MSI GeForce RTX 4070 Ti Super Ventus 3X

[W1zzard]

The MSI GeForce RTX 4070 Ti Super Ventus 3X comes a triple-slot, triple-fan cooler that achieves excellent noise levels. Unfortunately the card required two BIOS updates to bring it closer to expected performance, and the story might not be over yet.

Show full review

 

[tvshacker]

Thanks for the review. The average performance comparison is really eye opening. The 4070Ti seems to make more sense now, just bad pricing.

 

[Nekajo]

The increase from 4070 Ti 12GB to 4070 Ti SUPER 16GB is actually less than I thought Even in 4K UHD gaming which few people actually does without using upscaling to begin with.

Those 4K UHD minimum framerates just confirms that 12GB is plenty still and this probably won't change for years.

It seems that Nvidia just needs less vRAM than AMD to deliver same perf. Better memory compression? Better cache hit/miss system?

 

[5 o'clock Charlie]

Thank you again for another in-depth and informative review @W1zzard. On Page #3, I thought I saw a Zotac card for the first two images of "The Card" section for this MSI review. When I refreshed the page, they were gone. Was this quickly corrected or do I need more coffee this morning? Thanks.

 

[W1zzard]

I thought I saw a Zotac card for the first two images of "The Card" section for this MSI review. When I refreshed the page, they were gone. Was this quickly corrected or do I need more coffee this morning? Thanks.

Yeah I forgot to shoot those two MSI pics, but had to reassemble the card first, took me 5 minutes

 

[Deleted member 237269]

Thanks @W1zzard sadly you guys didn’t receive the MSI gaming slim version and the asus rog.

 

[W1zzard]

Thanks @W1zzard sadly you guys didn’t receive the MSI gaming slim version and the asus rog.
View attachment 331013View attachment 331014

MSI isn't sampling the gaming slim at all, I've asked for a sample a few days ago, they are discussing it internally

Reviews for cards more expensive than the MSRP, like the Strix, can't go live yet, just one more day

 

[Deleted member 237269]

MSI isn't sampling the gaming slim at all, I've asked for a sample a few days ago, they are discussing it internally

Reviews for cards more expensive than the MSRP, like the Strix, can't go live yet, just one more day

I got Jebaited, was thinking it was the MSI 4070 Ti Super Gaming Slim from another website and i was wondering why TPU didn’t posted yet.
https://www.nikktech.com/main/artic...tx-4070-gaming-slim-12gb-graphics-card-review

And it’s not looking good if they don’t want to give reviewers a sample... Gaming slim should be the higher model than the ventus i believe ?

 

[THU31]

This model definitely seems to have issues, noticeably worse results compared to the ASUS TUF for example.

But the drivers seem to have issues as well. In HUB's testing, the card was actually slower than the original 4070 Ti in Modern Warfare 3.

Are both the ASUS and MSI actually going to sell for $800? I thought leaks were mentioning $850-900, which would be ridiculous.

 

[Garrus]

So even if bios updates improve it, at best it is 10 percent faster than the Ti, but the 4080 is 15 percent faster than this one. I really don't get the positioning here, it didn't even hit the midway point between the two old cards (I'm referring to the Asus Tuf results)... much slower than expected.

 

[NvidiaAmdIntel]

MSI isn't sampling the gaming slim at all, I've asked for a sample a few days ago, they are discussing it internally

Reviews for cards more expensive than the MSRP, like the Strix, can't go live yet, just one more day

darn, thats one of the few 4070 TI Super cards that fits my small pc case lol. Guess i'll be buying blind.

based off the 4070 ti (non super) versions of the same cards would you go Gaming Slim or ProArt?

 

[W1zzard]

based off the 4070 ti (non super) versions of the same cards would you go Gaming Slim or ProArt?

I haven't tested either of those cards so I have no opinion. Buy a bigger case

Edit: I just rerendered the charts for the MSI review. Looks like some of them still had old data from before the 2nd BIOS update

 

[Deleted member 237269]

Edit: I just rerendered the charts for the MSI review. Looks like some of them still had old data from before the 2nd BIOS update

Don’t they fuckin test before sending samples and shipping the products to all retailers or what. This baffle me.
It’s not or two units but all reviewers with theses issues.

 

[W1zzard]

Don’t they fuckin test before sending samples and shipping the products to all retailers or what. This baffle me.
It’s not or two units but all reviewers with theses issues.

