Five Years Later: Revisiting the GeForce GTX 970

[Julio Franco]

Recently we've looked back at the GeForce GTX 980 Ti and the GTX 960, both popular GPUs from yesteryear. Those features have been warmly welcomed, but besides the overall positive responses what we noticed in common in your feedback was the request to test the GeForce GTX 970, which was the performance/value offering of the time and a GPU some of you are still rocking in today's games with some success.

Read the full article here.

https://www.techspot.com/review/1822-geforce-gtx-970-vs-radeon-r9-290/

 

[JimEs]

Nice revisit, I rate it 3.5/4

 

[NightAntilli]

Should have thrown an R9 390 in the mix.

 

[QuantumPhysics]

1080p gaming is easy. Most games are in no way demanding anymore unless you wanna play in ULTRA settings 4K 60FPS with Ray Tracing.

 

[JimEs]

Should have thrown an R9 390 in the mix.

To be fair, R9 390 is just R9 290 with extra VRAM and some slight overclock.

 

[deanimate]

Oh yes, looking forward to upgrading this 970 and 2600k cpu. Maybe it will be an AMD upgrade if this years releases fulfill their promise.

 

[NightAntilli]

By the way... The attribution given to driver improvement which is what supposedly improves the performance, is an assumption. It might as well be that the way games are developed has changed and uses the hardware more efficiently. It is well-known that AMD cards suffered from high CPU overhead due to API constraints... I suspect... Now that more cores are being used rather than throwing everything at one or two cores, the AMD card can stretch its legs just a bit more.

 

[Steve]

By the way... The attribution given to driver improvement which is what supposedly improves the performance, is an assumption. It might as well be that the way games are developed has changed and uses the hardware more efficiently. It is well-known that AMD cards suffered from high CPU overhead due to API constraints... I suspect... Now that more cores are being used rather than throwing everything at one or two cores, the AMD card can stretch its legs just a bit more.

Ahh no it's not mate, we've benchmarked the improvements brought about by new drivers in plenty of games over the years.

Should have thrown an R9 390 in the mix.

It's a 290 vs 970 comparison and frankly I didn't feel like benchmarking the same GPU in 33 games again.

 

[nnguy2]

The problem with AMD higher end cards (290/x, Fury/x, Vega 56/64) is that they need to be undervolted to truly unlock their potential. You can easily get 10-15% perf increase with less heat and powerdraw. Unfortunately this requires a very easy tweak via the newer drivers or third party like afterburner. For something that's easily done, unfortunately the majority of users will not bother and get the Nvidia card.

 

[Nightfire]

The R9 290 did a great job coming back after all these years.

Still, I doubt anyone regrets getting the GTX 970 so long as the price was comparable. It still does well in newer games, save Wolf 2. A used GTX 970 would have been a better choice than a 3 GB GTX 1060. Those cards are neck and neck it seems.

 

[NightAntilli]

Ahh no it's not mate, we've benchmarked the improvements brought about by new drivers in plenty of games over the years.

Well, you're convinced. I'm still not convinced. I'm not saying drivers didn't bring improvements. But I am saying that if it was (solely) due to drivers, why is it that if we take a list of say 2015 games, and a list of 2018 games, the R9 290 generally is a better competition in the newer games compared to the older games?

It's a 290 vs 970 comparison and frankly I didn't feel like benchmarking the same GPU in 33 games again.

I can understand that. Benchmarking is definitely a chore... But why did it need to be the R9 290 and not the R9 390 in the first place? The reason I say it, is because;

1) This review (https://www.techspot.com/review/1410-gtx-970-radeon-390-years-later/page12.html) by you, shows the R9 390 a lot closer to the GTX 970 than the R9 290 ever got back then. Also, just a tip for next time, please do mention if they are reference cards or not. I didn't see it in the article (or I'm really blind lol). If you had to benchmark reference cards, then I get it. The R9 390 only were available as reference cards from XFX and chances are you don't have one. If this is the case, disregard the other points below.

2) The R9 390 was a competition for the GTX 970 for slightly longer. 10 months after the R9 290, the GTX 970 was released as you mentioned. The R9 290 competed for 9 months with the GTX 970, until the R9 390 was released. It was slightly improved to be direct competition for the GTX 970 up until the RX480 was release, which was 12 months later. The GTX 970 was released between them. You basically gave nVidia the slight upper hand by having their card that released almost a year later specifically designed to tackle the competition, and (accidentally) robbing AMD of the same chance. And then you compare it to the GTX 1060 and RTX 2060...? What is this? Advertisement for nVidia?

