From WikiChip
Difference between revisions of "nvidia"

(→‎Microarchitectures)
 
(26 intermediate revisions by 11 users not shown)
Line 16: Line 16:
 
}}-->
 
}}-->
 
}}
 
}}
−
'''NVIDIA Corporation''' is an American [[fabless]] semiconductor company that focuses largely on [[graphics processing unit]]s for the gaming industry. They also design various other chips, SoC, and consumer electronics for the mobile and automotive markets.
+
'''NVIDIA Corporation''' is an American [[fabless]] semiconductor company that focuses largely on [[graphics processing unit]]s for the gaming industry and [[machine learning]] applications. They also design various other chips, SoC, and consumer electronics for the mobile and automotive markets.
  
 
== GPU Families ==
 
== GPU Families ==
 +
{{collist
 +
| count = 3
 +
|
 +
* {{\|NV1}}
 +
* {{\|NV2}}
 +
* {{\|Riva TNT}}
 +
* {{\|Riva TNT2}}
 +
* {{\|Riva 128}}
 +
* {{\|GeForce 256}}
 +
 +
* {{\|GeForce 2}}
 +
* {{\|GeForce 2 Go}}
 +
* {{\|GeForce 3}}
 +
* {{\|GeForce 4}}
 +
* {{\|GeForce 4 Go}}
 +
* {{\|GeForce FX}}
 +
* {{\|GeForce FX Go 5}}
 +
* {{\|GeForce 6}}
 +
* {{\|GeForce Go 6}}
 +
 +
* {{\|GeForce 7}}
 +
* {{\|GeForce Go 7}}
 +
* {{\|GeForce 8}}
 +
* {{\|GeForce 8M}}
 +
* {{\|GeForce 9}}
 +
* {{\|GeForce 9M}}
 +
 +
* {{\|GeForce 10}}
 +
* {{\|GeForce 16}}
 +
* {{\|Geforce RTX 20}}
 +
* {{\|Geforce RTX 30}}
 +
* {{\|Geforce RTX 40}}
 +
* ''{{\|Geforce RTX 50}}''
 +
 +
* {{\|GeForce 100}}
 +
* {{\|GeForce 100M}}
 +
* {{\|GeForce 200}}
 +
* {{\|GeForce 200M}}
 +
* {{\|GeForce 300}}
 +
* {{\|GeForce 300M}}
 +
* {{\|GeForce 400}}
 +
* {{\|GeForce 400M}}
 +
* {{\|GeForce 500}}
 +
* {{\|GeForce 500M}}
 +
* {{\|GeForce 600}}
 +
* {{\|GeForce 600M}}
 +
 +
* {{\|GeForce 700}}
 +
* {{\|GeForce 700M}}
 +
* {{\|GeForce 800}}
 +
* {{\|GeForce 800M}}
 +
* {{\|GeForce 900}}
 +
* {{\|GeForce 900M}}
 +
 +
* {{\|Quadro}}
 +
* {{\|Quadro Go}}
 +
* {{\|Quadro FX}}
 +
* {{\|Quadro FX Go}}
 +
* {{\|Quadro NVS}}
 +
* ''{{\|Quadro RTX}}''
 +
 +
* {{\|Tesla}}
 +
* {{\|XGPU}}
 +
* {{\|GRID}}
 +
* {{\|RSX}}
 +
* {{\|NVS}}
 +
* ''{{\|DGX}}''
 +
}}
 +
 +
== CPU Families ==
 +
:; {{\|GoForce}}
 +
:* GoForce
 +
 +
