Insider Claims NVIDIA GeForce RTX 50 Series Transitioning to Usage of SK hynix GDDR7 Memory Modules

[T0@st]

So far, NVIDIA's GeForce RTX 50xx graphics card models have shipped with Samsung GDDR7 memory modules onboard. According to a fresh MEGAsizeGPU (aka @Zed__Wang) claim, a change in vendor has already occurred. The tenured tracker of Team Green inside track information believes that the company has: "started to use SK hynix GDDR7 for the GeForce RTX 50 series graphics cards. Started with GeForce RTX 5070 first." Officially, NVIDIA's latest board designs can support GDDR7 modules produced by the "big three:" Samsung, SK hynix and Micron (see BIOS info below). Team Green's comfortable market leading position probably grants plenty of negotiation power to pick and choose the best component deals. Day one evaluators performed teardowns on GeForce RTX 50 series review samples; TechPowerUp's W1zzard found Samsung "K4VAF325ZC-SC32" GDDR7 units—rated for 32 Gbps—onboard various GeForce RTX 5080 16 GB models. As outlined by VideoCardz, the rest of NVIDIA's "Blackwell" gaming product stack sticks with 28 Gbps-rated Samsung GDDR7 modules, extending to its Mobile portfolio.



View at TechPowerUp Main Site | Source

 

[praescepter]

The big question : Will they clock the same? These Samsung chips overclock pretty well on these RTX50 cards.

 

[Legacy-ZA]

The big question : Will they clock the same? These Samsung chips overclock pretty well on these RTX50 cards.

Yes, this is why I am glad I snagged a early GPU.

nGreedia usually does this after a launch, they get them at a better price (for them), but their quality is lacking, they can never clock as high or last as long.

 

[bonehead123]

Yes, this is why I am glad I snagged a early GPU.

nGreedia usually does this after a launch, they get them at a better price (for them), but their quality is lacking, they can never clock as high or last as long.

But of course they do, this IS nGreediya we're talking about here after all, and they are ALL about achieving higher profits at the expense of the lowly end users, which they have proven over & over again that they don't give a rat's ass about..

I'm not opposed to component changes after the launch per se, as long as the mfgr is upfront & forthcoming about it (leaving ME with the choice of either buying the item or something else), which again, nGreediya never does & will never do, for the same reasons stated above..

 

[Calmmo]

Isn't this pretty common? One is going to be the better choice for OCing, 0 impact the vast majority of buyers/users.
Wonder how much (if at all) memory has been the limiting factor for 50 series availability.

 

[Wirko]

Why would Hynix keep making GDDR and stay poor when they can make HBM, which is better than printing money?

 

[wolf]

nGreedia

nGreediya

So original

If these achieve the same clocks, long term, I don't see an issue out of it, as the products will be identical. If they're worse somehow, then it's problematic.

 

[hsew]

I think they should transition RTX 50 from broken to fixed, first.

 

[Quicks]

Shame samsung just can't catch a break.

 

[nguyen]

SK hynix produce better DDR5 than samsung one, no idea if this translate to better GDDR7 module.

Samsung probably has yield issues too, so it's better for Nvidia to source GDDR7 from different vendors to increase Blackwell production rate.

 

[_roman_]

I agree with the point that component changes should be communicated before purchase. It's part the shop fault and part the manufacturer fault.

I think about m2- SSD- NVME and Storage where component swaps are worse.

-- I do not agree with ngreedia as sourcing components is a very important task in electronics. I assume the lowest price will be the most likely the most important factor. I comment quite often about poor build quality or poor material quality.

Note: Do not buy AMD graphic cards - they are the "RED DEVIL" and evil and bad and fancy and ... ... you know the rest.

 

[Legacy-ZA]

So original

If these achieve the same clocks, long term, I don't see an issue out of it, as the products will be identical. If they're worse somehow, then it's problematic.

Isn't it one of the "selling points" that it can overclock so well? Don't they and "influencers" use this OC headroom to say, "make your 5080 into a 4090 (which it still doesn't reach in performance), same for their 5070Ti into a 5080?

And yeah, their brand name for me is nGreedia, branded them like this, mmm, ages ago anyways, doesn't matter, it's been 5 years or more, I even said, that is what I will call them from then on (you can search for the post if you wish)

 

[wolf]

Isn't it one of the "selling points" that it can overclock so well?

The core is where 9/10ths of that performance comes from, but even then, no not really, you get the performance you get out of the box, Nvidia themselves isn't promoting any overclocking ability of the 50 series.

