2060 performs roughly on par with GTX 1070 Ti. It's probably around GTX 1070.

► Recommended GPU Projects: [email protected] , [email protected] (FP64) (AMD moreso) ► Other notable GPU projects: [email protected] (Nvidia), GPUGrid (Nvidia) ► Project list


AlphaC is online now  
post #39 of 81 (permalink) Old 02-11-2019, 07:53 AM
I dunno what I'm doing.
 
Blze001's Avatar
 
Join Date: Mar 2013
Posts: 1,356
Rep: 59
Quote: Originally Posted by m4fox90 View Post
NVIDIA's product stack sure is something. So many options just confuse consumers and the price points are just making them angry. No wonder the stock is tanking.
But it just works, though.

It is what it is.
Waterworld
(21 items)
CPU
i7-8700k
Motherboard
ASRock Fatal1ty Z370 Gaming-ITX/ac
GPU
GTX-1080ti Founder's Edition
RAM
Kingston HyperX
Hard Drive
Samsung M.2 950 PRO
Hard Drive
SanDisk Ultra II
Power Supply
Corsair RM650x
Cooling
Optimus V1 LGA-1151 Waterblock
Cooling
Nemesis GTS240
Cooling
Nemesis 120GTS
Cooling
EK D5 Revo
Cooling
Singularity Protium Pump Top/Case
Cooling
EKWB 10x0 Waterblock
Cooling
SavantPCs 240mm Reservoir
Cooling
Noctua NF-A12x25
Case
Fractal Nano S
Operating System
Windows 10 Pro
Monitor
Alienware AW3418DW
Keyboard
GMMK TKL
Mouse
Logitech G602
Audio
Creative X7
▲ hide details ▲
Blze001 is offline  
post #40 of 81 (permalink) Old 02-11-2019, 10:31 AM
H₂O Aficionado
 
Slaughtahouse's Avatar
 
Join Date: Jan 2013
Posts: 3,521
Rep: 117
https://wccftech.com/nvidia-geforce-...listings-leak/

-=]░▒▓███ [Official] NVIDIA GTX 780 Owner's Club ███▓▒░[=-

Stridsfordon 90
(20 items)
M1 Abrams
(18 items)
Asus G53
(0 items)
CPU
Intel 3930K
Motherboard
Asus Sabertooth X79
GPU
GTX 780 SC
RAM
Patriot Black Viper
Hard Drive
Samsung 840 SSD
Hard Drive
WD Black 7200 RPM
Power Supply
Corsair AX 850
Cooling
XSPC EX 360 Radiator
Cooling
XSPC Raystorm CPU Block
Cooling
XSPC Titan/780 GPU Block
Cooling
EK-XRES 100 Revo D5 - Acetal
Cooling
XSPC D5 Pump
Case
Fractal Design Meshify C
Operating System
Windows 10
Monitor
QNIX 27"
Keyboard
Logitech G710+
Mouse
Razer Death Adder 2013
Mousepad
Mouse Rug
Audio
Sennheiser HD 6XX
Audio
Bose Companion III's
CPU
i7 3930K
Motherboard
Asus Sabertooth X79
GPU
Geforce GTX 780 EVGA SC 3GB + XSPC Waterblock
RAM
Patriot Viper 3 16GB (4 x 4GB) DDR3 2133
Hard Drive
SAMSUNG 840 Series 250GB
Power Supply
Corsair AX850
Cooling
XIGMATEK Dark Knight II Night Hawk Edition
Cooling
XSPC 360mm radiator, Raystorm CPU block, D5 pump
Case
Corsair Vengeance Series C70 Military Green (Custom paint)
Operating System
Windows 10
Monitor
QNIX QX2710 27" 100hz
Monitor
Samsung 2233RZ 22" 120hz
Keyboard
Logitech g710+
Mouse
Razer DeathAdder 3.5G
Mouse
Mouse Rug
Audio
Bose Companion III's Speakers
Audio
Sennheiser HD 558 Headphones
Audio
Sennheiser HD 6XX Headphones
▲ hide details ▲
Slaughtahouse is offline  
Reply

Quick Reply
Message:
Options

Register Now

In order to be able to post messages on the Overclock.net - An Overclocking Community forums, you must first register.
Please enter your desired user name, your email address and other required details in the form below.
User Name:
If you do not want to register, fill this field only and the name will be used as user name for your post.
Password
Please enter a password for your user account. Note that passwords are case-sensitive.
Password:
Confirm Password:
Email Address
Please enter a valid email address for yourself.
Email Address:

Log-in



Currently Active Users Viewing This Thread: 2 (1 members and 1 guests)
rainzor
Thread Tools
Show Printable Version Show Printable Version
Email this Page Email this Page


Forum Jump: 

Posting Rules  
You may post new threads
You may post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off