Every single card is tested at the factory. NVIDIA sent these samples to reviewers, so I'd expect they did some testing, too, at least spot checks. A 5% delta can be missed very easily though

 

[Nekajo]

So even if bios updates improve it, at best it is 10 percent faster than the Ti, but the 4080 is 15 percent faster than this one. I really don't get the positioning here, it didn't even hit the midway point between the two old cards (I'm referring to the Asus Tuf results)... much slower than expected.

You are looking at 4K results, yet most PC gamers don't use 4K and the true 4K king is 4090 anyway. No other card comes close in 4K. 4080 SUPER will probably still be 15-20% slower than 4090 here.

The difference between 4070 Ti SUPER and 4080 in 1440p is only 12% and with a 400 dollar lower pricetag and lower power usage it seems fine to me.

Nvidia gave people what they wanted - BIGGER BUS and MORE VRAM yet the performance did not move much. Just shows that bus width and vRAM is not everything. Like many people seem to think on this forum (maybe they read AMDs vRAM marketing slide too many times)

I will be getting 4080 SUPER tho. One week and I will be back to team Green. Has tons of RTX games in my backlog and can't wait to play around with RTX Remix in older games.

 

[THU31]

Nvidia gave people what they wanted - BIGGER BUS and MORE VRAM yet the performance did not move much. Just shows that bus width and vRAM is not everything. Like many people seem to think on this forum (maybe they read AMDs vRAM marketing slide too many times)

Yeah, none of these cards really need the extra VRAM or bandwidth. You have to push them so hard to run out of VRAM that performance is too low anyway. But the main issue people have is that all cards this generation are very expensive, but they have limited VRAM at the same time. They should either be cheaper, or have more VRAM.

I'm actually really looking forward to the next generation of cards because of GDDR7. The 3 GB modules might finally end the VRAM discussions. 128-bit cards will be able to have 12 GB, 192-bit 18 GB and 256-bit 24 GB.
What I'm hoping for is 4080 performance at 200-220 W. The card could have a 192-bit bus with 18 GB of VRAM. At 32 Gbps, that would offer 768 GB/s of bandwidth, which is even more than the 4080 has.

 

[Nekajo]

Yeah, none of these cards really need the extra VRAM or bandwidth. You have to push them so hard to run out of VRAM that performance is too low anyway. But the main issue people have is that all cards this generation are very expensive, but they have limited VRAM at the same time. They should either be cheaper, or have more VRAM.

I'm actually really looking forward to the next generation of cards because of GDDR7. The 3 GB modules might finally end the VRAM discussions. 128-bit cards will be able to have 12 GB, 192-bit 18 GB and 256-bit 24 GB.
What I'm hoping for is 4080 performance at 200-220 W. The card could have a 192-bit bus with 18 GB of VRAM. At 32 Gbps, that would offer 768 GB/s of bandwidth, which is even more than the 4080 has.

I agree on GDDR7, looking great. RTX 5000 should get this I guess?

I don't think Radeon 8000 will get GDDR7, since AMD seems to focus on performance per dollar with this generation and GDDR7 will be expensive in the beginning


I could see RTX 5080 with 24 gigs GDDR7 on a 256 bit bus and RTX 5090 with 32 gigs on a 320 bit bus, or 36GB with 384 bit, looks weird tho. 18GB vRAM and 36GB vRAM looks kinda strange. Some rumours about 5090 said 512 bit bus too... Doubt it


As proces improves, prices go up. TSMC wants a bigger and bigger cut too.
3nm TSMC is probably 50% higher cost (per wafer) than 4-5nm TSMC.

I don't expect prices to fall anytime soon for higher end GPUs or chips in general

 

[THU31]

I could see RTX 5080 with 24 gigs GDDR7 on a 256 bit bus and RTX 5090 with 32 gigs on a 320 bit bus, or 36GB with 384 bit, looks weird tho. 18GB vRAM and 36GB vRAM looks kinda strange. Some rumours about 5090 said 512 bit bus too... Doubt it


As proces improves, prices go up. TSMC wants a bigger and bigger cut too.
3nm TSMC is probably 50% higher cost (per wafer) than 4-5nm TSMC.

I don't expect prices to fall anytime soon for higher end GPUs or chips in general

There will also be 2 GB GDDR7 modules available, so some cards could use those, depending on what bus they get. 256-bit bus can have 16 or 24 GB, 384-bit can have 24 or 36.
It's hard to believe that the 5090 could have a 512-bit bus. How much faster can the card be to need that bandwidth? The memory speed itself will offer a 40-50% increase over the fastest GDDR6X.
I don't think they would be happy putting 36 GB on the 5090. This is supposed to be a gaming card, not a professional card, 24 seems much more reasonable.