And yeah, their brand name for me is

I find it childish, and that it severely diminishes any objective point anyone is trying to make if they can't even avoid name-calling. I won't harp on about it, but I do call it out when I see it.

 

[Tomorrow]

The big question : Will they clock the same? These Samsung chips overclock pretty well on these RTX50 cards.

Depends on what they're even rated at. SK Hynix is pretty good at memory manufacturing currently (HBM3e, HBM4 and DDR5).
They showed 40Gbps speeds a while back. Also 3GB modules. Laptops may use the 3GB modules but i doubt we'll see 40Gbps before the mandatory Super refresh.

 

[Chomiq]

What's this memory OC people speak of:

Unfortunately NVIDIA is limiting the maximum overclocking for the GDDR7 memory chips to +375 MHz—usually NVIDIA doesn't have any OC limits.

AFAIK this is true for any GDDR7 equipped 50 series card, unless they changed something since reviews dropped.

 

[Legacy-ZA]

I find it childish, and that it severely diminishes any objective point anyone is trying to make if they can't even avoid name-calling. I won't harp on about it, but I do call it out when I see it.

Then, feel free to ignore me and my posts, but it will hopefully get the point across to new tech users. To me it's simple, if you don't want to be branded as something, then you probably should avoid that little problem people associate with you, don't you think? I think it's more childish to expect others to believe a fairy tale and a lie when the evidence is contrary.

The core is where 9/10ths of that performance comes from, but even then, no not really, you get the performance you get out of the box, Nvidia themselves isn't promoting any overclocking ability of the 50 series.

nGreedia not promoting it eh? Have you tried their G-Assistant? lol, "You seem to be GPU limited, here are blah blah methods blah blah" one includes overclocking. lol (granted, for all GPUs, not just the 5000 series, but more likely, as you need 12GB VRAM to use it)

Here is the thing, nGreedia might not have promoted it as such, but their "influencers" sure as hell did, you know the ones, probably, not gonna mention "names"

 

[nguyen]

Then, feel free to ignore me and my posts, but it will hopefully get the point across to new tech users. To me it's simple, if you don't want to be branded as something, then you probably should avoid that little problem people associate with you, don't you think? I think it's more childish to expect others to believe a fairy tale and a lie when the evidence is contrary.

nGreedia not promoting it eh? Have you tried their G-Assistant? lol, "You seem to be GPU limited, here are blah blah methods blah blah" one includes overclocking. lol (granted, for all GPUs, not just the 5000 series, but more likely, as you need 12GB VRAM to use it)

Here is the thing, nGreedia might not have promoted it as such, but their "influencers" sure as hell did, you know the ones, probably, not gonna mention "names"

Maybe, just maybe that Nvidia being greedy allow you to enjoy their products for decades and not fade away like 3dfx, Matrox, etc... .
Btw if you can afford a 5070Ti, I bet you are also greedy, as in earning more than minimum wages to survive eh, and i'm also greedy too .

If Nvidia were a little less greedy, they would be a monopoly already

 

[Legacy-ZA]

Maybe, just maybe that Nvidia being greedy allow you to enjoy their products for decades and not fade away like 3dfx, Matrox, etc... .
Btw if you can afford a 5070Ti, I bet you are also greedy, as in earning more than minimum wages to survive eh, and i'm also greedy too .

If Nvidia were a little less greedy, they would be a monopoly already

Nah, not greedy, don't just automatically assume, but I can see why you would draw such a conclusion and call me a hypocrite, I can't fault you for that. However, there was more to the story; You see, I am a realist, I had to buy it.

• My GTX1650 was only an interim card for my previously crippled 8GB VRAM 3070Ti, if it had 16GB VRAM, I would have kept it. I had planned on getting a new GPU 6 months prior to the 5000 series launch, I saved a little more, as I had crypto/covid vibes, as BTC was mooning.
• I knew South-Africa will be heavily sanctioned because of their treatment of the white minority, our Rand was already in the ****ter and was going to get worse, which it did and is still getting worse day by day. I dodged that bullet, this isn't even with the AIB's and nGreedias shenanigans, that is just extra icing on the cake.
• The markets were going to go apeshit with all the tariffs, which I knew would happen, it was inevitable.

