How do you judge Graphics cards?

[IR76]

Ive just learnt that the Nvidia Geforce 980 is faster than the 1050 and when i looked into the specs the 980 has pretty much triple the amount of cuda cores than the 1050 so my question is how do you compare GPUs? and know whether which one is better because i dont get why the 1050 is slower even though it was released later so why would Nvidia downgrade the performance? 

[Dedayog]

4 minutes ago, IR76 said:

Ive just learnt that the Nvidia Geforce 980 is faster than the 1050 and when i looked into the specs the 980 has pretty much triple the amount of cuda cores than the 1050 so my question is how do you compare GPUs? and know whether which one is better because i dont get why the 1050 is slower even though it was released later so why would Nvidia downgrade the performance? 

Basically, we look at how they perform in the given use case (usually gaming).

 

But Cuda cores coupled with bandwidth coupled with memory speeds, etc. are more the technical side of comparing.  

 

You're saying since a guy is taller, he's a better basketball player.  Bit more to it than that.  

 

But for me, I look at reviews/comparisons/common sense.  I don't give two shits if there is 1048 Cuda cores or 5 sticks of bubble gum in the thing.  How well does it make my games look and run.

[Needfuldoer]

8 minutes ago, IR76 said:

Ive just learnt that the Nvidia Geforce 980 is faster than the 1050 and when i looked into the specs the 980 has pretty much triple the amount of cuda cores than the 1050 so my question is how do you compare GPUs? and know whether which one is better because i dont get why the 1050 is slower even though it was released later so why would Nvidia downgrade the performance? 

They're not comparable products. The GTX 980 was a high end card for its generation, and the GTX 1050 was a low end card for its generation. There's only one generation between them, which isn't enough time for generational improvements to "trickle down" yesterday's flagship card's performance into a newer low to mid tier card.

 

Comparing the GTX 980 against the GTX 1070 or GTX 1080 would be fair.

[Dedayog]

14 minutes ago, IR76 said:

Ive just learnt that the Nvidia Geforce 980 is faster than the 1050 and when i looked into the specs the 980 has pretty much triple the amount of cuda cores than the 1050 so my question is how do you compare GPUs? and know whether which one is better because i dont get why the 1050 is slower even though it was released later so why would Nvidia downgrade the performance? 

You were shown how we figure this out:

 

 

[podkall]

Techpowerup is a good website for raster performance comparison,

 

or simply searching "X vs Y benchmark" on YouTube can give you a rough idea,

 

they key thing to remember, is the performance is both relative and it's only raster performance, this comparison doesn't bring in any other features that the card possesses into equation,

which is why a relatively slower Nvidia card can win against a relatively faster AMD card, because you enabled RayTracing in game:

 

 

• The reason why the 4070 loses in 4K here, is because it runs out of VRAM.

 

8 minutes ago, Needfuldoer said:

They're not comparable products. The GTX 980 was a high end card for its generation, and the GTX 1050 was a low to mid tier card for its generation.

 

Comparing the GTX 980 against the GTX 1070 or GTX 1080 would be fair.

that is also true,

 

it's almost similar to expect a 10 year old car be faster than a 20 year old car, despite the fact that the 20 year old car has max speed of 300mp/h while the 10 year old car has max speed of only 200mp/h,

 

the 10 year old car might have more efficient engine and uses less fuel over same speed and mileage, etc, etc,

 

you can understand where I'm going with this.

[podkall]

6 minutes ago, Dedayog said:

Basically, we look at how they perform in the given use case (usually gaming).

 

But Cuda cores coupled with bandwidth coupled with memory speeds, etc. are more the technical side of comparing.  

 

You're saying since a guy is taller, he's a better basketball player.  Bit more to it than that.  

 

But for me, I look at reviews/comparisons/common sense.  I don't give two shits if there is 1048 Cuda cores or 5 sticks of bubble gum in the thing.  How well does it make my games look and run.

yep, the 4070 has double cores compared to 7800 XT,

 

and is still on par or slightly slower than the AMD card (in raster), despite having double the core count:

[Stahlmann]

Do not look at the specs. Watch reviews. That's how you do it.

