Graphics Cards
This section will cover many different parts of a graphics card. What is important, brands to buy, pricing and everything in between!
What is there to compare? Well here is a list of important stats to compare
Stream Processors/Cuda Cores
The only time measuring these are if you are comparing 2 cards in the same reasons, such as a GTX 10xx, 9xx, 7xx or AMD R7-xxx, R9-xxx Rx-xxx. But comparing a GTX 10xx to a GTX 9xx, cuda cores are almost irrelevant. AMD seems to favor stream processors over core clock where NVidia favors core clock more than anything else.
Core Clock
Core clock is a useful measurement of the main processing power of a GPU. However, like with CPU's there is far more in play to determine raw power. Like core count on GPU's, these are only relevant when comparing 2 like cards in the same series. In the recent generations of NVidia has been pushing over 2000MHz core clock where AMD's flagships (RX vega 56/64) are only pushing 1300-1600MHz core clock. But a 1080 and Vega 64 are pretty even as is the Vega 56 and GTX 1070.
Vram (Video Memory)
Video memory or Vram is essentially memory on the graphics card. It stores textures and other things while the video card renders the previous frame. Higher resolution, higher resolution textures, antialiasing, and anisotropic filtering all requires more memory on the graphics card as these things take up more space. The more pixels per inch, the larger the picture file is a good way to look at it.
The minimum requirement for graphical memory is 2GB at 1080p, but we are at a point where 4GB does not hurt and above 1080p 6GB is ideal. Some games have such high resolution textures that they eat up video memory waiting to be rendered.
Memory Clock
This is part of the bandwidth calculation. Think of the memory clock as the speed of traffic on a highway. The faster the speed the faster you get there. The faster the memory clock, the faster the graphics card can render textures store in the video memory. These are measured in a few ways. In the case of GDDR5, you have the actual speed say 1250MHz and times it by 4 to have an effect speed of 5000MHz or 5GHz. In the case of GDDR3, you only multiply that by 2. This is why GDDR3 is recommended AGAINST for gaming. The higher the better.
Memory Interface
This is also part of the bandwidth calculation. Think of the memory interface as how wide the highway is. 4 lanes wide means you will not get into traffic jams, but 1 or 2 lanes wide you might, especially if the speed is slower. You generally will see 128, 256 384 and 512 bit interfaces on gaming graphics card. 192 has been popular as well. Historically speaking, the higher the better, but NVidia has come out with a way to compress data that has made it so much more efficient that faster speeds benefit much more than interface.
Bandwidth
This section can be extremely important as it brings everything together. Think of usable or required Vram as traffic on a highway. Memory interface is how many lanes there are and memory clock is how fast traffic (can) move. At 1920x1080, you will be using about 2GB of Vram in top tier games. Ideally you want a MINIMUM of 256bit interface (think of this as 2 lanes wide on a highway). 4 or 512 bit is a bit overkill and 384 bit or 3 lanes is perfect usually. Bandwidth is a combination of memory interface and memory clock which is basically how much data can be rendered per second.
Now as resolutions increase, AA increases or textures increase more Vram maybe required and 256bit won't cut it anymore. In these cases you want a wider memory interface and a faster memory clock. An R9-290 has a bandwidth of about 320, where as a GTX 780 TI is around 340. Even though the R9-290 has a wider interface 512 vs 384, the memory clock of the GTX 780 TI is 7GHz effective vs 5GHz effective. The theory is, if you have 3 lanes of traffic moving consistently at 70MPH, you will have more cars go from point A to point B than 4 lanes of traffic moving at 50MPH.
Memory bandwidth calculation is as such: Memory interface / 8 * memory clock * multiplier (2 for GDDR3 and 4 for GDDR5). So 512 / 8 * 1250 * 4 = 320gbps on the R9-290. The GTX 780 TI would be 384 / 8 * 1750 * 4 = 336gbps.
What is there to compare? Well here is a list of important stats to compare
- Stream Processors and Cuda Cores
- Core Clock
- Memory Clock
- Bandwidth
- Memory Interface
- Vram
- Architecture
Stream Processors/Cuda Cores
The only time measuring these are if you are comparing 2 cards in the same reasons, such as a GTX 10xx, 9xx, 7xx or AMD R7-xxx, R9-xxx Rx-xxx. But comparing a GTX 10xx to a GTX 9xx, cuda cores are almost irrelevant. AMD seems to favor stream processors over core clock where NVidia favors core clock more than anything else.
