Shadowizedits 3 x 3gb modules, instead of 4 x 2gb modules. Allowing them to deal with the shortages better. Surely you can understand why less silicon for the same or increased capacity is a good thing.

I can understand why it's a good thing, but that doesn't shake the feeling that 9 is just weird. Also, even though it's less silicon, NV isn't gonna pass up a chance to charge customers even more. Most people don't understand that it's less silicon, so NVIDIA is totally gonna be able to get away with a small upcharge.

YouseethisThe 60 class having more memory than the 80 class and tying the 80ti was funnier :roll:

Agreed lol.

RejZoRWell, we had Radeon HD7950 with 3GB VRAM, RTX 1080Ti with 11GB

True, but I didn't give a crap about computers then, so 9 still feels strange. *I would like to specify that I'm not saying it is strange, more VRAM is more VRAM, it simply doesn't look right to me after having worked with almost entirely even numbers for a while.*

RejZoRWell, we had Radeon HD7950 with 3GB VRAM, RTX 1080Ti with 11GB and we also had RTX 3080 with 10GB which is also a rather weird number... it's just very odd choice for a weak ass GPU like RTX 5050 to 1GB up compared to RTX 5060 8GB for some reason. Like, why does this product even exists?

It was a choice between 9 GB or 6 GB.

I question this bit:

Nomad76The report also claims NVIDIA is preparing adjustments higher up the stack. A revised GeForce RTX 5060 may transition to the larger GB205 die, currently used in the RTX 5070, in a cut-down configuration. That would reportedly reduce the GPU from 6144 CUDA cores and a 192-bit bus to 3840 cores and a 128-bit interface. Such a move would require board partners to adopt new PCBs compatible with GB205 and standard 8-pin power connectors. According to the leak, these changes are tied to supply constraints affecting lower-tier RTX 50-series GPUs, including the RTX 5050 and RTX 5060. The GeForce RTX 5060 12 GB, which appeared in earlier roadmaps, is now reportedly canceled.

A reduction in available resources despite moving up to a much larger core? Unheard of, even the TU104 2060 that EVGA cooked retained the same core configuration of any other TU106 2060. Not to mention this would reduce the GPU's performance by nearly 80%. There is no way they can pull this off and retain the same performance level unless they move 3 architectural generations ahead, so that bit can be written off as rumor/hearsay/I made it tf up.

YouseethisWhy would they do this? Are the 3x 3gb gddr7 chips cheaper that 4x 2gb gddr6? Is the ga207 with only 96bit working, that common? Is salvaged 96bit ga207+9gb gddr7 cheaper than the current 5050?
Perhaps the pricing of this thing might be close to a 5060, but the current one is already close. So 290$ msrp?

It is bewildering.

Probably, and with a spec bump to boot. But this would've zero negatives for consumers, as this GDDR7 configuration would be superior in both capacity and performance overall, provided no additional ROP clusters are disabled.

RejZoRWell, we had Radeon HD7950 with 3GB VRAM, RTX 1080Ti with 11GB and we also had RTX 3080 with 10GB which is also a rather weird number... it's just very odd choice for a weak ass GPU like RTX 5050 to 1GB up compared to RTX 5060 8GB for some reason. Like, why does this product even exists?

HD 7950/7970 have a 384 bit bus, so 3 GB isn't exactly bewildering if you look at the memory chips of the time. Reference design uses H5GQ2H24MFR-T2C, which is 2 Gbit (256 MB per chip), install 12 of them on a board and voilà, 3 GB

GTX 1080 Ti is just as easily explained, 352-bit interface. Take the exact same chip that goes into a Titan X Pascal, disable one memory controller, and you have the 1080 Ti. The performance is largely identical, and in games of the time, it was even marginally faster as Nvidia used slightly faster GDDR5X chips to compensate for the loss in bandwidth, most games didn't scale to hardware that powerful yet.

In the case of the RTX 3080, in addition to the shader clusters, ROPs and other areas of the GPU, they disabled two memory controllers and installed 10 8 Gbit chips to achieve its 10 GB memory capacity over a 320-bit interface.