[Killjoy_NS]

3 hours ago, IR76 said:

Ive just learnt that the Nvidia Geforce 980 is faster than the 1050 and when i looked into the specs the 980 has pretty much triple the amount of cuda cores than the 1050 so my question is how do you compare GPUs? and know whether which one is better because i dont get why the 1050 is slower even though it was released later so why would Nvidia downgrade the performance? 

Especially today....very hard for me.
Back in the day was "the more, the merrier"...but these days...

Well I have pretty old RX 470 with 4GB of vram...GDDR5 and 256bit memory bus...and I can't understand how is some newer GPU with 8GB and 128bit bus with lets say GDDR6 vram "FASTER" than my GPU.

 

Comparing vram, yes 8GB is better than 4GB...it can load more/bigger textures etc.

But I can't understand how is GDDR6/128bit better than GDDR5/256bit...

 

Maybe there is some "compensation" between bus and VRAM.....I don't know, honestly.

[YoungBlade]

46 minutes ago, Killjoy_NS said:

Comparing vram, yes 8GB is better than 4GB...it can load more/bigger textures etc.

But I can't understand how is GDDR6/128bit better than GDDR5/256bit...

 

Maybe there is some "compensation" between bus and VRAM.....I don't know, honestly

When it comes to raw bandwidth, both the bus width and the speed of the memory need to be taken into account. The 128-bit bus is half the width of the 256-bit bus. However, this can be compensated for if the VRAM itself is faster. In the case of something like the GTX 1070 vs the RTX 4060 Ti, it's that the improved performance of the GDDR6 memory is overcoming the reduction in bus width, giving the RTX 4060 Ti 288GB/s of bandwidth vs the 256GB/s of the GTX 1070.

 

You can actually arrive at those figures by just taking the speed of the individual memory chips and multiplying to get the total bandwidth. So, for the GTX 1070, the memory's bandwidth is 8Gbps, each chip connects with a 32-bit bus, and there are 8 of them (8 x 32 = 256). So 8 x 32 x 8 = 2048Gbps / 8 = 256GB/s (8 bits per byte). Then, with the RTX 4060 Ti, you have 18Gbps memory, each chip with a 32-bit bus, but there are now just 4 of them (4 x 32 = 128). So 18 x 32 x 4 = 2304Gbps / 8 = 288GB/s.

[Killjoy_NS]

1 minute ago, YoungBlade said:

When it comes to raw bandwidth, both the bus width and the speed of the memory need to be taken into account. The 128-bit bus is half the width of the 256-bit bus. However, this can be compensated for if the VRAM itself is faster. In the case of something like the GTX 1070 vs the RTX 4060 Ti, it's that the improved performance of the GDDR6 memory is overcoming the reduction in bus width, giving the RTX 4060 Ti 288GB/s of bandwidth vs the 256GB/s of the GTX 1070.

 

You can actually arrive at those figures by just taking the speed of the individual memory chips and multiplying to get the total bandwidth. So, for the GTX 1070, the memory's bandwidth is 8Gbps, each chip connects with a 32-bit bus, and there are 8 of them (8 x 32 = 256). So 8 x 32 x 8 = 2048Gbps / 8 = 256GB/s (8 bits per byte). Then, with the RTX 4060 Ti, you have 18Gbps memory, each chip with a 32-bit bus, but there are now just 4 of them (4 x 32 = 128). So 18 x 32 x 4 = 2304Gbps / 8 = 288GB/s.

This is the most awesome answer I've seen.

OK, I have XFX RX 470 TripleX gpu: 4GB, 256bit, GDDR5.

I will love to have AMD....don't like nVidia...

 

I will atleast need 8GB of vram.

Which gpu would be significant upgrade for me?

I'm talking about 100% + performance comparing to my GPU.

[Dedayog]

6 minutes ago, Killjoy_NS said:

This is the most awesome answer I've seen.

OK, I have XFX RX 470 TripleX gpu: 4GB, 256bit, GDDR5.

I will love to have AMD....don't like nVidia...