3) Considering the R9 390 AIBs were around ~15% faster than the reference R9 290, had double the RAM, and was equal or cheaper in price than the GTX 970... Yeah... It's not exactly as black and white as this article is painting that exact same GPU. The previously linked article of yours comparing AIB versions of the GTX 970 and the R9 390 tells quite a different story. It makes this article seem like a way to discredit the ones that recommended AMD back then, despite there being completely legitimate reasons to do so. If I had to do so today, I would still recommend an AIB R9 390 over any GTX 970.

4) It was well-known that the R9 390 stretched its legs more at 1440p compared to the GTX 970. Two reasons. CPU overhead, and RAM. Yes. I get it. Looking at the 1080p framerates, it doesn't seem that these cards are viable for 1440p anymore. But that didn't stop you back then in your article stated at #1. Also, the R9 290 was not viable for 1440p due to its 4GB RAM limitation. This would be a great way to test if RAM really is a limit or not, by comparing the 8GB vs the 3.5GB at 1440p. The current argument that 3.5GB is fine, not only do I find it incomplete, I honestly find it appalling to put a product in a good light after nVidia deliberately deceived its customers.

Cheers.

 

[Markoni35]

Did you lie before, or are you lying now? Because you're contradicting your own benchmarks.
In almost every benchmark you did before, Nvidia GTX 970 was faster than AMD R9 290. As can be seen on your own pages:
https://www.techspot.com/review/885-nvidia-geforce-gtx-970-gtx-980/page3.html

But now you've changed the story, and GTX-970 is now 3% slower than R9-290. How is that possible?
1. Did you rig the benchmarks?
2. Did you choose games that support the new version of the truth?
3. Or does Nvidia degrade their drivers for older video cards? Which would again contradict your own findings, where you said that Nvidia is not doing that.

What's the truth? Because there can't be 10 different truths. There's only one. The rest are lies or mistakes.

 

[Vulcanproject]

I think a lot of the gain is still the amount of memory the AMD card has but particularly the memory bandwidth advantage, which is huge.

AMD do typically improve their drivers slower. Or more accurately they simply don't get the most out of their cards early on.....

Its the same scenario if you look at a 7970GHz 3GB and a GTX680 2GB. It's definitely partly memory, but the fact is when the card launched AMD's drivers for it were rubbish. It took about 2-3 years for it to show its true potential and then it would equal or often exceed the Nvidia card.

This is a good thing if you buy a high end card and still have it 3+ years later. However I suspect for most people that buy those cards new its not really relevant at that point.

New cards have come, beaten them and use a lot less power while not costing exorbitant amounts. If you bought high end you probably replaced it already, if you didn't then you likely will consider it soon. Even as this demonstrates a $130 RX570 4GB is more than a match for a GTX970, and a $220 GTX1660 obliterates it.

 

[Lew Zealand]

Did you lie before, or are you lying now? Because you're contradicting your own benchmarks.
In almost every benchmark you did before, Nvidia GTX 970 was faster than AMD R9 290. As can be seen on your own pages:
https://www.techspot.com/review/885-nvidia-geforce-gtx-970-gtx-980/page3.html

But now you've changed the story, and GTX-970 is now 3% slower than R9-290. How is that possible?
1. Did you rig the benchmarks?
2. Did you choose games that support the new version of the truth?
3. Or does Nvidia degrade their drivers for older video cards? Which would again contradict your own findings, where you said that Nvidia is not doing that.

What's the truth? Because there can't be 10 different truths. There's only one. The rest are lies or mistakes.

I dunno what conspiracy you're looking for but there ain't one here.

That was 2014. The problem with 2014 is that games released in 2015-2019 hadn't happened yet because 2015-2019 hadn't happened yet.

The.
Games.
Are.
DIFFERENT.

Steve benchmarked todays games because *today* people are playing *today's* games. Weird but true, eh? He also commented just above there —have a look— that NVidia doesn't degrade their drivers. He's done the tests.

 

[STbob]

GTX 970 still kicking butt in 2019. I have one in my 2nd gaming system and it still plays all the titles well at 1080p no compelling reason to change it either.

