Rumored Price of RTX 4070 starting at 750 USD

[porina]

11 hours ago, V3ptur said:

Lets hope the launch of this card lowers the 30 series cards by some margin so that it can become more affordable. Over here in the UK 30 series cards are way over prcied, even second hand as well.  

Not going to happen. 30 cards are only still being sold because of over stock. Once it is flushed out, time to launch and sell the lower 40 series. Pricing of new higher 30 series is kinda meaningless now, especially when it costs more than higher performing 40 series. I'd consider 3080 12GB and above to be essentially sold out. The ones that are left are the stupidly priced ones and not representative of what actually sold. 3080 10GB is as low as £680 new in stock at Scan right now and seems to be the highest performing 30 card still with some volume behind it. 3070 is £490 at Scan, new in stock. Whoever you're looking at is more expensive, and I don't consider Scan on the cheap end either. 

[SolarNova]

TBH when it comes to used cards of a previous generation, I don't consider prices 'normal' until we're looking at about 30% off or more and, crucially, has to be below a 'normal' MSRP, not the hyper inflated pricing of the 20, 30 and 40 series cards..

 

For example, during the 20 series launch I bought a 1080ti used WITH a waterblock for £500.

 

The 1080ti was ~£700 card new, - 30% is £490. Other cards at the time were around £400-450 iirc, the one i got had a waterblock which i needed so worked out about right.

 

Bare in mind this was during the 20 series launching which was a terrible generation compared to the 10 series with the only improvement beyond the 1080ti being the 2080ti which cost waaaay more than it should have and started the trend of increasing pricing by an entire tier, which we now see throughout the entire GPU lineup.

As such you could easily argue that under normal circumstances, should the 20 series had been a good generation, the 1080ti would have been cheaper on the used market.

 

Compare that to today, and your not getting anywhere near 30% off  (below 'normal' MSRP) on used cards, and even if you did, the abnormal original MSRP are so high that ur still paying what you would have paid in the past for a brand new top tier GPU.

E.G 3080ti used for £650-£750, (50%-60% off 'new' MSRP) the price of a flagship  'New' 10 series,  9 series etc.

 

Long story short, the GPU market both new and used is in a terrible state for normal 'Joe Blogs' consumers, be they those who usually buy mid tier, or those who buy high tier.

 

For 1080ti like performance your looking at a used RTX 3060 (non ti), which sell for between £300 and £350. Sounds good until u realize the 1080ti released 6 years ago. 6 years and the performance of the card has only dropped in value by half when looking at a newer equivalent used.

Ironically a GTX 1080ti sells for £200-£250 nowadays.

 

The RTX 4070 needs to be around £500 +/- 10% for it not to be DOA. Reality check ..it wont be. So its DOA.

 

[DuckDodgers]

[Rumor] NVIDIA GeForce RTX 4070 to retail at $599: 

https://videocardz.com/newz/nvidia-geforce-rtx-4070-to-retail-at-599

 

Quote

GeForce RTX 4070 MSRP to match RTX 3070 Ti

 

The company has reportedly settled on RTX 4070 non-Ti SKU price at $599. This information was confirmed during a press briefing, which usually takes place before each launch. Reviewers are now expecting their RTX 4070 test units to arrive early next week. Although we confirmed this price with three sources, just be aware that NVIDIA is known for last-minute changes in regard to pricing.

 

At $599, the RTX 4070 will be a lot cheaper than the RTX 4070 Ti, which was initially priced at $899. As a reminder, the Ti model was initially called RTX 4080 12GB, but after a community backlash NVIDIA made a surprising decision to ‘unlaunch’ the card. The same card but with a new name was released at $799 a few weeks later.

 

[porina]

If true this could be the first volume GPU of current gen? Same core count as 3070 doesn't sound like much, do we know if there was much generational IPC vs Ampere? Otherwise it falls down to clock uplift for performance delta, about +40% for nominal boost clocks. Same memory bandwidth as 3070, but with the increased cache. This could put it in 3080 performance ball park (native rendering) which is where you'd expect it to be. Some will never be happy about pricing but I think this is where Ada goes mainstream.

[HenrySalayne]

27 minutes ago, porina said:

Same core count as 3070 doesn't sound like much, do we know if there was much generational IPC vs Ampere?

The 4080 should be 42.5% faster than a 3080 TI considering only shading units and clock speeds.

¯\_(ツ)_/¯

[porina]

35 minutes ago, HenrySalayne said:

The 4080 should be 42.5% faster than a 3080 TI considering only shading units and clock speeds.