{{collist
 +
| count = 3
 +
| width = 200px
 +
|
 +
:; {{\|Tegra}}
 +
* {{nvidia|tegra/2|Tegra 2}} • AP20H/AP25/T20/T25 (VEC4)
 +
* {{nvidia|tegra/3|Tegra 3}} (Kal-El) • T30/T30L/T33/AP33
 +
* {{nvidia|tegra/4|Tegra 4}} (Wayne) • T114 (VEC4)
 +
* {{nvidia|tegra/4i|Tegra 4i}} (Grey) • T148? (VEC4)
 +
* {{nvidia|tegra/k1|Tegra K1}} (Logan) • T124 • T132 ({{nvidia|Denver|l=arch}})
 +
* {{nvidia|tegra/x1|Tegra X1}} (Erista) • T210 (Maxwell)
 +
* {{nvidia|tegra/x1+|Tegra X1+}} (Mariko) • T214 (Maxwell)
 +
* {{nvidia|tegra/x2|Tegra X2}} (Parker) • T186 ({{nvidia|Denver|l=arch}} 2)
 +
* {{nvidia|tegra/xavier|Tegra Xavier}} • T194 ({{nvidia|Carmel|l=arch}}/Volta)
 +
* {{nvidia|tegra/orin|Tegra Orin}} • T234 (Hercules) • T239 (Drake)
 +
* {{nvidia|tegra/grace|Tegra Grace}} • T241 ([[Neoverse]] V2)
 +
* {{nvidia|tegra/thor|Tegra Thor}} • T264? ([[Neoverse]] V3AE)
 +
}}
 +
{{clear}}
 +
<!--
 +
<pre>
 +
Orin • T234 • GPU: "Ampere"; CPU: 12x ARM Cortex-A78AE (Hercules)
 +
Grace • T241 • GPU: "Hopper"; CPU: 72x ARM Neoverse V2 cores (AI & HPC)
 +
Atlan • T254? • GPU: "Ada Lovelace"; CPU: Grace-Nex, cancelled model 2022
 +
Thor • T264? • GPU: "Blackwell"; CPU: ARM Neoverse V3AE
 +
Series/Generation
 +
Tegra 2
 +
Tegra 3 (Kal-El)
 +
Tegra 4 (Wayne)
 +
Tegra 4i (Grey)
 +
Tegra K1 (Logan)
 +
Tegra X1 (Erista)
 +
Tegra X1+ (Mariko)
 +
Tegra X2 (Parker)
 +
Tegra Xavier
 +
Tegra Orin
 +
Tegra Thor
 +
CPU • Model/Device
 +
AP20H/AP25/T20/T25 • T30/T30L/AP33/T33 • T114 • T148? • T124 • T132 • T210 • T214 • T186 • T194 • T234 • T264?
 +
 +
{| class="wikitable" cellpadding="3px" style="border: 1px solid black; border-spacing: 0px; width: 100%; text-align:center;"
 +
|+ Tegra Models comparison
 +
! colspan="2" |Series /<br>Generation
 +
!Tegra 2
 +
!Tegra 3<br>(''Kal-El'')
 +
!Tegra 4<br>(''Wayne'')
 +
!Tegra 4i<br>(''Grey'')
 +
! colspan="2" |Tegra K1<br>(''Logan/Stark'')+
 +
!Tegra X1/X1+<br>(''Erista/Mariko'')
 +
!Tegra X2<br>(''Parker'')
 +
!Tegra <br>''Xavier''
 +
!Tegra <br>''Orin''
 +
|-
 +
! colspan="2" |Model /<br>Device
 +
|AP25/T25
 +
|T30/T30L<br>AP33/T33
 +
|T114
 +
|T148?
 +
|T124
 +
|T132
 +
|T210/T214
 +
|T186
 +
|T194<br>(AGX/NX)
 +
|T234
