Nvidia GeForce RTX 4070 Ti Review: A Costly 70-Class GPU
Nvidia’s GeForce RTX 4070 Ti may rank as the worst “secret” graphics card in decades.originally in October RTX 4080 12GB, which later became “unreleased”, Core specs and hardware are unchanged. Only the price and name are new.follow in the footsteps of more powerful people RTX4090 When RTX4080, The card will officially go on sale tomorrow, January 5, 2023. Reviews can go up a day early, though, so potential buyers should at least get some advice before taking the plunge.
The RTX 4070 Ti is best graphics cardIt’s a tough question and we’ll go into more detail in this review. GPU Benchmark Hierarchy It’s faster than most 30-series offerings in our tests, essentially tying the RTX 3090 in at a lower price point. That’s good news. The bad news is that this is his xx70-class graphics card, the most expensive Nvidia has ever released. Prices have gone up from $329 for the GTX 970 to $379 for the GTX 1070, $499 for the RTX 2070 and 2070 Super, $499 for the RTX 3070 and $599 for the 3070 Ti. RTX 4070 Ti.
So what we’re really talking about is RTX3080 12GBNvidia hasn’t revealed an official suggested retail price, but we’ve briefly seen prices dip into the $700-$800 range before apparently running out of supply. Without the crypto boom of 2020-2022, the 3080 12GB and 3080 Ti would probably be in the $800-$900 range.now Ethereum is now Proof of Stake The profits of cryptocurrency mining are in the toilet, but we still seem to be looking at mining-inflated graphics card prices.
There is also another way of looking at things. There are no hard numbers on how many cards have been sold, but all RTX 4090 cards sold at retail are still selling for over $2,000. RTX 4080 card (opens in new tab) On the other hand, while it looks relatively available at a price close to its base price of $1,200, nothing actually sells for less than that price.the same can be said AMD’s New RX 7900 XTX When RX7900XT (opens in new tab)the latter can be found for $899, while the former is mostly sold out and often Costs over $1,200 (opens in new tab).
In other words, there’s clearly enough demand that getting a newer, faster graphics card at a lower price point is a fantasy we’ll never experience. We’ve heard speculation that the price of the RTX 4090 in particular has skyrocketed due to demand from the professional sector. The $2,000 4090 is actually a pretty good deal compared to $5,000-$10,000 if you just want a monster processing numbers for AI research. For a real pro card that might not be that fast! But how far down will the stack go before such experts simply choose the faster card?
Let’s hope it doesn’t drop below $800. Likewise, I hope the scalper doesn’t buy up all the cards and try to resell them on eBay. But it probably takes more than hoped for any of these to come to fruition.
previously covered Nvidia’s Ada Lovelace Architecture, so the only real question is how performance scales down to the AD104 GPU with shaders, memory, caches, shrinking memory interfaces, etc. Leaked details from the rebadged RTX 4080 12GB suggest it will be around 20-25%. It’s slower than the RTX 4080, so it’s in RTX 3090 territory. Along with AMD’s new RX 7900 offering, let’s take a look at the specs for the new and previous generation Nvidia cards.
graphics card | RTX 4070 Ti | RTX4090 | RTX4080 | RTX 3090 Ti | RTX 3080 Ti | RTX3080 | RTX 3070 Ti | RX7900XTX | RX7900XT |
---|---|---|---|---|---|---|---|---|---|
architecture | AD104 | AD102 | AD103 | GA102 | GA102 | GA102 | GA104 | Navi 31 | Navi 31 |
process technology | TSMC 4N | TSMC 4N | TSMC 4N | Samsung 8N | Samsung 8N | Samsung 8N | Samsung 8N | TSMC N5+N6 | TSMC N5+N6 |
Transistor (billion) | 35.8 | 76.3 | 45.9 | 28.3 | 28.3 | 28.3 | 17.4 | 45.6 + 6x 2.05 | 45.6 + 5x 2.05 |
Die size (mm^2) | 294.5 | 608.4 | 378.6 | 628.4 | 628.4 | 628.4 | 392.5 | 300 + 222 | 300 + 185 |
SMS | 60 | 128 | 76 | 84 | 80 | 68 | 48 | 96 | 84 |
GPU shader | 7680 | 16384 | 9728 | 10752 | 10240 | 8704 | 6144 | 12288 | 10752 |
Tensor cores | 240 | 512 | 304 | 336 | 320 | 272 | 192 | none | none |
Ray Tracing “Core” | 60 | 128 | 76 | 84 | 80 | 68 | 48 | 96 | 84 |
Boost Clock (MHz) | 2610 | 2520 | 2505 | 1860 | 1665 | 1710 | 1765 | 2500 | 2400 |
