NVIDIA RTX 4000 Ada Lovelace GPU Roughly Equivalent to GeForce RTX 3060 Ti, Consumes 65% Less Power

[T0@st]

The NVIDIA RTX 4000 SFF Ada Generation graphics card was released to the public in late April, but very few reviews and benchmarks have emerged since then. Jisaku Hibi, a Japanese hardware site, has published an in-depth evaluation that focuses mostly on gaming performance. The RTX 4000 Ada SFF has been designed as a compact workstation graphics card, but its usage of an AD104 die makes it a sibling of NVIDIA's GeForce RTX 4070 and 4070 Ti gaming-oriented cards. Several PC hardware sites have posited that the 70 W RTX 4000 Ada SFF would "offer GeForce RTX 3070-like performance," but Jisaku Hibi's investigation points to the RTX 3060 Ti being the closest equivalent card (in terms of benchmark results).

According to the TPU GPU database: "NVIDIA has disabled some shading units on the RTX 4000 SFF Ada Generation to reach the product's target shader count. It features 6144 shading units, 192 texture mapping units, and 80 ROPs. Also included are 192 tensor cores which help improve the speed of machine learning applications. The card also has 48 ray tracing acceleration cores. NVIDIA has paired 20 GB GDDR6 memory with the RTX 4000 SFF Ada Generation, which are connected using a 160-bit memory interface. The GPU is operating at a frequency of 1290 MHz, which can be boosted up to 1565 MHz, memory is running at 1750 MHz (14 Gbps effective)." The SKU's 70 W TGP and limited memory interface are seen as the card's main weak points, resulting in average clock speeds and a maximum memory bandwidth of only 280 GB/s.




Jisaku Hibi's benchmark tests show that the RTX 4000 SFF performs nearest to the Ampere-based (GA104) GeForce RTX 3060 Ti, while consuming 65% less power. The older card was a bit faster, by a maximum 13% margin, but its total graphics power (TGP) is 200 W. The RTX 4000 SFF's newer architecture allows for impressive efficiency, and it beats the RTX 3060 (non-Ti) 170 W card by a 20% margin. It cannot keep pace with the substantially larger and power hungry RTX 3070 and 4070 GPUs.



It is good to know that the small form factor RTX 4000 GPU is versatile beyond its core purpose - but the high asking price ($1250-1500) makes it a sensible choice for visual professional-type buyers only.

View at TechPowerUp Main Site | Source

 

[Pooch]

Here we are again with what seems to be a crippled 4000 series card. How do they do it? Easy, just hobble that memory so it can't go anywhere (280gb/s) but give it plenty of space to move around the room. 20GB Vram (Misery)

This seems to be the new way to stretch out product skus while they partner with board manufacturers to pump out 10 versions of the same card. I said to another guy on here, Im not into playing the game they set up, I will wait. They capitalize on impatient people. Don't give them what they want, which is for you to run out and but the first thing they show you, and then after that they show you the one that has enough ram on it, and you buy that too. So Ill end by stating that the 3060 12GB model, in most situations, is the optimal pick. And if you don't know how to make a 3060 with 12 gigabytes work for you? Then I'm sorry but you are no so good at computer stuff.

My stuff
5600x
16 GB 3600 xmp
x570
wd sn770 1tb
evga 1660 super oc 6gb

 

[lemonadesoda]

Looks great! Small form factor, half height, half length, double wide. This comparison picture shows it all.

Performance/watt is best in class, and although clearly not as fast as the best gaming 4x series, it is good for HD, 2.5K in all cases, and 4K in some. That's pretty impressive as well as providing all the Quadro-purposed CUDA and professional encoding video attributes.

Not too noisy, low power, can put two into the same PC/workstation.

Unfortunately, at a "professional price". ooof.

 

[Pooch]

Of course at a pro price, you didn't actually think that it would be at a price point you could afford did you?

 

[killowatt]

Here we are again with what seems to be a crippled 4000 series card. How do they do it? Easy, just hobble that memory so it can't go anywhere (280gb/s) but give it plenty of space to move around the room. 20GB Vram (Misery)

This seems to be the new way to stretch out product skus while they partner with board manufacturers to pump out 10 versions of the same card. I said to another guy on here, Im not into playing the game they set up, I will wait. They capitalize on impatient people. Don't give them what they want, which is for you to run out and but the first thing they show you, and then after that they show you the one that has enough ram on it, and you buy that too. So Ill end by stating that the 3060 12GB model, in most situations, is the optimal pick. And if you don't know how to make a 3060 with 12 gigabytes work for you? Then I'm sorry but you are no so good at computer stuff.

My stuff
5600x
16 GB 3600 xmp
x570
wd sn770 1tb
evga 1660 super oc 6gb

I'm fairly confident this card is not being marketed to us..