 +
|
 +
|-
 +
</pre>
 +
-->
 +
 +
== Microarchitectures ==
 +
 +
'''GPUs:'''
 +
{| border="0" cellpadding="2" width="100%"
 +
|-
 +
|width="20%" valign="top" align="left"|
 +
* NV3 (1997)
 +
* NV4 (''{{nvidia|Fahrenheit|l=arch}}'') (1998)
 +
* NV10 (''{{nvidia|Celsius|l=arch}}'') (1999)
 +
* NV20 (''{{nvidia|Kelvin|l=arch}}'') (2001)
 +
* NV30 (''{{nvidia|Rankine|l=arch}}'') (2003)
 +
* NV40 (''{{nvidia|Curie|l=arch}}'') (2004)
 +
|width="23%" valign="top" align="left"|
 +
* {{nvidia|Tesla|l=arch}}
 +
** G80/G92 (1 Gen) (2006)
 +
** G200/GT215 (2 Gen) (2008)
 +
* {{nvidia|Fermi|l=arch}}
 +
** GF100/GF104 (1 Gen) (2010)
 +
** GF110/GF114 (2 Gen) (2010)
 +
|width="23%" valign="top" align="left|
 +
* {{nvidia|Kepler|l=arch}}
 +
** GK104/GK110 (1 Gen) (2012)
 +
** GK208 (2 Gen) (2013)
 +
* {{nvidia|Maxwell|l=arch}}
 +
** GM107 (1 Gen) (2014)
 +
** GM204 (2 Gen) (2014)
 +
|width="17%" valign="top" align="left"|
 +
* {{nvidia|Pascal|l=arch}} (2016)
 +
* {{nvidia|Volta|l=arch}} (2017)
 +
* {{nvidia|Turing|l=arch}} (2018)
 +
* {{nvidia|Ampere|l=arch}} (2020)
 +
* {{nvidia|Hopper|l=arch}} (2022)
 +
* ''{{nvidia|Ada Lovelace|l=arch}}'' (2022)
 +
|width="17%" valign="top" align="left"|
 +
* {{nvidia|Blackwell|l=arch}} (2024)
 +
* ''{{nvidia|Rubin|l=arch}}'' (2026)
 +
* ''{{nvidia|Rubin Ultra|l=arch}}'' (2027)
 +
* ''{{nvidia|Feynman|l=arch}}'' (2028)
 +
|}
 +
<!--
 +
NVIDIA
 +
NV3 (1997)
 +
NV4 (Fahrenheit) (1998)
 +
NV10 (Celsius) (1999)
 +
NV20 (Kelvin) (2001)
 +
NV30 (Rankin) (2003)
 +
NV40 (Curie) (2004)
 +
* ''{{nvidia|Celsius|l=arch}}''
 +
* ''{{nvidia|Curie|l=arch}}''
 +
* ''{{nvidia|Kelvin|l=arch}}''
 +
* ''{{nvidia|Rankin|l=arch}}''
 +
* ''{{nvidia|Rubin|l=arch}}''
 +
G80/G92 (1st gen Tesla) (2006)
 +
G200/GT215 (2nd Generation Tesla) (2008)
 +
1st generation Fermi (GF100/GF104) (2010)
 +
2nd Generation Fermi (GF110/GF114) (2010)
 +
1st generation Kepler (GK104/GK110) (2012)
 +
2nd Generation Kepler (GK208) (2013)
 +
1st generation Maxwell (GM107) (2014)
 +
2nd Generation Maxwell (GM204) (2014)
 +
Pascal (2016)
 +
Volta (2017)
 +
Turing (2018)
 +
Ampere (2020)
 +
Ada Lovelace (2022)
 +
Blackwell (2025)
 +
-->
 +
 +
'''CPU:'''
 