Core Clock
Core clock is a useful measurement of the main processing power of a GPU. However, like with CPU's there is far more in play to determine raw power. Like core count on GPU's, these are only relevant when comparing 2 like cards in the same series. In the recent generations of NVidia has been pushing over 2000MHz core clock where AMD's flagships (RX vega 56/64) are only pushing 1300-1600MHz core clock. But a 1080 and Vega 64 are pretty even as is the Vega 56 and GTX 1070.
Vram (Video Memory)
Video memory or Vram is essentially memory on the graphics card. It stores textures and other things while the video card renders the previous frame. Higher resolution, higher resolution textures, antialiasing, and anisotropic filtering all requires more memory on the graphics card as these things take up more space. The more pixels per inch, the larger the picture file is a good way to look at it.
The minimum requirement for graphical memory is 2GB at 1080p, but we are at a point where 4GB does not hurt and above 1080p 6GB is ideal. Some games have such high resolution textures that they eat up video memory waiting to be rendered.
Memory Clock
This is part of the bandwidth calculation. Think of the memory clock as the speed of traffic on a highway. The faster the speed the faster you get there. The faster the memory clock, the faster the graphics card can render textures store in the video memory. These are measured in a few ways. In the case of GDDR5, you have the actual speed say 1250MHz and times it by 4 to have an effect speed of 5000MHz or 5GHz. In the case of GDDR3, you only multiply that by 2. This is why GDDR3 is recommended AGAINST for gaming. The higher the better.
Memory Interface
This is also part of the bandwidth calculation. Think of the memory interface as how wide the highway is. 4 lanes wide means you will not get into traffic jams, but 1 or 2 lanes wide you might, especially if the speed is slower. You generally will see 128, 256 384 and 512 bit interfaces on gaming graphics card. 192 has been popular as well. Historically speaking, the higher the better, but NVidia has come out with a way to compress data that has made it so much more efficient that faster speeds benefit much more than interface.
Bandwidth
This section can be extremely important as it brings everything together. Think of usable or required Vram as traffic on a highway. Memory interface is how many lanes there are and memory clock is how fast traffic (can) move. At 1920x1080, you will be using about 2GB of Vram in top tier games. Ideally you want a MINIMUM of 256bit interface (think of this as 2 lanes wide on a highway). 4 or 512 bit is a bit overkill and 384 bit or 3 lanes is perfect usually. Bandwidth is a combination of memory interface and memory clock which is basically how much data can be rendered per second.
Now as resolutions increase, AA increases or textures increase more Vram maybe required and 256bit won't cut it anymore. In these cases you want a wider memory interface and a faster memory clock. An R9-290 has a bandwidth of about 320, where as a GTX 780 TI is around 340. Even though the R9-290 has a wider interface 512 vs 384, the memory clock of the GTX 780 TI is 7GHz effective vs 5GHz effective. The theory is, if you have 3 lanes of traffic moving consistently at 70MPH, you will have more cars go from point A to point B than 4 lanes of traffic moving at 50MPH.
Memory bandwidth calculation is as such: Memory interface / 8 * memory clock * multiplier (2 for GDDR3 and 4 for GDDR5). So 512 / 8 * 1250 * 4 = 320gbps on the R9-290. The GTX 780 TI would be 384 / 8 * 1750 * 4 = 336gbps.
**Disclaimer** The suggested performance is a mix of medium-near max settings. Generally things like AA/Shadows/SSAO are turned down a few clicks while other enhancements can be increased.
Cheap Under $100
Used
There are really no cards under $100 worth buying as the APU's outrun most cards at this level.
Entry Level $150
NVidia Geforce GTX 1650 4GB GDDR5
I personally wouldn't recommend this card as the competition flat out, outruns it in every scenario. But if you're a NVidia fan, this is the best New card for less than $150.
AMD RX570 4GB GDDR5
This card can be found brand new for under $150 all day and as cheap as $120 in some cases. If you can snag this card for $140 or less, it is a great deal. Can play most modern games, 1080p medium-high settings and eSports games at near max detail! Looking at 1080p 60 fps
Mid Ranged $175-225
NVidia Geforce GTX 1660 6GB GDDR5
This is quite a capable card. Able to push above 60 FPS on high settings in most modern titles, this is where the value starts to slow down. Past here, the ROI begins to dwindle. 1080p 100 FPS/1440p 75 FPS+ is to be expected.
AMD RX 590 8GB GDDR5
This card is a direct competitor of the 1660 and is also a good buy. Similar performance, however with AMD you can get an inexpensive freesync monitor to pair with this. 1080p 100 FPS/1440p 75 FPS+ is to be expected.