 

[RegaeRevaeb]

Considering we were stuck with the 1650 for far too long (Ampere A's were eaten by miners), this bodes well for the future when Ada's successor comes in. This is to say we should see this card's price drop to something -- hopefully -- suitable by then/then. It's especially likely, I believe, if the garden variety VRAM floor keeps getting raised for GPUS in general. Of course, I'm crossing my fingers with some doubt nonetheless.

 

[Lew Zealand]

I think the problem with the 75W class of slot-power GPUs is that performance GPUs are getting higher and higher power limits each generation so the 75W line in the sand becomes relatively less adequate with each generation. The A2000 shows that you can still produce it but the larger die size will simply cost more (ignoring the A2000's other costs). Die size increased quite a bit from 1050 Ti to 1650 to A2000 but I think once you exceed a $200 cost for this GPU, it'll just be better getting a power supply and a good $200 GPU like the RX 6600 or whatever it's replaced with.

 

[TheinsanegamerN]

I think the problem with the 75W class of slot-power GPUs is that performance GPUs are getting higher and higher power limits each generation so the 75W line in the sand becomes relatively less adequate with each generation.

That's utterly rediculous. Look at the sheer performance gains per gen, they at times look more impressive then cards like the 4080 that dont have said limits.

The A2000 shows that you can still produce it but the larger die size will simply cost more (ignoring the A2000's other costs). Die size increased quite a bit from 1050 Ti to 1650 to A2000 but I think once you exceed a $200 cost for this GPU, it'll just be better getting a power supply and a good $200 GPU like the RX 6600 or whatever it's replaced with.

The large die is not mandatory. Nvidia made 3060s that pulled only 75 watts. They were used in mobile.

There is no reason said GPUs could not be packaged into desktop cards. Nobody was willing to do it.

 

[Tomgang]

I would loop to to get this gpu. But since this is only about rtx 3060 TI/3070 performance for rtx 4080 price and i all ready have a rtx a2000 and rtx 4090. I can't justify buying this.

Rtx a2000 is around 3050/3060 none TI performance for a msrp of 450 usd. Rtx 4000 ssf is 3060 TI/3070 for triple the price or 1250 usd msrp. Just to put it in to a perspective. Sure you do pay some of it with 20 GB vram. But still, it's a very expensive card for the performance you get regardless of it low power consumption.

Maybe sometime in the future when it is used a price has come down to a acceptable level I might jump on it. But for now, it's to expensive for the performance you get just for gaming at least.

If you are after a low power SFF slot powered gpu. I would rather recommend a2000. Much more reasonable price now it is used amd for sale. Gaming performance aren't that bad either. It handle dead island 2 just fine example.

 

[Vayra86]

We can start a series

"Ada, what could have been"
after
"Ampere, what could have been"
and...
"RDNA3... what could have been"

Man...

 

[TheinsanegamerN]

I would loop to to get this gpu. But since this is only about rtx 3060 TI/3070 performance for rtx 4080 price and i all ready have a rtx a2000 and rtx 4090. I can't justify buying this.

Rtx a2000 is around 3050/3060 none TI performance for a msrp of 450 usd. Rtx 4000 ssf is 3060 TI/3070 for triple the price or 1250 usd msrp. Just to put it in to a perspective.

for perspective: that MSRP was a pure fantasy. Multiple tech sites claimed it would be a $450 MSRP, but nvidia's own press release makes no mention of the sort:

NVIDIA Makes RTX Technology Accessible to More Professionals

NVIDIA RTX A2000 is our most compact, power-efficient GPU for a wide range of standard and small-form-factor workstations.

blogs.nvidia.com

The ACTUAL launch price of the A2000 was $799. $999 for the 12GB version. After about 18 months the prices came way down to $250-300 on ebay for one. The launch was also only a few months into the hyperinflationary spiral we are in today, that is going to play a major factor.

Sure you do pay some of it with 20 GB vram. But still, it's a very expensive card for the performance you get regardless of it low power consumption.

Well yeah, going from 6GB to 20GB of ECC VRAM is going to cost you a huge penny. And you are paying for pro tier drivers, not consumer drivers. Those dont come cheap, and they are factored into the price of the card itself.

Maybe sometime in the future when it is used a price has come down to a acceptable level I might jump on it. But for now, it's to expensive for the performance you get just for gaming at least.

This has always been, and always will be, the trend with pro cards for home users. Same thing applied to quaddro for over a decade.

If you are after a low power SFF slot powered gpu. I would rather recommend a2000. Much more reasonable price now it is used amd for sale. Gaming performance aren't that bad either. It handle dead island 2 just fine example.

The A2000 is good, and if you have one, upgrading to the A4000 will never make sense. The next gen up never does.

 

[Lew Zealand]

That's utterly rediculous. Look at the sheer performance gains per gen, they at times look more impressive then cards like the 4080 that dont have said limits.