¯\_(ツ)_/¯

I tried to compare them on paper at least, but instead I'm going to put 3090 Ti vs 4070 Ti as both are near enough 40 TFLOPS at nominal boost clock. There are no existing 30 to 40 pairings with similar/same core counts to compare. Anyway, looking at the gaming perf of the above, they're near enough the same! This is suggestive of not having a significant IPC change for gaming, gen on gen. Of course changes elsewhere, and gaming mix which might use different GPU features could swing it over time.

[starsmine]

2 hours ago, DuckDodgers said:

[Rumor] NVIDIA GeForce RTX 4070 to retail at $599: 

https://videocardz.com/newz/nvidia-geforce-rtx-4070-to-retail-at-599

 

 

assuming both leaks valid, its nice to see nvidia didnt hard commit to the price (pricing is to the frustration of AIB is not decided until right before launch)

I still dont think the value is great, but at least (because im anchored with ampere pricing) its not offensively bad

3080 performance for the price of a 3070ti.... yay?

[leadeater]

33 minutes ago, porina said:

I tried to compare them on paper at least, but instead I'm going to put 3090 Ti vs 4070 Ti as both are near enough 40 TFLOPS at nominal boost clock.

If you are wanting to figure out IPC gain isn't this the wrong way to do it? You've normalized for compute performance rather than "per core". This would show how well each architecture is able to attain performance from theoretical compute performance but that wouldn't be "IPC". It's quite tricky, is for CPU too as well realistically.

 

Right now the closest two I can see is the RTX 3080 TI (10240/320:112:80:320) and the RTX 4080 (9728/304:112:76:304), 512 CUDA core difference. Edit: Forgot to adjust for clocks. Will edit in after I figure that out. Edit 2: huh so looks like effective net zero IPC based on FP32 and adjusted for boost clocks? Did I miss something?

[porina]

3 minutes ago, leadeater said:

If you are wanting to figure out IPC gain isn't this the wrong way to do it? You've normalized for compute performance rather than "per core". This would show how well each architecture is able to attain performance from theoretical compute performance but that wouldn't be "IPC". It's quite tricky, is for CPU too as well realistically.

I have a hypothetical GPU with 100 cores running at 1 GHz. I have another GPU of same generation running 200 cores at 0.5 GHz. Both are same TFLOPS, and for an ideally scaling workload (gaming or otherwise) should be equal. If I increase IPC by 10%, you'd expect both to increase also 10% right? The cores and clocks kinda cancel each other out. There is a small gotcha in that not all workloads scale ideally. It is often difficult to make optimal use of wider configurations. In case of any ambiguity, IPC is "per core", not over the whole width.

 

The main reason for wanting them to be similar when comparing GPUs is because in normal running the cores are not the only contributing factor. If they're kept roughly the same then hopefully external factors would kinda similarly impact each other too. If I were to test what I'd call peak CPU IPC, I'd intentionally run as low a core count and clock as I can stand, so that the cores are the limiting factor and not anything else. Without getting my hands physically on samples of 40 series, which I'm not willing to pay for so far, I can't do my own testing. And no mainstream tester is going to do anything like this.

 

3 minutes ago, leadeater said:

Right now the closest two I can see is the RTX 3080 TI (10240/320:112:80:320) and the RTX 4070 Ti (7680/240:80:60:240), 512 CUDA core difference. Problem I see is quite large difference in TGP.

TGP will impact clocks, which will cancel out with suitable consideration. BTW please check your numbers, I don't think I see what you are claiming there.

[Blademaster91]

Not surprised at the price, but its still stupidly expensive for a x70 tier card, the launch price for the 3070 was $499 USD.

Also the $599 price will probably only be for the FE card, so it'll be more like $650 for any decent AIB card. I don't really see the point of buying it when a 6950XT is in that price range and is about as fast as a 3080Ti.

[leadeater]

3 minutes ago, porina said:

TGP will impact clocks, which will cancel out with suitable consideration. BTW please check your numbers, I don't think I see what you are claiming there.

Sorry you'll have to re-look at my post, I meant to use the RTX 4080 since I noticed that was actually closer and then dropped the ball on the TGP difference changing between comparison GPUs. 3am, slack needs to be given lol.

[starsmine]

4 minutes ago, porina said:

It is often difficult to make optimal use of wider configurations. In case of any ambiguity, IPC is "per core", not over the whole width.

IPC is instructions per clock/per cycle. Its basicly the end number of how you pipeline/design the architecture. Instructions per core does not even make sense. over how many clocks? over how much time?  