 

[rrwards]

I recently moved to a much smaller mITX case. Unfortunately, my 290x is still running everything pretty well at 1080p. I wish there was a good upgrade for AMD that didn't use a ton of power/produce a ton of heat. As it stands, my only viable budget upgrade path is NVidia and the 1660 Ti, but idk if it's worth the cash for 40ish% gains.

 

[SirDigby]

Thanks for this Steve!

 

[ZackL04]

Where are the overclocked results. The 970 overclocks way better than the 290 no?

 

[harm9963]

Still have my 290x CFX, and will always love it , but upgraded to a 1080Ti .
https://www.youtube.com/watch?v=NObuADv10Nw
https://www.youtube.com/watch?v=LqAM38HoUiU

 

[Steve]

Did you lie before, or are you lying now? Because you're contradicting your own benchmarks.
In almost every benchmark you did before, Nvidia GTX 970 was faster than AMD R9 290. As can be seen on your own pages:
https://www.techspot.com/review/885-nvidia-geforce-gtx-970-gtx-980/page3.html

But now you've changed the story, and GTX-970 is now 3% slower than R9-290. How is that possible?
1. Did you rig the benchmarks?
2. Did you choose games that support the new version of the truth?
3. Or does Nvidia degrade their drivers for older video cards? Which would again contradict your own findings, where you said that Nvidia is not doing that.

What's the truth? Because there can't be 10 different truths. There's only one. The rest are lies or mistakes.

LOL

 

[Evernessince]

Where are the overclocked results. The 970 overclocks way better than the 290 no?

I don't think there was a general conclusion on that. While there were 970 samples that did overclock very well, there were also bad samples that did not. The 970 was highly silicon dependent. The R9 290 on the otherhand was more heat restricted. If you installed a good cooler or it came with one, you'd be able to get pretty good gains.

 

[kmo911]

Tryed to buy a amd rx 390 but it was faulty. just like having pyramids and pink in screen. got money back faster then... .-)yeah know where you get if youre bad)
I was usin rx 290 tripple core(somehov) for a long time.
then changed to nvidia gf 960 4 gb.
that changing it out wit a rx 480 8gb later was even better showed in 3dmark 99-2006 11 13 fire strike. you had to have atleast a i7-6700k to run decent in 4k 2160p. some games can be run in 4096x2160p too if you got that screen. 3k half to 4k.
youtube channal with 4x2p are supported. but you gonna needing a good line. 100mb-1gb. 10g price in nok 149 000 are still to exspensive from altibox signal.
tryed many games with better cpu gp ram. ddr 3 4.
so if you got the money to go to rx 20xx 70 80 ti you can blaze trough many games wit latest i9-9900 (k f m) versions. we just have to wait for rx 2080 ti on laptops too with full support for rog gaming and so pcs desktop o.c too.
im at i7-9700k asus prime z370 p II and not yet buildt but in box gigabyte z390 gaming sli rev 1.0 and gtx 1090 ti 11gb (can run rts now) in april drivers but only low. the rtx 2070 can bring it up to better fps o.c
ive gave avay my i7 6700k and my gtx 1070 8gb and rx 480 8gb. just hope it get used. tanks D uncle fireman sam keept his promise. C got a good gpu cpu.

 

[treetops]

Ahh no it's not mate, we've benchmarked the improvements brought about by new drivers in plenty of games over the years.

Well, you're convinced. I'm still not convinced. I'm not saying drivers didn't bring improvements. But I am saying that if it was (solely) due to drivers, why is it that if we take a list of say 2015 games, and a list of 2018 games, the R9 290 generally is a better competition in the newer games compared to the older games?

It's a 290 vs 970 comparison and frankly I didn't feel like benchmarking the same GPU in 33 games again.

I can understand that. Benchmarking is definitely a chore... But why did it need to be the R9 290 and not the R9 390 in the first place? The reason I say it, is because;

1) This review (https://www.techspot.com/review/1410-gtx-970-radeon-390-years-later/page12.html) by you, shows the R9 390 a lot closer to the GTX 970 than the R9 290 ever got back then. Also, just a tip for next time, please do mention if they are reference cards or not. I didn't see it in the article (or I'm really blind lol). If you had to benchmark reference cards, then I get it. The R9 390 only were available as reference cards from XFX and chances are you don't have one. If this is the case, disregard the other points below.