As for TSMC, early rumors suggested a 25% price increase for 3 nm wafers ($20k vs. $16k), but that was a while ago, so it could change.

I'm fine with high prices as long as they give good value. Efficiency is the main thing I care about currently, I don't want another heater like the 3080 I had, it was pretty much unusable in the summer. If I can get 40-50% more performance over the 4070 at similar power with enough VRAM, I can pay.

 

[Nekajo]

There will also be 2 GB GDDR7 modules available, so some cards could use those, depending on what bus they get. 256-bit bus can have 16 or 24 GB, 384-bit can have 24 or 36.
It's hard to believe that the 5090 could have a 512-bit bus. How much faster can the card be to need that bandwidth? The memory speed itself will offer a 40-50% increase over the fastest GDDR6X.
I don't think they would be happy putting 36 GB on the 5090. This is supposed to be a gaming card, not a professional card, 24 seems much more reasonable.

As for TSMC, early rumors suggested a 25% price increase for 3 nm wafers ($20k vs. $16k), but that was a while ago, so it could change.

I'm fine with high prices as long as they give good value. Efficiency is the main thing I care about currently, I don't want another heater like the 3080 I had, it was pretty much unusable in the summer. If I can get 40-50% more performance over the 4070 at similar power with enough VRAM, I can pay.

I don't believe the 512 bit rumours either. Seem excessive and wasted. Especially when more cache is able to lower bus width requirement and faster GDDR7 memory will increase bandwidth alot on its own.

 

[pancake155]

The real question is, does a 4070 Super ventus 3x suffer from this bios issue too? We'll never know because no one reviewed it lol

 

[lexluthermiester]

Unfortunately the card required two BIOS updates to bring it closer to expected performance, and the story might not be over yet.

Thanks for the heads up on that! Might be wise to give MSI some time to sort themselves out

 

[The Von Matrices]

@Wizzard The conclusion on all of these 4070 TI Super reviews incorrectly states that going from a 192-bit to a 256-bit bus is a 25% increase. This must be a slip of the mind because the introductions correctly list it as 33% increase; I'm not sure how that error happened in only the conclusions.

There will also be 2 GB GDDR7 modules available, so some cards could use those, depending on what bus they get. 256-bit bus can have 16 or 24 GB, 384-bit can have 24 or 36.
It's hard to believe that the 5090 could have a 512-bit bus. How much faster can the card be to need that bandwidth? The memory speed itself will offer a 40-50% increase over the fastest GDDR6X.
I don't think they would be happy putting 36 GB on the 5090. This is supposed to be a gaming card, not a professional card, 24 seems much more reasonable.

As for TSMC, early rumors suggested a 25% price increase for 3 nm wafers ($20k vs. $16k), but that was a while ago, so it could change.

I'm fine with high prices as long as they give good value. Efficiency is the main thing I care about currently, I don't want another heater like the 3080 I had, it was pretty much unusable in the summer. If I can get 40-50% more performance over the 4070 at similar power with enough VRAM, I can pay.

I could totally see NVIDIA putting 36GB on their top GPUs. The xx90 series is positioned as a "prosumer" GPU, so its memory capacity isn't solely tied to the amount of memory needed by games. 36GB also isn't enough to eat into their workstation/server GPUs market.

I'm squarely in the market that NVIDIA is targeting with their xx90 cards. I use my GPU when I'm not gaming to experiment with AI models. I can't justify spending twice (or more) money to buy a workstation class card just for this, but I can justify the upgrade cost from a xx80 to a xx90 GPU for this extra capability.

 

[W1zzard]

192-bit to a 256-bit bus is a 25% increase

Correct, it's a 25% decrease from 256 to 192, will fix

 

[Deleted member 237269]

The real question is, does a 4070 Super ventus 3x suffer from this bios issue too? We'll never know because no one reviewed it lol

All the 6× MSI 4070 TI SUPER.
https://videocardz.com/newz/msi-apologizes-for-rtx-4070-ti-super-bios-issues

 

[Night]

There will also be 2 GB GDDR7 modules available, so some cards could use those, depending on what bus they get. 256-bit bus can have 16 or 24 GB, 384-bit can have 24 or 36.

Memory size configuration isn't strictly connected to bus width. 384-bit wide bus can have 12 memory chips on each side of the PCB, doubling the amount of memory per chip's location if needed. For example the RTX 4060 Ti 16 GB version still runs on 128-bit interface, instead of moving to 256-bit.