Upper Mid Ranged $300-400
AMD RX 5700 8GB GDDR6
This is where things start to really heat up. Literally, this card runs so incredibly hot and loud. Buy a card with an after market cooler from the board partners. 1080p 144 FPS/1440p 100 FPS+ is to be expected.
NVidia Geforce RTX 2060 6GB GDDR6 and 2060 Super 8GB GDDR6
The 2060 Super is over $400 right now making it not a fantastic buy. But there are a few non-Supers under $350 which is not a horrible deal. The RX 5700 beats both the 2060 and 2060 super offering a better value but not Gsync capable like the 2060's are. 1080p 144 FPS/1440p 100 FPS+ is to be expected.
High End $400-500
NVidia Geforce RTX 2070 8 GB GDDR6 and 2070 Super 8GB GDDR6
This new super card kind of makes sense, at least when it launched. It showed the 2070 super beating the RX 5700 XT, except a few things changed. AMD lowered the price by $50, they found ways to cool it better and thus even the Super Card does not beat the 5700 XT. The 2070 is a decent buy but not for over $400 and the super cards are over $500 at the moment. 1440p 120 FPS+ is to be expected.
AMD RX 5700 XT 8GB GDDR6
Like it's little brother, unless you wait until the custom cooled ones come out, there are issues with the reference design. But you get RTX 2070 Super Performance for $400. 1440p 120 FPS+ is to be expected.
Top Tier $500+
NVidia Geforce RTX 2080 8GB GDDR6 & More
Anything this tier and above, you probably don't need my help. But these cards are designed for 4k gaming 60Hz+ at this point.
Cheap Under $100
Used
There are really no cards under $100 worth buying as the APU's outrun most cards at this level.
Entry Level $150
NVidia Geforce GTX 1650 4GB GDDR5
I personally wouldn't recommend this card as the competition flat out, outruns it in every scenario. But if you're a NVidia fan, this is the best New card for less than $150.
AMD RX570 4GB GDDR5
This card can be found brand new for under $150 all day and as cheap as $120 in some cases. If you can snag this card for $140 or less, it is a great deal. Can play most modern games, 1080p medium-high settings and eSports games at near max detail! Looking at 1080p 60 fps
Mid Ranged $175-225
NVidia Geforce GTX 1660 6GB GDDR5
This is quite a capable card. Able to push above 60 FPS on high settings in most modern titles, this is where the value starts to slow down. Past here, the ROI begins to dwindle. 1080p 100 FPS/1440p 75 FPS+ is to be expected.
AMD RX 590 8GB GDDR5
This card is a direct competitor of the 1660 and is also a good buy. Similar performance, however with AMD you can get an inexpensive freesync monitor to pair with this. 1080p 100 FPS/1440p 75 FPS+ is to be expected.
Upper Mid Ranged $300-400
AMD RX 5700 8GB GDDR6
This is where things start to really heat up. Literally, this card runs so incredibly hot and loud. Buy a card with an after market cooler from the board partners. 1080p 144 FPS/1440p 100 FPS+ is to be expected.
NVidia Geforce RTX 2060 6GB GDDR6 and 2060 Super 8GB GDDR6
The 2060 Super is over $400 right now making it not a fantastic buy. But there are a few non-Supers under $350 which is not a horrible deal. The RX 5700 beats both the 2060 and 2060 super offering a better value but not Gsync capable like the 2060's are. 1080p 144 FPS/1440p 100 FPS+ is to be expected.
High End $400-500
NVidia Geforce RTX 2070 8 GB GDDR6 and 2070 Super 8GB GDDR6
This new super card kind of makes sense, at least when it launched. It showed the 2070 super beating the RX 5700 XT, except a few things changed. AMD lowered the price by $50, they found ways to cool it better and thus even the Super Card does not beat the 5700 XT. The 2070 is a decent buy but not for over $400 and the super cards are over $500 at the moment. 1440p 120 FPS+ is to be expected.
AMD RX 5700 XT 8GB GDDR6
Like it's little brother, unless you wait until the custom cooled ones come out, there are issues with the reference design. But you get RTX 2070 Super Performance for $400. 1440p 120 FPS+ is to be expected.
Top Tier $500+
NVidia Geforce RTX 2080 8GB GDDR6 & More
Anything this tier and above, you probably don't need my help. But these cards are designed for 4k gaming 60Hz+ at this point.