 

scaling workloads in terms of GPUs over multiple cores can generally be ignored unless getting really into the weeds as the workloads are inherently parallel over n-threads. 

[leadeater]

6 minutes ago, porina said:

I have a hypothetical GPU with 100 cores running at 1 GHz. I have another GPU of same generation running 200 cores at 0.5 GHz. Both are same TFLOPS, and for an ideally scaling workload (gaming or otherwise) should be equal. If I increase IPC by 10%, you'd expect both to increase also 10% right? The cores and clocks kinda cancel each other out. There is a small gotcha in that not all workloads scale ideally. It is often difficult to make optimal use of wider configurations. In case of any ambiguity, IPC is "per core", not over the whole width.

True, good point. I ended up coming to the same conclusion you did via the method I choose, seems to all be clock speed improvements rather than IPC. The cache seems to solely lower then memory bandwidth requirement rather than providing meaningful uplift in IPC, or rather prevents IPC loss through memory bandwidth limitations.

[porina]

4 minutes ago, leadeater said:

3am, slack needs to be given lol.

Get some sleep? Forum will still be here tomorrow. Probably.

 

3 minutes ago, starsmine said:

IPC is instructions per clock/per cycle. 

I got that, added the clarification to pre-empt possible misunderstanding.

[leadeater]

4 minutes ago, porina said:

Get some sleep? Forum will still be here tomorrow. Probably.

But I'm listening to Steve yell at the clouds about PSU's and ATX12VO

[ToboRobot]

1 minute ago, leadeater said:

But I'm listening to Steve yell at the clouds about PSU's and ATX12VO

I want to Steve do more yelling at clouds.  The rants this week have been lit.

[SolarNova]

While the latest rumor of £600 is better than $750 ,, its still WAY to high for a xx70 card.

hell even the 30 series $500 was to high.

The xX70 cards should be ~ $470. included the absurd recent inflation.

[HenrySalayne]

18 minutes ago, SolarNova said:

hell even the 30 series $500 was to high.

The xX70 cards should be ~ $470. included the absurd recent inflation.

The 2070 was $500 as well. The 1070 was $380 back then (along with the rest of the 10 series very good value).

[Sagefive]

Buy Intel Arc

[GamerDude]

On 4/1/2023 at 1:04 AM, HenrySalayne said:

The 2070 was $500 as well. The 1070 was $380 back then (along with the rest of the 10 series very good value).

That's where nGreedia realized their mistake, arming the GTX xx70 cards with 8GB VRAM buffer (which was more than enough back then) making them pretty useful even now. Pretty sure some GTX 1070 owners are hanging on to their cards for dear life. And the following two gens, RTX 2070/3070, nGreedia kept VRAM buffer at 8GB fulling knowing that it'd prove insufficient in the long run (NOT everyone upgrade their GPUs yearly or once every two years like some peeps here do). 

 

Meanwhile, AMD kept its 5700 cards at 8GB, but went 12GB on its 6700 XT cards. nV boys could argue, well, what about the RX 6600 series, they still have 8GB. While undeniably true, it can be argued that the RX 6600 series aren't powerful enough to levearge the extra VRAM......you can't argue that with the RTX 2070S/3070/3070 Ti, and perhaps even the RTX 3060 Ti. These cards are powerful enough to leverage the extra VRAM, but have been kneecapped performance-wise due to 8GB VRAM. This deficiency in VRAM buffer has finally reared its ugly head in 2023 with certain games, and I'd bet, Jensen is giving himself a toast and patting himself on the back for a job well done.

 

I feel nGreedia is doing the same for the RTX 4070/4070 Ti, these cards are powerful, yet are allotted with just 12GB VRAM buffer. Some games at nipping at the 12GB VRAM mark at max setting + RT. IF the RX 7800 XT launches at a similar price to the RTX 4070 (which it shouldn't) and is armed with 16GB VRAM buffer, it'd be a repeat of the RTX 3070/3070 Ti/3060 Ti vs RX 6700XT/6800 situation again, with AMD cards aging better than its nGreedia counterparts.

[Deadpool2onBlu-Ray]

Availability seems to be good today so far.

[maartendc]

Uhm, can we all just take a moment to appreciate the fact that this "leaked price" for the RTX 4070 was utter B/S?

 

And lo and behold, who was the source for this "leak"? The Youtuber "Moore's law is dead".

