Jump to content

What Are Cuda Cores

misterman

a cuda core is just a core on say a cpu, Nvidia calls there cores cuda cores and amd called them stream processors

so a 1080 will have 2560 cuda cores, meaning it has 2560 cores. thats a really basic example

but don't get cpu and GPU cores mixed up

My speakers dont even fit on or under my desk...PA's FTW

Link to comment
Share on other sites

Link to post
Share on other sites

Each cuda core is a compute unit. More means better performance, but direct comparison is only possible within the same generation. Also clock speed play an important role in judging each cores' performance.

CPU: i7-2600K 4751MHz 1.44V (software) --> 1.47V at the back of the socket Motherboard: Asrock Z77 Extreme4 (BCLK: 103.3MHz) CPU Cooler: Noctua NH-D15 RAM: Adata XPG 2x8GB DDR3 (XMP: 2133MHz 10-11-11-30 CR2, custom: 2203MHz 10-11-10-26 CR1 tRFC:230 tREFI:14000) GPU: Asus GTX 1070 Dual (Super Jetstream vbios, +70(2025-2088MHz)/+400(8.8Gbps)) SSD: Samsung 840 Pro 256GB (main boot drive), Transcend SSD370 128GB PSU: Seasonic X-660 80+ Gold Case: Antec P110 Silent, 5 intakes 1 exhaust Monitor: AOC G2460PF 1080p 144Hz (150Hz max w/ DP, 121Hz max w/ HDMI) TN panel Keyboard: Logitech G610 Orion (Cherry MX Blue) with SteelSeries Apex M260 keycaps Mouse: BenQ Zowie FK1

 

Model: HP Omen 17 17-an110ca CPU: i7-8750H (0.125V core & cache, 50mV SA undervolt) GPU: GTX 1060 6GB Mobile (+80/+450, 1650MHz~1750MHz 0.78V~0.85V) RAM: 8+8GB DDR4-2400 18-17-17-39 2T Storage: HP EX920 1TB PCIe x4 M.2 SSD + Crucial MX500 1TB 2.5" SATA SSD, 128GB Toshiba PCIe x2 M.2 SSD (KBG30ZMV128G) gone cooking externally, 1TB Seagate 7200RPM 2.5" HDD (ST1000LM049-2GH172) left outside Monitor: 1080p 126Hz IPS G-sync

 

Desktop benching:

Cinebench R15 Single thread:168 Multi-thread: 833 

SuperPi (v1.5 from Techpowerup, PI value output) 16K: 0.100s 1M: 8.255s 32M: 7m 45.93s

Link to comment
Share on other sites

Link to post
Share on other sites

57 minutes ago, Jurrunio said:

Each cuda core is a compute unit. More means better performance, but direct comparison is only possible within the same generation. Also clock speed play an important role in judging each cores' performance.

nope, a compute unit is a core. On a GPU a core contains shaders which are called cuda cores.

 

59 minutes ago, scobie56 said:

a cuda core is just a core on say a cpu, Nvidia calls there cores cuda cores and amd called them stream processors

so a 1080 will have 2560 cuda cores, meaning it has 2560 cores. thats a really basic example

but don't get cpu and GPU cores mixed up

nope, it has 2560 shaders not cores.

 

Cuda cores are shaders, stream processors are shaders. The true question is whats a shader.

Ever since GPUs came out for dx10, they use a unified architecture. What this means is that the GPU cores contain more general purpose units. These units are usually called shaders. Each shader is basically like some kind of tiny processor part of a larger one (like the SSE units in an x86 CPU) that can perform some tasks, not every.

 

So each GPU core contains a decoder to decode instructions for the core itself. The workload is then sent to the various different units in the core depending on the work itself. Each GPU core consists of shaders (nvidia calls them cuda cores, nvidia calls theirs stream processors), TMUs, ROPs and in some GPUs they have other units as well. Each unit has a contribution to graphics so a GPU with more of a particular unit is faster at a particular graphics setting.

TMUs do everything texture related, even AF as well. More vram allows higher quality textures along with more TMUs

ROPs deal with post processing and AA

Shaders deal with math, stuff like 3d mesh calculations, shadows and so on. Stuff that needs calculations including effects.

 

A GPU with so many shaders isnt necessarily fast, it can be bottlenecked by other things like vram bandwidth, too few TMUs, too few ROPs (vega).

 

Remember a GPU has cores which itself has a bunch of shaders. So a full chip vega has 64 cores and not 4096 cores. Those 64 cores each has a bunch of shaders (like 64 each per core).

 

So shader count cannot be compared between different GPUs, only same generation GPUs. An AMD GCN (GCN, polaris, vega, fiji, etc) is not the same as a pascal shader, and a pascal shader is not the same as a kepler shader or even a fermi shader.

 

on intel, each shader on an intel IGP is actually a core too as each intel shadaer is an x86 core with a simplified decoder. So each intel shader is also a core which is what gives intel it's strong IGP IPC performance vs nvidia and AMD GPUs but intel only includes too few of its IGP cores.

Link to comment
Share on other sites

Link to post
Share on other sites

Cuda core is pretty much compute unit in terms of gpu Put it simply.

 

the more the better. But also have to factor in more than just cores alone, clock speed mem speed all come into play too one way or another.

CPU: Intel Core i7-7700K | Motherboard: ASUS ROG STRIX Z270H | Graphics Card: ASUS ROG STRIX GTX 1080 Ti OCEdition | RAM: 16GB G.Skill Ripjaws V 3000MHz |Storage: 1 x Samsung 830 EVO Series 250GB | 1 x Samsung 960 PRO Series 512GB | 1 x Western Digital Blue 1TB | 1 x Western Digital Blue 4TB | PSU: Corsair RM750x 750W 80+ Gold Power Supply | Case: Cooler Master MasterCase 5 Pro |

Cooling: Corsair H100i v2 // 4x Corsair ML140 RED Fans // 2x Corsair ML120 RED Fans 
---

Monitor: ASUS ROG Swift PG279Q 1440p 165Hz IPS G-Sync | Keyboard: Corsair K70 LUX Red LED, Cherry MX Brown Switches | Mouse: Corsair Glaive RGB | Speakers: Logitech Z623 THX Certified Speakers

Link to comment
Share on other sites

Link to post
Share on other sites

27 minutes ago, xsimplyjosh said:

Cuda core is pretty much compute unit in terms of gpu Put it simply.

 

the more the better. But also have to factor in more than just cores alone, clock speed mem speed all come into play too one way or another.

its not a compute unit. a compute unit is a GPU core and not a shader or cuda core. If you dont believe me simply query using openCL. A shader or CUDA core as you call it is simply part of a GPU core (like a CPU core). It cannot function independently and relies on the GPU core for operation. For instance if you give an array of numbers to multiply by another array of numbers, the GPU core will receive it and divide the elements to give to each shader (cuda core) to perform the work. Hence a shader (cuda core) is simply a unit part of a processor.

 

a GPU consists of a bunch of processor cores which is called a compute unit in compute terms . Each core/compute unit has a bunch of shaders, TMUs and ROPs.

 

For instance the vega 64 has 64 cores and 4096 shaders in total. 

 

When you look at a GPU block diagram, you see the shaders arranged in blocks. Each block is a GPU core.

 

Seriously guys, why do you keep getting this wrong. GPU shaders arent cores. Nvidia calls their shaders cuda cores, AMD calls their shaders stream processors, while intel just simplifies their x86 CPUs into GPU cores which also function as shaders. Hence only intel can call their shaders cores.

Link to comment
Share on other sites

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×