2) The R9 390 was a competition for the GTX 970 for slightly longer. 10 months after the R9 290, the GTX 970 was released as you mentioned. The R9 290 competed for 9 months with the GTX 970, until the R9 390 was released. It was slightly improved to be direct competition for the GTX 970 up until the RX480 was release, which was 12 months later. The GTX 970 was released between them. You basically gave nVidia the slight upper hand by having their card that released almost a year later specifically designed to tackle the competition, and (accidentally) robbing AMD of the same chance. And then you compare it to the GTX 1060 and RTX 2060...? What is this? Advertisement for nVidia?

3) Considering the R9 390 AIBs were around ~15% faster than the reference R9 290, had double the RAM, and was equal or cheaper in price than the GTX 970... Yeah... It's not exactly as black and white as this article is painting that exact same GPU. The previously linked article of yours comparing AIB versions of the GTX 970 and the R9 390 tells quite a different story. It makes this article seem like a way to discredit the ones that recommended AMD back then, despite there being completely legitimate reasons to do so. If I had to do so today, I would still recommend an AIB R9 390 over any GTX 970.

4) It was well-known that the R9 390 stretched its legs more at 1440p compared to the GTX 970. Two reasons. CPU overhead, and RAM. Yes. I get it. Looking at the 1080p framerates, it doesn't seem that these cards are viable for 1440p anymore. But that didn't stop you back then in your article stated at #1. Also, the R9 290 was not viable for 1440p due to its 4GB RAM limitation. This would be a great way to test if RAM really is a limit or not, by comparing the 8GB vs the 3.5GB at 1440p. The current argument that 3.5GB is fine, not only do I find it incomplete, I honestly find it appalling to put a product in a good light after nVidia deliberately deceived its customers.

Cheers.

Yeah it's crazy 970s still cost more even used. You should keep your complex mathematics to yourself so all the crypto miners don't catch on and make a fortune!

 

[codgerface]

Great revisit, Steve! This will be very helpful for plenty of 290/970 owners contemplating their path forward. Appreciate the hard work.

...

2) The R9 390 was a competition for the GTX 970 for slightly longer. 10 months after the R9 290, the GTX 970 was released as you mentioned. The R9 290 competed for 9 months with the GTX 970, until the R9 390 was released. It was slightly improved to be direct competition for the GTX 970 up until the RX480 was release, which was 12 months later. The GTX 970 was released between them. You basically gave nVidia the slight upper hand by having their card that released almost a year later specifically designed to tackle the competition, and (accidentally) robbing AMD of the same chance. And then you compare it to the GTX 1060 and RTX 2060...? What is this? Advertisement for nVidia?

...

Oh, you haven't seen the Hardware Unboxed videos where he unboxes the monthly cash stipend from nvidia? It's beautiful. Sometimes he streams it live and you can watch as he straight up just types in extra fps for the nvidia cards based on how many stacks of hundreds they sent him.

Sound preposterous? Yes. Cool, I was just trying to keep my story as whack-a-doo as your constant "AMD oppression" conspiracy hunt.

I mean, you couldn't pick a worse target to accuse of clandestine bias... the gold standard of rigorous benchmarking who uses n=30+ for his data and conclusions, and who consistently recommends AMD products when the data/value warrants it.

Also, they're 5 year old cards! I feel like I'm taking crazy pills!

 

[NightAntilli]

Great revisit, Steve! This will be very helpful for plenty of 290/970 owners contemplating their path forward. Appreciate the hard work.



Oh, you haven't seen the Hardware Unboxed videos where he unboxes the monthly cash stipend from nvidia? It's beautiful. Sometimes he streams it live and you can watch as he straight up just types in extra fps for the nvidia cards based on how many stacks of hundreds they sent him.

Sound preposterous? Yes. Cool, I was just trying to keep my story as whack-a-doo as your constant "AMD oppression" conspiracy hunt.

I mean, you couldn't pick a worse target to accuse of clandestine bias... the gold standard of rigorous benchmarking who uses n=30+ for his data and conclusions, and who consistently recommends AMD products when the data/value warrants it.

Also, they're 5 year old cards! I feel like I'm taking crazy pills!

And yet you can bring not one relevant fact to prove me wrong. False equivalencies and appeals to ridicule are cheap tactics that don't convince anyone that is capable of critical thinking.