Basics of a Graphics Card

[AoD]

Hey guys, so I practically understand the basics of all the components in a computer, but I still don't fully understand the specifications of a graphics card that you look at when choosing to purchase one? Could anyone please explain this to me?

[Crunchy Dragon]

Specifications to look for when picking a GPU is a little complicated, I admit it's something that I personally never really paid a lot of attention to.

 

I usually want over 3-4Gb VRAM and good performance, but beyond that, I don't really care too much as long as it's not a power-hungry monster like the R9 295X2 or 390/390X.

[xriqn]

6 minutes ago, AoD said:

Hey guys, so I practically understand the basics of all the components in a computer, but I still don't fully understand the specifications of a graphics card that you look at when choosing to purchase one? Could anyone please explain this to me?

Look for the Video RAM (measured in GB) and the clock speed (measured in MHz but more commonly GHz now). Higher on both is better.

EDIT: Check the RAM type too, e.g GDDR3, GDDR5 or even GDDR5X

[minervx]

look at benchmarks and real-world performance.  otherwise, youll end up chasing numbers that may not matter to your use case.

 

 

[AoD]

4 minutes ago, Crunchy Dragon said:

Specifications to look for when picking a GPU is a little complicated, I admit it's something that I personally never really paid a lot of attention to.

 

I usually want over 3-4Gb VRAM and good performance, but beyond that, I don't really care too much as long as it's not a power-hungry monster like the R9 295X2 or 390/390X.

That's basically how I am looking at things right now  

[AoD]

4 minutes ago, xriqn said:

Look for the Video RAM (measured in GB) and the clock speed (measured in MHz but more commonly GHz now). Higher on both is better.

EDIT: Check the RAM type too, e.g GDDR3, GDDR5 or even GDDR5X

What's a good clock speed for running 1080p? Or just for running triple A games?

 

[AoD]

4 minutes ago, minervx said:

look at benchmarks and real-world performance.  otherwise, youll end up chasing numbers that may not matter to your use case.

 

 

Thanks for the response 

[minervx]

1 minute ago, AoD said:

What's a good clock speed for running 1080p? Or just for running triple A games?

 

gtx 1050 ti , gtx 1060 gb or some card benched between them is good.

 

its hard to describe a clock speed for a processor that has lots and lots of cores.

[Mira Yurizaki]

Just now, AoD said:

Hey guys, so I practically understand the basics of all the components in a computer, but I still don't fully understand the specifications of a graphics card that you look at when choosing to purchase one? Could anyone please explain this to me?

The basics boil down to the following, with this caveat: You cannot directly compare these specs between manufacturers. So you cannot compare an NVIDIA card against an AMD card and judge how well they do based on their raw specs. You must find benchmarks to see how they compare.

 

In any case, here are the common ones you find in stores and on the box

• GPU Core
  • Shader unit count: These are basic "worker" units of a GPU. So the more of these there are, the better. AMD calls these stream processors (sometimes they're clumped into "Compute Units", which are 64 of these). Intel calls these Execution Units (EUs). NVIDIA calls these CUDA cores.
  • Base clock speed: This is the normal operating speed of the GPU.
  • Boost clock speed: For NVIDIA and AMD GPUs, they will try to automatically clock themselves faster if their cooling system can keep up with it (faster speed = more heat). Note that the speed on the box doesn't appear to mean much as I've seen my card routinely go past boost clock speeds.
• Memory
  • Note that memory is something you shouldn't really pay much attention to, as most cards are designed with the memory specs they need. The only exception is if you find two cards with different memory configurations. In that case, you should opt for the better one if you can afford it.
  • Memory Capacity:  GPUs need their own RAM to do work. For the most part in games, this largely affects the following
    • The resolution for anything, such as screen resolution and shadow resolution
    • Texture quality
    • How much MSAA you can do. Other anti-aliasing methods like FXAA, MLAA, and TXAA consume little, if any VRAM.
  • Memory Interface: How many bits at a time the GPU memory can transfer at once
  • Memory speed: How fast the memory can operate, which translates to how much data per second the memory can transfer.
• API support: This is the maximum API support they have or what they were designed in mind. This is only important if you want to play games that use a particular API like DirectX 11 or 12 or OpenGL.
• Power connectors: Most cards these days need additional power than they can get from the PCI Express slot. They either require a 6-pin, 8-pin, or both. You need a power supply with matching connectors to use the card safely.

