General Questions About Graphics cards

[Fez760]

So i don't know much about selected bits Graphics Cards Being one What are the important things to look for As in the following -ill be Using the EVGA 04G-P4-2989-KR GeForce GTX 980 as an example

 

Core and boost clock, what significance do these hold?

 

CUDA cores <- What are these?? Do they matter?

 

The Effective Memory Clock? or is this just self explanatory

 

Memory Interface? Is that like DDR2,DDR3,DDR4 GDDR5?

 

Difference Between Radeon And GeForce? or are they just another competition like AMD and Intel

 

These are probably stupid question but i thank you in advance for your help!

 

 

[Glenwing]

What you look for is performance benchmarks. The spec sheets don't help you compare graphics cards. Go read reviews from Anandtech.

Radeon is AMD's graphics card branding, GeForce is NVIDIA's branding.

[Guest Kloaked]

What you look for is performance benchmarks. The spec sheets don't help you compare graphics cards. Go read reviews from Anandtech.

Radeon is AMD's graphics card branding, GeForce is NVIDIA's branding.

Yup. Unless you're wanting to just learn more about how the cards work, you just need to look at benchmarks to see performance, which is the more important thing when deciding on a card.

[minibois]

-snip-

Specs are not that important in GPU's, as you want to see real life performance.

Look up benchmarks.

That's about it for core and boost clock.

 

Cuda cores are sometimes important for rendering stuff, but not that important for regular consumers. My earlier statement applies here too.

Memory clock same.

 

Pretty much all modern GPU's use GDDR5. Not much to say about that. Older cards use DDR3 and AMD's 300 series might use another type.

 

Radeon = AMD and Geforce = Nvidia. 

They are competitors like Intel and AMD. They both have very good cards and the GTX 980 is very good too.

 

 

This might help:

[Glenwing]

Specs are not that important in GPU's, as you want to see real life performance.

Look up benchmarks.

That's about it for core and boost clock.

 

Cuda cores are sometimes important for rendering stuff, but not that important for regular consumers. My earlier statement applies here too.

Memory clock same.

 

Pretty much all modern GPU's use GDDR5. Not much to say about that. Older cards use DDR3 and AMD's 300 series might use another type.

 

Radeon = AMD and Geforce = Nvidia. 

They are competitors like Intel and AMD. They both have very good cards and the GTX 980 is very good too.

 

 

This might help:

CUDA cores are what NVIDIA calls regular cores, so they are kinda central to the GPU's whole operation actually

[Fez760]

Specs are not that important in GPU's, as you want to see real life performance.

Look up benchmarks.

That's about it for core and boost clock.

 

Cuda cores are sometimes important for rendering stuff, but not that important for regular consumers. My earlier statement applies here too.

Memory clock same.

 

Pretty much all modern GPU's use GDDR5. Not much to say about that. Older cards use DDR3 and AMD's 300 series might use another type.

 

Radeon = AMD and Geforce = Nvidia. 

They are competitors like Intel and AMD. They both have very good cards and the GTX 980 is very good too.

 

 

This might help:

 

 

Yup. Unless you're wanting to just learn more about how the cards work, you just need to look at benchmarks to see performance, which is the more important thing when deciding on a card.

 

 

What you look for is performance benchmarks. The spec sheets don't help you compare graphics cards. Go read reviews from Anandtech.

Radeon is AMD's graphics card branding, GeForce is NVIDIA's branding.

Yes Yes Im more Generaly curious for knowledge purposes

[minibois]

CUDA cores are what NVIDIA calls regular cores, so they are kinda central to the GPU's whole operation actually

Alright, I knew CUDA was an Nvidia only thing, but I wasn't 100% on what they do.

As I said, I knew quite a bit of people used some Nvidia cards for cuda rendering (the 750 Ti mainly, because low power consumption), but wasn't sure how to explain.

[Glenwing]

Yes Yes Im more Generaly curious for knowledge purposes

Ok.

Digital processors like a GPU are regulated by a clock to keep all their pieces synchronized. Every time the clock ticks, everything associated with that clock does whatever it does once (known as a "cycle"). GPUs have two main clocks that we consumers are concerned with, the core clock and the memory clock. Keep in mind the clock frequency is not a measure of speed or power. Computational power is a factor of: how much stuff the GPU gets done every time the clock ticks, and how many times the clock ticks per second. Clock frequency only tells you the second. The first can't be quantified on a spec sheet, this is why you need to look at benchmarks instead.

A GPU has many processing cores, it's similar to a CPU except it's designed with hundreds of really small cores instead of a few strong ones. GPUs can perform several operations but by far the most common task is 32-bit floating point calculations (FP32) so the bulk of the GPU's cores are FP32 cores. Every time the core clock ticks, they do some amount of FP32 calculation (how much depends on the specific core design). NVIDIA calls these CUDA cores, AMD calls them stream processors. Twice as many cores is theoretically twice as many calculations, if the core design and clock frequency are the same.

The core clock may operate at a frequency of 800-1200MHz typically, though you can usually change it from the default setting. Each GPU does have a limit to how fast it can operate though, and won't be able to keep up with the clock if you raise it past a certain point, which will result in device failure. (Software or even hardware depending on if you change the voltage). Modern GPUs vary their clock speed to save power. When the GPU is not under heavy load, the clock slows down to a very low level, and when it's under a heavy load the clock speed rises. The boost clock is the maximum clock frequency, the minimum is not listed. Base clock is not really meaningful.

The GPU processes a lot of data and has dedicated RAM for storing that data. Graphics cards usually use specialized RAM which is a lot faster than the RAM used on CPUs (although higher latency so not good for use with CPUs). Most graphics cards use GDDR5, which is based on DDR3 but optimized for graphics cards. Previous graphics cards used GDDR4 (based on DDR3), GDDR3 (DDR2), GDDR2 (DDR), or GDDR (DDR). This is called the memory type.

The memory interface is the connection between the GPU and the RAM. GDDR5 is connected via a 32-bit bus, which means every time the memory clock ticks, the GPU can send 32 ones or zeroes across the memory bus simultaneously. The memory clock may tick at a frequency of 6-7GHz or more (6GHz = 6 billion ticks per second). Most graphics cards combine several memory busses together for even more bandwidth, so they'll have maybe 8 or even 12 GDDR5 memory busses being used simultaneously, for an effective 256-bit or 384-bit bus.

As GPUs are powerful and work with a lot of data, memory bandwidth is helpful to a point. Memory capacity can also be helpful to a point, that really depends on what it's being used for.

[typographie]

Alright, I knew CUDA was an Nvidia only thing, but I wasn't 100% on what they do.

As I said, I knew quite a bit of people used some Nvidia cards for cuda rendering (the 750 Ti mainly, because low power consumption), but wasn't sure how to explain.

 

Nvidia's naming is pretty confusing. All processors (whether CPUs or GPUs) have "cores" of some sort. Nvidia calls theirs "CUDA cores," AMD usually calls their own stream processors. Nvidia's CUDA rendering API is another topic entirely.

[dougdangger]

In general:

 

AMD

slightly cheaper than Nvidia performance per dollar

better looking design

 

Nvidia

quality components

frequently updated rock solid drivers