Nice attitude, you missed the point. If your top GPU goes from 250 to 450W then your 75W GPU, while it takes advantage of increased new-gen efficiency, is further behind from the top GPU in the gen. And thus more ignored by the public. Nvidia skips 75W GPUs pretty frequently and likely from this lack of relative performance/demand. Why build when nobody will buy?

The large die is not mandatory. Nvidia made 3060s that pulled only 75 watts. They were used in mobile.

There is no reason said GPUs could not be packaged into desktop cards. Nobody was willing to do it.

That's the point. When high Wattage is part of high performance, then 75W performance is too low for the cost to build and nobody's gonna buy.

 

[TheinsanegamerN]

Nice attitude, you missed the point. If your top GPU goes from 250 to 450W then your 75W GPU, while it takes advantage of increased new-gen efficiency, is further behind from the top GPU in the gen. And thus more ignored by the public. Nvidia skips 75W GPUs pretty frequently and likely from this lack of relative performance/demand. Why build when nobody will buy?



That's the point. When high Wattage is part of high performance, then 75W performance is too low for the cost to build and nobody's gonna buy.

so you think because the RTX 4090 pulls 450 watt that the sub 100w market just doesnt think new GPUs are good anymore?

Have you ever heard of correlation does not equal causation? someone who wants a slot powered GPU does not, in any way, care about the 4090. All they care about is how well the new cards in the appropriate TDP range perform compared to their predecessor.

What ratio, in your mind, produces proper results? At what ratio does the low end fall off? ?There has to be something backing this idea up, right?

 

[Lew Zealand]

so you think because the RTX 4090 pulls 450 watt that the sub 100w market just doesnt think new GPUs are good anymore?

Have you ever heard of correlation does not equal causation? someone who wants a slot powered GPU does not, in any way, care about the 4090. All they care about is how well the new cards in the appropriate TDP range perform compared to their predecessor.

What ratio, in your mind, produces proper results? At what ratio does the low end fall off? ?There has to be something backing this idea up, right?

I claimed no correlation, but keep asking those asking empty questions.

If there's such a big market for 75W GPUs then where is the 30 series one? With such demand, Nvidia should surely have made one.

My assumption is that performance is too low relative to the 3050 (already not known as a good value which is bad at the low end) to make one worth selling for the price to make it. Ie.: getting the power down to 75W while maintaining enough performance and margins to make a viable product is a problem. Not a technical, just marketing. Seeing the sales of the 1650 vs the 1650 Super could have answered that question for Nvidia but I don't know those numbers. $10 more for 33% more performance seems worth considering a power supply replacement for.

Or it's just that 75W gamers are easy to satisfy with older lower-end parts. If the 1650 is selling enough to make it the top card on the Steam survey, then why make a new one?

The large die is not mandatory. Nvidia made 3060s that pulled only 75 watts. They were used in mobile.

Forgot this: That's a large die for a low-power and entry-level GPU, larger than the 3050 and the same one that the 3060 uses. There's no way Nvidia would sell a 3060 die GPU for less than a 3050. The 3040 or whatever needs to exist cut from the 3050 die with 2048 Cores, maybe slower GDDR6 and 75W. Maybe once all the 4000 series releases we'll see something like that but it should have been made already. Again I assume this is a profit thing, there's not enough demand vs the 3050 to satisfy someone in marketing.

 

[tvshacker]

Does anyone have Steve's number (from HU), we "need" a new video comparison:
RTX A4000 vs RTX 4000 ADA

This one shows that RTX A4000 was a bit slower than the 3070 when there was no "overflow" of VRAM. Now with lower memory bandwidth could be a valid preview of what to expect from the 4060TI 16GB behaviour.

 

[Metroid]

Lets hope the price is the same if not then boycott it.

 

[thegnome]

We can start a series

"Ada, what could have been"
after
"Ampere, what could have been"
and...
"RDNA3... what could have been"

Man...

All great, aside from RDNA3 being worse than expected. The problem mainly lies (as always) with bad vram descisions, bad pricing and generally bad sku tiers. All easily solved, but money rulezzz

 

[mechtech]

It would be nice if there was more choice in the SFF space................I know several people that bought refurb mini PCs from dell, hp, etc. that are all low profile cases................best they can do for the money is the RX6400

 

[wolf]

Really loving the little RTX A2000 in an Ultra SFF I built, and will almost certainly pick one of these up at some stage.

 

[LupintheIII]

A bit disappointed with the performance to be honest, the 160bit-bus is probably suffocating the chip.
I've already made more than one build with the RTX A2000, I'm waiting for a waterblock right now, hopefully it will be compatibile with the RTX 4000 SFF (externally they look the same), but I doubt it.
Looking forward to grab one of those as soon as price come down to not-insane territory.