{{collist
 
{{collist
 
| count = 4
 
| count = 4
 
|
 
|
−
* {{nvidia|NV1}}
+
:'''ARM:'''
−
* {{nvidia|NV2}}
+
* {{nvidia|Denver|l=arch}}
−
* {{nvidia|NV3}}
+
* {{nvidia|Denver 2|l=arch}}
−
* {{nvidia|NV4}}
+
* {{nvidia|Carmel|l=arch}}
−
* {{nvidia|NV5}}
+
* {{nvidia|Tegra|l=arch}}
−
* {{nvidia|GeForce 256}}
+
 
−
* {{nvidia|GeForce 2}}
+
:'''RISC-V:'''
−
* {{nvidia|GeForce 2 Go}}
+
* {{nvidia|NV-RISCV|l=arch}}
−
* {{nvidia|GeForce 3}}
+
'''[[NPU]]:'''
−
* {{nvidia|GeForce 4}}
+
* {{nvidia|NVDLA|l=arch}}
−
* {{nvidia|GeForce 4 Go}}
 
−
* {{nvidia|GeForce FX}}
 
−
* {{nvidia|GeForce FX Go 5}}
 
−
* {{nvidia|GeForce 6}}
 
−
* {{nvidia|GeForce Go 6}}
 
−
* {{nvidia|GeForce 7}}
 
−
* {{nvidia|GeForce Go 7}}
 
−
* {{nvidia|GeForce 8}}
 
−
* {{nvidia|GeForce 8M}}
 
−
* {{nvidia|GeForce 9}}
 
−
* {{nvidia|GeForce 9M}}
 
−
* {{nvidia|GeForce 100}}
 
−
* {{nvidia|GeForce 100M}}
 
−
* {{nvidia|GeForce 200}}
 
−
* {{nvidia|GeForce 200M}}
 
−
* {{nvidia|GeForce 300}}
 
−
* {{nvidia|GeForce 300M}}
 
−
* {{nvidia|GeForce 400}}
 
−
* {{nvidia|GeForce 400M}}
 
−
* {{nvidia|GeForce 500}}
 
−
* {{nvidia|GeForce 500M}}
 
−
* {{nvidia|GeForce 600}}
 
−
* {{nvidia|GeForce 600M}}
 
−
* {{nvidia|GeForce 700}}
 
−
* {{nvidia|GeForce 700M}}
 
−
* {{nvidia|GeForce 800M}}
 
−
* {{nvidia|GeForce 900}}
 
−
* {{nvidia|GeForce 900M}}
 
−
* {{nvidia|GeForce 10}}
 
−
* {{nvidia|Quadro}}
 
−
* {{nvidia|Quadro FX}}
 
−
* {{nvidia|Quadro NVS}}
 
−
* {{nvidia|Tesla}}
 
−
* {{nvidia|Quadro Go}}
 
−
* {{nvidia|Quadro FX Go}}
 
−
* {{nvidia|NVS}}
 
−
* {{nvidia|GRID}}
 
−
* {{nvidia|XGPU}}
 
−
* {{nvidia|RSX}}
 
 
}}
 
}}
 +
 +
== Comparison ==
 +
:;Tegra
 +
 +
{|class="wikitable" style="font-size: 95%; text-align: center;"
 +
|+
 +
! colspan="2" |Generation
 +
!Tegra 2
 +
!Tegra 3 <br>(Kal-El)
 +
!Tegra 4 <br>(Wayne)
 +
!Tegra 4i <br>(Grey)
 +
! colspan="2" |Tegra K1 <br>(Logan)
 +
!Tegra X1 <br>(Erista)
 +
!Tegra X1+ <br>(Mariko)
 +
!Tegra X2 <br>(Parker)
 +
!{{nvidia|Tegra Xavier|Tegra <br>Xavier}}
 +
!Tegra <br>Orin
 +
!Tegra <br>Thor
 +
|-
 +
! rowspan="6" |CPU
 +
!Models
 +
| AP25/T25 || T30/<!--T30L/AP33/-->T33 || T114 || T148 || T124 || T132 || T210 || T214 || T186 || T194 || T234 || T264?
 +
|-
 +
!Cores
 +
|2x ARM <br>Cortex-A9
 +
|4+1x ARM <br>Cortex-A9
 +
|4+1x ARM <br>Cortex-A15
 +
|4+1x ARM <br>Cortex-A9
 +
|4+1x ARM <br>Cortex-A15
 +
|2x <br>'''{{nvidia|Denver|l=arch}}'''
 +
| colspan="2" |4x ARM Cortex-A53 <br>(disabled) +<br>4x ARM Cortex-A57
 +
|2x '''{{nvidia|Denver|l=arch}}''' 2 <br>+ 4x ARM <br>Cortex-A57
 +
|8x <br>'''{{nvidia|Carmel|l=arch}}'''
 +
|12x <br>Cortex<br>-A78AE
 +