VRAM Speed (Gbps) | twenty one | twenty one | 22.4 | twenty one | 19 | 19 | 19 | 20 | 20 |
VRAM (GB) | 12 | twenty four | 16 | twenty four | 12 | Ten | 8 | twenty four | 20 |
VRAM bus width | 192 | 384 | 256 | 384 | 384 | 320 | 256 | 384 | 320 |
L2 cache | 48 | 72 | 64 | 6 | 6 | Five | Four | 96 | 80 |
ROP | 80 | 176 | 112 | 112 | 112 | 96 | 96 | 192 | 192 |
TMU | 240 | 512 | 304 | 336 | 320 | 272 | 192 | 384 | 336 |
TFLOPS FP32 | 40.1 | 82.6 | 48.7 | 40 | 34.1 | 29.8 | 21.7 | 61.4 | 51.6 |
TFLOPS FP16 (FP8/INT8) | 321 (641) | 661 (1321) | 390 (780) | 160 (320) | 136 (273) | 119 (238) | 87 (174) | 123 (123) | 103 (103) |
Bandwidth (GBps) | 504 | 1008 | 717 | 1008 | 912 | 760 | 608 | 960 | 800 |
TBP (Watts) | 285 | 450 | 320 | 450 | 350 | 320 | 290 | 355 | 300 |
release date | January 23rd | October 22nd | November 22 | March 22 | June 21st | September 20 | June 21st | December 22nd | December 22nd |
launch price | $799 | $1,599 | $1,199 | $1,999 | $1,199 | $699 | $599 | $999 | $899 |
We’ve mentioned before that Nvidia appears to be pushing the range of performance offered with the RTX 40 series, and the same thing happened with the RTX 30 series. Up to RTX 20 series GPUs, RTX 2080 Ti performance was approximately double that of RTX2060On the other hand, the previous generation RTX 3090 Ti was three times faster than RTX3050We don’t even have a regular RTX 4070 at the moment. Never mind the future RTX 4050. Plus, we’re already looking at the possibility of doubling the performance from the bigger and more expensive RTX 4090.
Of course, things don’t scale perfectly, but the 4090 has more than twice as many GPU shaders, Tensor Cores, RT Cores, etc. The memory and memory bandwidth are also exactly doubled, but the L2 cache is only 50% more. Workloads like AI training and inference come to mind, but a card that costs twice as much will give you twice the performance of his and consume twice the power. But we’ll get to the benchmark soon.
What about AMD’s RX 7900 series cards? The RX 7900 XT costs $100 more than the 4070 Ti and looks to have better specs in at least some respects. For example, the teraflops numbers favor AMD’s GPUs by nearly 30%, with AMD giving users 67% more memory and 59% more memory bandwidth. Nvidia will undoubtedly continue to dominate when it comes to ray tracing performance and deep learning workloads, but AMD may have an edge when it comes to rasterization. But we must also ask: Want to spend nearly $1,000 on a graphics card that relegates ray tracing to tier two status?
We know for sure: there aren’t many games that need ray tracing yet. Also, quite a few games that support ray tracing seem to use it only to slightly improve image fidelity, with a significant performance penalty.but there that is Games like Control and Cyberpunk 2077 can make a difference, so spending a little more (and potentially saving money in this case) to get a better feature set is something many gamers will appreciate. seems to be happy to do so.
In addition to all the usual graphics hardware, Nvidia’s latest Ada Lovelace GPUs have some extra features. AMD does not yet have a Tensor Core equivalent. Its AI accelerator shares the same execution unit as the GPU shader. In other words, the 4070 Ti could potentially triple the throughput of the RX 7900 XT for FP16 workloads. And for AI inference work that can take advantage of FP8 hardware, Nvidia doubles the performance even more.
It also has an enhanced Optical Flow Accelerator (OFA) that enhances the DLSS 3 algorithm and can be used for other video-related tasks. According to Nvidia, more than 50 games using DLSS 3 are in the works, with about 15 currently available. Even better news is that all of these games with DLSS 3 also support his DLSS 2 and Reflex. That means anyone with an RTX card can benefit — and not just frame generation.
If you’re interested in AI research, especially if you’re interested in a professional workstation without spending a fortune, Nvidia’s GeForce cards are worth a look. Quite a few AI tools are already available, such as Stable Diffusion, ChatGPT. It’s possible to run these tools on non-Nvidia GPUs, but after months of searching for good AI benchmarks, many of the open source tools are Nvidia-focused, while AMD (or Intel) GPUs are You could say it’s an afterthought at best.