 

[cvaldes]

Here we are again with what seems to be a crippled 4000 series card. How do they do it? Easy, just hobble that memory so it can't go anywhere (280gb/s) but give it plenty of space to move around the room. 20GB Vram (Misery)

This seems to be the new way to stretch out product skus while they partner with board manufacturers to pump out 10 versions of the same card. I said to another guy on here, Im not into playing the game they set up, I will wait. They capitalize on impatient people.

killowatt is correct.

There are graphics card usage cases that aren't gaming.

If it doesn't say GeForce, it's probably not intended for consumers. Consumer usage cases are only a subset of the entirety of applications for GPUs.

The performance-per-dollar metric for this SKU isn't aligned for the consumer gaming marketplace. The fact that you find the device unappealing as a PC gaming component is deliberate. After all, it would be stupid for a manufacturer to market a pro card that's identical in specs and performance to an AIB's consumer product at three times the price.

 

[Pooch]

Nothing of what I said was incorrect. They are doing the same things to commercial/developers as they are to gaming consumers. They are pulling the same tricks on the industry and perhaps even more so because they know that there is more profit per product in the professional/developer/enterprise arena.

 

[Minus Infinity]

Lets hope the price is the same if not then boycott it.

$1200!

 

[TheinsanegamerN]

I claimed no correlation, but keep asking those asking empty questions.

You're the one claiming that 450watt GPUs make 100w GPUs undesirable, but keep moving those goalposts bud. It'll work eventually.

If there's such a big market for 75W GPUs then where is the 30 series one? With such demand, Nvidia should surely have made one.

They made a sub 100w GPU. The 3050. It just want under 75 w.

My assumption is that performance is too low relative to the 3050 (already not known as a good value which is bad at the low end) to make one worth selling for the price to make it. Ie.: getting the power down to 75W while maintaining enough performance and margins to make a viable product is a problem. Not a technical, just marketing. Seeing the sales of the 1650 vs the 1650 Super could have answered that question for Nvidia but I don't know those numbers. $10 more for 33% more performance seems worth considering a power supply replacement for.

Or it's just that 75W gamers are easy to satisfy with older lower-end parts. If the 1650 is selling enough to make it the top card on the Steam survey, then why make a new one?

Pretty sure gamers would be a lot more satisfied with a proper 75w 4050 then a 5 year old 1650. Not once has the "theyre fine with older hardware" mem actually worked in the face of new hardware.

Forgot this: That's a large die for a low-power and entry-level GPU, larger than the 3050 and the same one that the 3060 uses. There's no way Nvidia would sell a 3060 die GPU for less than a 3050. The 3040 or whatever needs to exist cut from the 3050 die with 2048 Cores, maybe slower GDDR6 and 75W. Maybe once all the 4000 series releases we'll see something like that but it should have been made already. Again I assume this is a profit thing, there's not enough demand vs the 3050 to satisfy someone in marketing.

It's a whopping 76mm larger then the 1650 mobile. That's not that big.

 

[Lew Zealand]

You're the one claiming that 450watt GPUs make 100w GPUs undesirable, but keep moving those goalposts bud. It'll work eventually.

Nvidia is selling their performance at a premium and that currently comes along with high power usage. Nvidia didn't make a desktop GPU even close to 75W in Ampere. My suggestion is that 75W Ampere performance would be too low to sell it at a reasonable price and make a profit. So they stuck with the old, underpowered 1650.

They made a sub 100w GPU. The 3050. It just want under 75 w.

Wrong, 130W in the spec:

NVIDIA GeForce RTX 3050 8 GB Specs

NVIDIA GA106, 1777 MHz, 2560 Cores, 80 TMUs, 32 ROPs, 8192 MB GDDR6, 1750 MHz, 128 bit

www.techpowerup.com

133W Gaming power as tested at TPU:

MSI GeForce RTX 3050 Gaming X Review

The MSI GeForce RTX 3050 Gaming X features a large triple-slot, dual-fan thermal solution that achieves excellent temperatures. Also included is a medium-sized overclock out of the box and the highly popular idle-fan-stop capability.

www.techpowerup.com

Pretty sure gamers would be a lot more satisfied with a proper 75w 4050 then a 5 year old 1650. Not once has the "theyre fine with older hardware" mem actually worked in the face of new hardware.

Then where is the new hardware? Why isn't it being made?

It's a whopping 76mm larger then the 1650 mobile. That's not that big.

A larger die on a newer, more expensive node = $$. Same can happen with a possible 75W Ada GPU though the node shrink should be more significant this time but also more expensive again. If the 1650 sells for $155+ on a cheap node, what will a 75W Ada sell for? I hope they make one but the price may be prohibitively expensive for upgrading slot powered computers looking for a cheap GPU.