|14x <br>'''[[Neoverse]]''' <br>V3AE
 +
|-
 +
!Instruction <br>set
 +
| colspan="5" |[[ARMv7|ARMv7&#8209;A]] <br>(32&#8209;bit)
 +
| colspan="4" |[[ARMv8|ARMv8&#8209;A]] <br>(64&#8209;bit)
 +
| colspan="2" |[[ARMv8|ARMv8.2&#8209;A]] <br>(64&#8209;bit)
 +
| [[ARMv9|ARMv9.2&#8209;A]] <br>(64&#8209;bit)
 +
|-
 +
!L1 cache <br>(I/D)
 +
| colspan="5" |32/32KB
 +
|128/64KB
 +
| colspan="2" |32/32KB <br>+ 64/32KB
 +
|128/64KB <br>+ 48/32KB
 +
|128/64KB
 +
| colspan="2" |64/64KB
 +
|-
 +
!L2 cache
 +
| colspan="2" |1 MB
 +
| colspan="4" |2 MB
 +
| colspan="2" |128KB + 2MB
 +
|2MB + 2MB
 +
|8 MB
 +
|3 MB
 +
|14 MB
 +
|-
 +
!L3 cache
 +
| colspan="9" |N/A
 +
|4 MB
 +
|6 MB
 +
|16 MB
 +
|-
 +
! rowspan="4" |GPU
 +
! Architecture
 +
| colspan="4" |{{nvidia|Vec4|l=arch}}
 +
| colspan="2" |{{nvidia|Kepler|l=arch}}
 +
| colspan="2" |{{nvidia|Maxwell|l=arch}}
 +
|{{nvidia|Pascal|l=arch}}
 +
|{{nvidia|Volta|l=arch}}
 +
|{{nvidia|Ampere|l=arch}}
 +
|{{nvidia|Blackwell|l=arch}}
 +
|-
 +
!CUDA cores
 +
|4+4*
 +
|8+4*
 +
|48+24*
 +
|48+12*
 +
| colspan="2" |192
 +
| colspan="3" |256
 +
|512
 +
|2048
 +
|2540
 +
|-
 +
!Tensor cores
 +
| colspan="9" |N/A
 +
| colspan="2" |64
 +
|96
 +
|-
 +
!RT cores
 +
| colspan="10" |N/A
 +
|8
 +
|?
 +
|-
 +
! rowspan="3" |RAM
 +
!Protocol
 +
|DDR2/ <br>LPDDR2
 +
|DDR3/ <br>LPDDR2
 +
| colspan="4" |DDR3/ <br>LPDDR3
 +
|LPDDR3/ <br>LPDDR4
 +
| colspan="3" |LPDDR4/ <br>LPDDR4X
 +
| colspan="2" |LPDDR5/5X
 +
|-
 +
!Max. size
 +
|1&nbsp;GB
 +
|2&nbsp;GB
 +
| colspan="2" |4&nbsp;GB
 +
| colspan="5" |8&nbsp;GB
 +
| colspan="2" |64&nbsp;GB
 +
|128&nbsp;GB
 +
|-
 +
!Bandwidth
 +
|2.7 GB/s
 +
| colspan="2" |6.4 GB/s
 +
|7.5 GB/s
 +
| colspan="2" |14.9 GB/s
 +
|25.6 <br>GB/s
 +
|34.1 <br>GB/s
 +
|59.7 <br>GB/s
 +
|136.5 <br>GB/s
 +
|204.8 <br>GB/s
 +
|273.0 <br>GB/s
 +
|-
 +
! colspan="2" |Process
 +
| colspan="2" |[[40 nm]]
 +
| colspan="4" |[[28 nm]]
 +
|[[20 nm]]
 +
| colspan="2" |[[16 nm]]
 +
|[[12 nm]]
 +
|[[8 nm]]
 +
|[[5 nm|4 nm]]
 +
|-
 +
|}
 +
* [[VLIW]]-based Vec4: Pixel shaders + Vertex shaders. Since Kepler, Unified shaders are used.
 +
 +
== Technology ==
 +
* {{\\|NVLink}}
 +
* {{\\|NVSwitch}}

Latest revision as of 15:41, 4 May 2025

NVIDIA
nvidia logo.svg
Type Public
Founded 1993
Founder Jen-Hsun Huang
Chris Malachowsky
Curtis Priem
Headquarters Santa Clara, California
Website http://www.nvidia.com

NVIDIA Corporation is an American fabless semiconductor company that focuses largely on graphics processing units for the gaming industry and machine learning applications. They also design various other chips, SoC, and consumer electronics for the mobile and automotive markets.