 

I will atleast need 8GB of vram.

Which gpu would be significant upgrade for me?

I'm talking about 100% + performance comparing to my GPU.

Anything RX6600 8GB and above will be 2x (100%) better.

 

I'd recommend the 6650XT bang for buck.

[YoungBlade]

1 minute ago, Killjoy_NS said:

This is the most awesome answer I've seen.

OK, I have XFX RX 470 TripleX gpu: 4GB, 256bit, GDDR5.

I will love to have AMD....don't like nVidia...

 

I will atleast need 8GB of vram.

Which gpu would be significant upgrade for me?

I'm talking about 100% + performance comparing to my GPU.

You can use a site like TechPowerUp to start your search. According to their Relative Performance chart, the first GPU with at least a 100% performance uplift is the RX 6600 (207% of the performance of the RX 470), which does have 8GB of VRAM. At current pricing, if you're in the US, it's going for $210 new.

 

If you want to go with a used option, the RX 5700XT is a bit faster than the RX 6600, and goes for around $150.

 

Or you could go up to the RX 6650XT for $250 or the RX 7600 for $270 if you want some more performance.

[Killjoy_NS]

6 minutes ago, YoungBlade said:

You can use a site like TechPowerUp to start your search. According to their Relative Performance chart, the first GPU with at least a 100% performance uplift is the RX 6600 (207% of the performance of the RX 470), which does have 8GB of VRAM. At current pricing, if you're in the US, it's going for $210 new.

 

If you want to go with a used option, the RX 5700XT is a bit faster than the RX 6600, and goes for around $150.

 

Or you could go up to the RX 6650XT for $250 or the RX 7600 for $270 if you want some more performance.

No "second hand" components. No thanks.
I had bad experience.

 

I can't see any comparing options on website....is there such a thing....it was on "GameDebate"...

 

P.S. I'm planing to do a small upgrade....some B550 mobo and Ryzen 5600...since I have B350 and Ryzen 2600...
Any bottlenecks I should be aware of?

[Tan3l6]

4 minutes ago, Killjoy_NS said:

I had bad experience.

Happens. I've had around 0 problems with 2nd hand PC parts in the past couple of decades.

But yeah, depends where you source your hardware from I guess.

[YoungBlade]

3 minutes ago, Killjoy_NS said:

No "second hand" components. No thanks.
I had bad experience.

 

I can't see any comparing options on website....is there such a thing....it was on "GameDebate"...

 

P.S. I'm planing to do a small upgrade....some B550 mobo and Ryzen 5600...since I have B350 and Ryzen 2600...
Any bottlenecks I should be aware of?

I wouldn't recommend upgrading your motherboard to one with the same socket. With a BIOS update, your B350 board should be able to accept the Ryzen 5 5600, and if your motherboard can handle a Ryzen 5 2600, it can handle a Ryzen 5 5600, as it draws the same amount of power under load. Why do you want to get a new motherboard if the one you have supports the CPU you want?

 

The comparison is the "Relative Performance" section. You can scroll through that and see the performance increase/decrease between cards. The card you are currently looking at is set to 100%, so anything with less than 100% is a performance decrease and anything with greater than 100% is a performance increase. 200% is a 100% uplift.

 

[Killjoy_NS]

9 minutes ago, Tan3l6 said:

Happens. I've had around 0 problems with 2nd hand PC parts in the past couple of decades.

But yead, depends where you source your hardware from I guess.

Well, local websites....sales sites...cause I guess shipping from USA could be more than a component that I'm buying.

 

Well, these are cheapest models in a local store.

 

RX 6600 Challenger

 

XFX SWFT 210 RX 6600

 

Are these any good?

[Tan3l6]

2 minutes ago, Killjoy_NS said:

Are these any good?

Those links lead to nowhere.

 

RX 6600XT would be a tad better but 6600 is decent on 1080p

[Killjoy_NS]

4 minutes ago, YoungBlade said:

I wouldn't recommend upgrading your motherboard to one with the same socket. With a BIOS update, your B350 board should be able to accept the Ryzen 5 5600, and if your motherboard can handle a Ryzen 5 2600, it can handle a Ryzen 5 5600, as it draws the same amount of power under load. Why do you want to get a new motherboard if the one you have supports the CPU you want?