You may have noticed that the standard GDDR memory controller bus width is 32 bit and has been for some time. ;)

FatloaderI can understand why it's a good thing, but that doesn't shake the feeling that 9 is just weird. Also, even though it's less silicon, NV isn't gonna pass up a chance to charge customers even more. Most people don't understand that it's less silicon, so NVIDIA is totally gonna be able to get away with a small upcharge.

Would you rather they use three 2 GB chips so you'd land with a 6 GB model? It's the same reason why the RTX 3060 had 12 GB and the 3060 Ti 8. Sometimes, it's not worth it to go so low. The 24 Gbit ICs solve that problem.

Anyway, I don't see why people are crazy about the RAM this GPU has. 9 or 8 GB is enough for the people that this product targets.

It's more than a little surprising that the GB207 can use GDDR7.

Dr. DroHD 7950/7970 have a 384 bit bus, so 3 GB isn't exactly bewildering if you look at the memory chips of the time. Reference design uses H5GQ2H24MFR-T2C, which is 2 Gbit (256 MB per chip), install 12 of them on a board and voilà, 3 GB

GTX 1080 Ti is just as easily explained, 352-bit interface. Take the exact same chip that goes into a Titan X Pascal, disable one memory controller, and you have the 1080 Ti. The performance is largely identical, and in games of the time, it was even marginally faster as Nvidia used slightly faster GDDR5X chips to compensate for the loss in bandwidth, most games didn't scale to hardware that powerful yet.

In the case of the RTX 3080, in addition to the shader clusters, ROPs and other areas of the GPU, they disabled two memory controllers and installed 10 8 Gbit chips to achieve its 10 GB memory capacity over a 320-bit interface.

It's all binary after all.
9 = 1001
10 = 1010
11 = 1011
I'm more bothered by the fact that memory speeds are still multiples of 33.333333333333 sometimes. Even LPDDR is at 7467 or 10667 Mbit/s. A legacy of the 80386-33, I guess.

Getting gtx970 4gb flashbacks :(

HyderzGetting gtx970 4gb flashbacks :(

Completely impossible to occur in this scenario, and the generational uplift from GDDR7 easily overcomes the bus width reduction. This is just assembling a card with three, much better performing, 24 Gbit chips. Fourth slot present, but vacant, with the associated memory controller physically disabled. It'd be an otherwise straight reduction, wasn't for the not-insignificant IC upgrade.

WirkoIt's all binary after all.
9 = 1001
10 = 1010
11 = 1011
I'm more bothered by the fact that memory speeds are still multiples of 33.333333333333 sometimes. Even LPDDR is at 7467 or 10667 Mbit/s. A legacy of the 80386-33, I guess.

I always thought it was to avoid having "the number of the beast" in spec sheets. Haha. Like, DDR-667 because it's actually 666.6 ;)

mrnagantThis would be really nice. But that would require a 60W TDP decrease. You are asking for reduced CUDA cores and like half the clock speed.
Would be cool though if they launched the MX line on desktop to offer a card to keep within the power budget of PCIe.

I already tested the equivalent, limiting the Laptop 4060 to 70W still makes it ~75% faster than the current best 70W slot powered consumer GPU, the RTX 3050 6GB. This is an excellent 70W GPU for recently decommissioned office PCs and the 5050 GDDR7 would perform about the same. I'd buy one in an instant.

RejZoRWell, we had Radeon HD7950 with 3GB VRAM, RTX 1080Ti with 11GB and we also had RTX 3080 with 10GB which is also a rather weird number... it's just very odd choice for a weak ass GPU like RTX 5050 to 1GB up compared to RTX 5060 8GB for some reason. Like, why does this product even exists?

I suppose for folks like myself that are still rocking at GTX 1660 super 6gb. Though I'm sure there is better value in the used flagship market compared to this 5050 refresh.

RejZoRWell, we had Radeon HD7950 with 3GB VRAM, RTX 1080Ti with 11GB and we also had RTX 3080 with 10GB which is also a rather weird number... it's just very odd choice for a weak ass GPU like RTX 5050 to 1GB up compared to RTX 5060 8GB for some reason. Like, why does this product even exists?

Don't forget the awkward GTX 570 with 1.25GB (well, technically 1280MB) and 2.5GB (2560MB) models.

neatfeatguyDon't forget the awkward GTX 570 with 1.25GB (well, technically 1280MB) and 2.5GB (2560MB) models.