I'm probably forgetting a few things, but those are the big ones. I could throw in more if you want.

[Paddi01]

8 minutes ago, Crunchy Dragon said:

I usually want over 3-4Gb VRAM

For 1080p, i wouldn't go below 4.

[Crunchy Dragon]

3 minutes ago, AoD said:

That's basically how I am looking at things right now  

Another way I usually see recommended is to get the highest performing card that fits your budget, which I personally did on my build(GTX 780 for nothing was one heck of a deal).

1 minute ago, AoD said:

What's a good clock speed for running 1080p? Or just for running triple A games?

I run my GTX 780 at 1105Mhz(1.1Ghz) at 1080p, for well-optimized games it holds up pretty well. PUBG is really hard to run, I'm stuck at 720p Lowest settings to get a playable framerate. Usually if you get a card that runs at least 1Ghz, you're good to go.

[Crunchy Dragon]

1 minute ago, Paddi01 said:

For 1080p, i wouldn't go below 4.

For the games I play, I've done fine on 3 and would consider it the minimum for games in this day and age, although that may certainly be invalid by 2019.

[AoD]

3 minutes ago, minervx said:

gtx 1050 ti , gtx 1060 gb or some card benched between them is good.

 

its hard to describe a clock speed for a processor that has lots and lots of cores.

Ok cool no worries, thanks

 

[AoD]

2 minutes ago, M.Yurizaki said:

The basics boil down to the following, with this caveat: You cannot directly compare these specs between manufacturers. So you cannot compare an NVIDIA card against an AMD card and judge how well they do based on their raw specs. You must find benchmarks to see how they compare.

 

In any case, here are the common ones you find in stores and on the box

• GPU Core
  • Shader unit count: These are basic "worker" units of a GPU. So the more of these there are, the better. AMD calls these stream processors (sometimes they're clumped into "Compute Units", which are 64 of these). Intel calls these Execution Units (EUs). NVIDIA calls these CUDA cores.
  • Base clock speed: This is the normal operating speed of the GPU.
  • Boost clock speed: For NVIDIA and AMD GPUs, they will try to automatically clock themselves faster if their cooling system can keep up with it (faster speed = more heat). Note that the speed on the box doesn't appear to mean much as I've seen my card routinely go past boost clock speeds.
• Memory
  • Note that memory is something you shouldn't really pay much attention to, as most cards are designed with the memory specs they need. The only exception is if you find two cards with different memory configurations. In that case, you should opt for the better one if you can afford it.
  • Memory Capacity:  GPUs need their own RAM to do work. For the most part in games, this largely affects the following
    • The resolution for anything, such as screen resolution and shadow resolution
    • Texture quality
    • How much MSAA you can do. Other anti-aliasing methods like FXAA, MLAA, and TXAA consume little, if any VRAM.
  • Memory Interface: How many bits at a time the GPU memory can transfer at once
  • Memory speed: How fast the memory can operate, which translates to how much data per second the memory can transfer.
• API support: This is the maximum API support they have or what they were designed in mind. This is only important if you want to play games that use a particular API like DirectX 11 or 12 or OpenGL.
• Power connectors: Most cards these days need additional power than they can get from the PCI Express slot. They either require a 6-pin, 8-pin, or both. You need a power supply with matching connectors to use the card safely.

I'm probably forgetting a few things, but those are the big ones. I could throw in more if you want.

Oh flip that's quite alot, but I understand some more now, thanks

[Paddi01]

6 minutes ago, M.Yurizaki said:

Memory Interface: How many bits at a time the GPU memory can transfer at once

GPU's earlier had like 512 bits but were much slower. Is there a simple example you can do about this?

I sometime read it's like cars on a highway just with more lanes.

[Mira Yurizaki]

10 minutes ago, Paddi01 said:

GPU's earlier had like 512 bits but were much slower. Is there a simple example you can do about this?

I sometime read it's like cars on a highway just with more lanes.

Memory bandwidth is bits times speed. The effective memory speed of cards of that time topped out around 2.5 GHz. Today, top-end cards top out at around 10 GHz. Note that doesn't mean they actually operate at that speed, but with pumping, you can use a slower speed to create an effectively faster one.

 

And the reason why cards started using less bits per transfer is because apparently wider memory interfaces are power hogs. Plus for every bit you have, you need a trace coming off the memory chip to the GPU.

 

EDIT: So yes, bits are like lanes in a highway, while clock speed is the speed limit.