 

The comparison is the "Relative Performance" section. You can scroll through that and see the performance increase/decrease between cards. The card you are currently looking at is set to 100%, so anything with less than 100% is a performance decrease and anything with greater than 100% is a performance increase. 200% is a 100% uplift.

 

Well, my mobo has PciE 3.....isn't that a "bottleneck" for a gpu which uses PciE 4?

This is my mobo.

 

Gigabyte AB350M-DS3H V2

[Killjoy_NS]

3 minutes ago, Tan3l6 said:

Those links lead to nowhere.

 

RX 6600XT would be a tad better but 6600 is decent on 1080p

I've fixed them. Sorry. Usually, LINK is "downstairs" and "upstairs" is a text....here is opposite.

[YoungBlade]

1 minute ago, Killjoy_NS said:

Well, my mobo has PciE 3.....isn't that a "bottleneck" for a gpu which uses PciE 4?

This is my mobo.

 

https://www.gigabyte.com/Motherboard/GA-AB350M-DS3H-V2-rev-11#kf

Not for a card that has at least a x8 connection, which the RX 6600 has. On average, you'd lose about 5% performance. But if you're worried about that, then just spend the additional money you would spend on the motherboard, and get a faster card to compensate. With the money spent on a B550 board, you could go up to an RX 6700 XT for $330, which has a full x16 connection, eliminating the concern, and also is much faster - it has almost 3x the performance of your RX 470 and comes with 12GB of VRAM.

 

Basically, your money could be better spent on making the GPU faster, rather than getting a new motherboard to improve the RX 6600 by a few percent.

[Killjoy_NS]

4 minutes ago, YoungBlade said:

Not for a card that has at least a x8 connection, which the RX 6600 has. On average, you'd lose about 5% performance. But if you're worried about that, then just spend the additional money you would spend on the motherboard, and get a faster card to compensate. With the money spent on a B550 board, you could go up to an RX 6700 XT for $330, which has a full x16 connection, eliminating the concern, and also is much faster - it has almost 3x the performance of your RX 470 and comes with 12GB of VRAM.

 

Basically, your money could be better spent on making the GPU faster, rather than getting a new motherboard to improve the RX 6600 by a few percent.

OK, I thought that PciE3 is far slower than PciE4...

So let's say....this GPU???

6700XT

 

And this GPU is 50 000,00 RSD which is 463$ in my country and we have "lower" standard than USA.

So.....it is WOW...

[Tan3l6]

17 minutes ago, Killjoy_NS said:

OK, I thought that PciE3 is far slower than PciE4...

So let's say....this GPU???

6700XT

 

And this GPU is 50 000,00 RSD which is 463$ in my country and we have "lower" standard than USA.

So.....it is WOW...

The 6700XT is moderately better than 6600 .. good for 60fps 1440p I guess. Not in all games, but decent.

[YoungBlade]

9 minutes ago, Killjoy_NS said:

OK, I thought that PciE3 is far slower than PciE4...

So let's say....this GPU???

6700XT

 

And this GPU is 50 000,00 RSD which is 463$ in my country and we have "lower" standard than USA.

So.....it is WOW...

PCIe 4 is twice as fast as PCIe 3, but this only matters if the PCIe bandwidth itself is a performance limiting factor, which it rarely is. It matters for cards like the RX 6500XT, because it only has a x4 connection. However, when it comes to cards like the RX 6600 or RTX 4060, it doesn't really matter in practice that they have a x8 connection. Even the mighty RTX 4090 doesn't show a meaningful performance loss when going down to a PCIe 3 connection, because it has a full x16 interface.

 

It's not that there's no performance loss at all when using a x8 interface going from PCIe 4 to 3, but it's not meaningful. For the vast majority of games, it's within 5%.

 

That's unfortunate that the RX 6700XT is so much more expensive in your country. It's probably not worth paying 2x the price for, as it's only about 40-50% faster than the RX 6600.