 

I have said it before, but I'll say it again: this guy doesn't have any reputable "sources". And all his "leaks" and "rumors" from "industry sources" are just COMPLETELY made up out of thin air. Why is anyone giving ANY credibility or attention to this guy? He is just some guy in his moms attic making shit up.

[starsmine]

2 hours ago, maartendc said:

Uhm, can we all just take a moment to appreciate the fact that this "leaked price" for the RTX 4070 was utter B/S?

 

And lo and behold, who was the source for this "leak"? The Youtuber "Moore's law is dead".

 

I have said it before, but I'll say it again: this guy doesn't have any reputable "sources". And all his "leaks" and "rumors" from "industry sources" are just COMPLETELY made up out of thin air. Why is anyone giving ANY credibility or attention to this guy? He is just some guy in his moms attic making shit up.

uh... No?

Afaik the leaked price was correct at the time. 

Plans have this wonderful thing of having the ability of being changed, plans are not set in stone.  That is also why its internal leaks, even if a company has full intention of doing a thing,  Annoy companies. Even if they have the intention of doing a thing, they have not committed to doing a thing. They dont like to announce early outside of internal channels when there is any possibility of it changing. Think of all the video games you know that finish pre production. get half way through production, and then get canceled before an announcement trailer even happens. Those games being part of the plan at the time is still true, It was never "BS"

You would have a point if the leak was never true. But all I have found do not indicate that. Nvida at that point was kicking around the idea in a serious way and talking to AIBs about that as a potential MSRP to target and build around. It is well understood the prices are not final until they are final on all sides, to the annoyance of AIBs in most cases. Though here AIBs were not confident in the 750 price so didnt build line ups that could not hit 600. 

Sometimes companies will take advantage of leaks like this and see if they can turn it into goodwill/ positive VR by going, hey we are over a month out, we can drop the price/eat the opportunity cost. They may have gone to the table and saw the 40% drop in PC sales across the market, and realized the 750 would no longer hit their targets in any scenarios on the supply demand curve. Some consultant with sway with those who have influence on final pricing could have convinced them, even though they considered 750, that it would be a poor choice. None of these incidents happening after the initial leak. disqualify the initial leak of being true.

 

[maartendc]

45 minutes ago, starsmine said:

uh... No?

Afaik the leaked price was correct at the time. 

Plans have this wonderful thing of having the ability of being changed, plans are not set in stone.  That is also why its internal leaks, even if a company has full intention of doing a thing,  Annoy companies. Even if they have the intention of doing a thing, they have not committed to doing a thing. They dont like to announce early outside of internal channels when there is any possibility of it changing. Think of all the video games you know that finish pre production. get half way through production, and then get canceled before an announcement trailer even happens. Those games being part of the plan at the time is still true, It was never "BS"

You would have a point if the leak was never true. But all I have found do not indicate that. Nvida at that point was kicking around the idea in a serious way and talking to AIBs about that as a potential MSRP to target and build around. It is well understood the prices are not final until they are final on all sides, to the annoyance of AIBs in most cases. Though here AIBs were not confident in the 750 price so didnt build line ups that could not hit 600. 

Sometimes companies will take advantage of leaks like this and see if they can turn it into goodwill/ positive VR by going, hey we are over a month out, we can drop the price/eat the opportunity cost. They may have gone to the table and saw the 40% drop in PC sales across the market, and realized the 750 would no longer hit their targets in any scenarios on the supply demand curve. Some consultant with sway with those who have influence on final pricing could have convinced them, even though they considered 750, that it would be a poor choice. None of these incidents happening after the initial leak. disqualify the initial leak of being true.

 

Haha I knew someone was going to say that.

 

Bro, MLID always prefaces every supposed "leak" he comes up with out of thin air by saying "of course this is still subject to change".

 

So basically he can never be wrong eh?

If he was wrong, there was just a "change of plans". 

If he was right, there happens to have been no change of plans.

Genius.

 

Even a broken clock is right twice a day.

 

[Spotty]

3 hours ago, maartendc said:

Uhm, can we all just take a moment to appreciate the fact that this "leaked price" for the RTX 4070 was utter B/S?

 

And lo and behold, who was the source for this "leak"? The Youtuber "Moore's law is dead".

 

I have said it before, but I'll say it again: this guy doesn't have any reputable "sources". And all his "leaks" and "rumors" from "industry sources" are just COMPLETELY made up out of thin air. Why is anyone giving ANY credibility or attention to this guy? He is just some guy in his moms attic making shit up.

I don't know, I'm still holding out for that RTX Titan Ada he "leaked" a few months ago. That's definitely coming, right? /s