GPU Families[edit]

CPU Families[edit]

GoForce
  • GoForce
Tegra

Microarchitectures[edit]

GPUs:

  • Tesla
    • G80/G92 (1 Gen) (2006)
    • G200/GT215 (2 Gen) (2008)
  • Fermi
    • GF100/GF104 (1 Gen) (2010)
    • GF110/GF114 (2 Gen) (2010)
  • Kepler
    • GK104/GK110 (1 Gen) (2012)
    • GK208 (2 Gen) (2013)
  • Maxwell
    • GM107 (1 Gen) (2014)
    • GM204 (2 Gen) (2014)

CPU:

ARM:
RISC-V:

NPU:

Comparison[edit]

Tegra
Generation Tegra 2 Tegra 3
(Kal-El) 		Tegra 4
(Wayne) 		Tegra 4i
(Grey) 		Tegra K1
(Logan) 		Tegra X1
(Erista) 		Tegra X1+
(Mariko) 		Tegra X2
(Parker) 		Tegra
Xavier 		Tegra
Orin 		Tegra
Thor
CPU Models AP25/T25 T30/T33 T114 T148 T124 T132 T210 T214 T186 T194 T234 T264?
Cores 2x ARM
Cortex-A9 		4+1x ARM
Cortex-A9 		4+1x ARM
Cortex-A15 		4+1x ARM
Cortex-A9 		4+1x ARM
Cortex-A15 		2x
Denver 		4x ARM Cortex-A53
(disabled) +
4x ARM Cortex-A57 		2x Denver 2
+ 4x ARM
Cortex-A57 		8x
Carmel 		12x
Cortex
-A78AE 		14x
Neoverse
V3AE
Instruction
set 		ARMv7‑A
(32‑bit) 		ARMv8‑A
(64‑bit) 		ARMv8.2‑A
(64‑bit) 		ARMv9.2‑A
(64‑bit)
L1 cache
(I/D) 		32/32KB 128/64KB 32/32KB
+ 64/32KB 		128/64KB
+ 48/32KB 		128/64KB 64/64KB
L2 cache 1 MB 2 MB 128KB + 2MB 2MB + 2MB 8 MB 3 MB 14 MB
L3 cache N/A 4 MB 6 MB 16 MB
GPU Architecture Vec4 Kepler Maxwell Pascal Volta Ampere Blackwell
CUDA cores 4+4* 8+4* 48+24* 48+12* 192 256 512 2048 2540
Tensor cores N/A 64 96
RT cores N/A 8 ?
RAM Protocol DDR2/
LPDDR2 		DDR3/
LPDDR2 		DDR3/
LPDDR3 		LPDDR3/
LPDDR4 		LPDDR4/
LPDDR4X 		LPDDR5/5X
Max. size 1 GB 2 GB 4 GB 8 GB 64 GB 128 GB
Bandwidth 2.7 GB/s 6.4 GB/s 7.5 GB/s 14.9 GB/s 25.6
GB/s 		34.1
GB/s 		59.7
GB/s 		136.5
GB/s 		204.8
GB/s 		273.0
GB/s
Process 40 nm 28 nm 20 nm 16 nm 12 nm 8 nm 4 nm
  • VLIW-based Vec4: Pixel shaders + Vertex shaders. Since Kepler, Unified shaders are used.

Technology[edit]

Retrieved from "https://en.wikichip.org/w/index.php?title=nvidia&oldid=103392"
Facts about "Nvidia"
RDF feed
company typepublic +
founded1993 +
founderJen-Hsun Huang +, Chris Malachowsky + and Curtis Priem +
full page namenvidia +
headquartersSanta Clara, California +
instance ofsemiconductor company +
nameNVIDIA +
websitehttp://www.nvidia.com +
wikidata idQ182477 +