"1.25 GB" is perfectly correct, commonly used and unambiguous. We just need to incorrectly assume, just like Intel/AMD/Nvidia and all RAM manufacturers do, that G and M are binary units.

I just don't get why... a bump to 12G on 24Gb modules, sure, whatever, makes sense. 3060 12GB all over again. 9? A singular gigabyte up? What purpose would that serve other than to prevent a total dropoff when something edges too closely to an 8GB limit? Is that the entire reason for this card hitting market? To pad memory just enough to allow for DLSS in tandem with game presets targeted at 8GB cards?

yfn_ratchetI just don't get why... a bump to 12G on 24Gb modules, sure, whatever, makes sense. 3060 12GB all over again. 9? A singular gigabyte up? What purpose would that serve other than to prevent a total dropoff when something edges too closely to an 8GB limit? Is that the entire reason for this card hitting market? To pad memory just enough to allow for DLSS in tandem with game presets targeted at 8GB cards?

Obviously, the intention is to keep the product in the same tier. The objective is reducing BoM (bill of materials) without reducing its performance and specification. In this case, given GDDR7's performance advantage and the higher density of the chips, it actually gets a minor spec buff.

I can't be the only guy that actually likes the idea of this 9GB G7 5050? Especially if they can reduce power consumption figures a little bit with the removal of the fourth memory chip.

Dr. DroEspecially if they can reduce power consumption figures a little bit with the removal of the fourth memory chip.

They'd rather up the core clock to fully utilise the 130 W power budget it has had ever since. Impossible to tell in a blind test most likely anyway.

Dr. DroI can't be the only guy that actually likes the idea of this 9GB G7 5050?

I could theoretically unleash my seal of approval but when they sell 4060 Ti (yes, 8 GB but still) for less money on the 2nd hand market I hardly ever see a point in going 5050, no matter 8 gigs or 9.

Macro DeviceI could theoretically unleash my seal of approval but when they sell 4060 Ti (yes, 8 GB but still) for less money on the 2nd hand market I hardly ever see a point in going 5050, no matter 8 gigs or 9.

If you can find a used 8GB 4060 Ti for similar or even a slightly higher price used than either 5050 variant, that's totally a better deal.

Nvidia remain idiots on VRAM.

This isn't a bad refresh, if they don't increase the price. Could it have been better? Yeah, but we've had less impressive refreshes than this before. *cough* RDNA2 refresh

windwhirlIt was a choice between 9 GB or 6 GB.

Yeah, and we know 6GB is definitely not going to cut it anymore. Unless they intent to make a 5030 6GB for grunt work, but I think the days of such cards are over.

So... no supers this year. But a 5050 9 gig. So a 5050 super. Possibly the most useless card ever made. Watch them price it at 350.

windwhirlAnyway, I don't see why people are crazy about the RAM this GPU has. 9 or 8 GB is enough for the people that this product targets.

To some degree yes, but the B580 is around that price with 12GB, and now multi frame generation, so the point of purchase starts to fall apart

This is obviously a modest GPU. As someone who's been gaming on modest GPUs for over 20 years, I find this GPU to be kinda cool, even tho I've been a Radeon guy since the 7500/8500 days in the early aughts.

I bought a 9060XT 8GB almost 9 months ago and still find it amazing for my 1080p needs. Hopefully this 5050 is less than the $259 I paid lol

Lotsa ultracrepidarian elitists here, as usual, still crying about VRAM. Learn your settings smh

RejZoRToo weak to really run anything, given 9GB of VRAM so it looks better than RTX 5060 8GB when it runs out of VRAM. What the F is NVIDIA smoking these days?

Perfect for retro style Indie games.

Would be nice if the low cards and all cards just started off with x16 lanes and forget about x8 being the ceiling

RTX 5050 isn't so bad for entry level 1080p gaming actually, performance is around 3060 12GB from review. Having extra 1GB might help it a bit, though not that much.

Are they optimising against RAM prices

I see your 9 GB and raise lower you 3.5 GB!

Lord Leatherjacket is altering the VRAM, pray he doesn't alter it any further.