I skipped all that noise by being quick, I got my GPU at MSRP though, or I wouldn't have bought it, so there is that, unfortunately I still had to pay the ASUS tax, but w/e. I originally wanted a Zotac, but they decided to go ape**** 1 hour after the launch and pumped those prices right-up and retailers happily adjusted. I knew about a computer store I used as a child and never grew as big as most and they had 1x ASUS card, which I snagged. They didn't have a 5080 which I wanted, but alas, probably for the best, as I knew I would be CPU limited with the 5070Ti, but I am waiting for the 10800X3D before upgrading that part of my system, then there is the fact that the 5080 draws near 400W on these ridiculous 16-pin, which at present, is not such a good idea, my 5070Ti only pulls 180W to 260W currently (Undervolted)


P.S
I miss your kittty avatar. :*(

 

[wolf]

Then, feel free to ignore me and my posts, but it will hopefully get the point across to new tech users. To me it's simple, if you don't want to be branded as something, then you probably should avoid that little problem people associate with you, don't you think? I think it's more childish to expect others to believe a fairy tale and a lie when the evidence is contrary.

Nah I won't ignore you, I'd miss all this and not get my chance to counterpoint, and there might be interesting discussion in there, albeit weakened by the name calling. I'd like to think new (and existing) users are also mature enough to see name calling for what it is - a detractor to any point you're trying to make rather than something that strengthens it.

Here is the thing, nGreedia might not have promoted it as such, but their "influencers" sure as hell did, you know the ones, probably, not gonna mention "names"

Thank you for agreeing with my point. And as far as anyone not Nvidia, I pay attention to respected tech press that do thorough reviews, and while they mention it, Nvidia still doesn't and a difference in memory suppliers doesn't change a thing - except hopefully more supply of cards overall as they may be restricted by GDDR7 availability at the moment. More supply is a good thing, as it will drive down prices.

usually does this after a launch, they get them at a better price (for them), but their quality is lacking, they can never clock as high or last as long.

Now, looking past the silly name calling, this statement suggests it'll be easy for you to show some receipts for how Nvidia "usually does this after a launch, but their quality is lacking, they can never clock as high or last as long".

I'd be interested in your evidence for the claim that it 1. usually happens, and when it does it always 2. lacks quality/longevity/oc headroom.

Or is it a baseless/exaggerated/insufficient statistical data kind of claim? I act on the best information I have at the time, if you can prove the statement is true, I'll happily agree with you on it.

 

[Wirko]

I agree with the point that component changes should be communicated before purchase. It's part the shop fault and part the manufacturer fault.

I think about m2- SSD- NVME and Storage where component swaps are worse.

SSD makers are careful about what to put and what to not put in the specs. The latest one reviewed, for example, says TLC, but number of channels is "up to 8". Graphics card and DDR module makers do the same.

I do not agree with ngreedia as sourcing components is a very important task in electronics. I assume the lowest price will be the most likely the most important factor. I comment quite often about poor build quality or poor material quality.

Price, sure, but acceptable terms of delivery (quantity, delivery time, flexibility if needed), sure are very important too. Also, quantity discounts - if you always buy from a single supplier, they can offer chips for 1.75% less.

 

[Legacy-ZA]

Nah I won't ignore you, I'd miss all this and not get my chance to counterpoint, and there might be interesting discussion in there, albeit weakened by the name calling. I'd like to think new (and existing) users are also mature enough to see name calling for what it is - a detractor to any point you're trying to make rather than something that strengthens it.

I won't stop calling them that though, get used to it.

Thank you for agreeing with my point. And as far as anyone not Nvidia, I pay attention to respected tech press that do thorough reviews, and while they mention it, Nvidia still doesn't and a difference in memory suppliers doesn't change a thing - except hopefully more supply of cards overall as they may be restricted by GDDR7 availability at the moment. More supply is a good thing, as it will drive down prices.

Yes.

However, I want to add; let's not pretend that these "influencers" are not influenced by nGreedia, I am pretty sure calls were made, especially since said influencers made follow up videos bringing the OC headroom to the forefront when people were extremely disappointed with the performance uplift compared to the previous generation, and they heavily stressed that you can OC it one tier up, when that simply isn't the case, it's closer, but not the case.

The average user doesn't OC, they buy a card and they expect next generation uplift in performance. nGreedia most likely influenced in this manner, otherwards the backlash would have been even greater, and they sure didn't want to un-launch another xx80 card and slot it lower, but it sure as heck deserves the same treatment.

Now, looking past the silly name calling, this statement suggests it'll be easy for you to show some receipts for how Nvidia "usually does this after a launch, but their quality is lacking, they can never clock as high or last as long".

I'd be interested in your evidence for the claim that it 1. usually happens, and when it does it always 2. lacks quality/longevity/oc headroom.

Or is it a baseless/exaggerated/insufficient statistical data kind of claim? I act on the best information I have at the time, if you can prove the statement is true, I'll happily agree with you on it.

Not just a personal experience, but this happened to me with my GTX1070 and with many others, they switched from Samsung to Micron, you couldn't nearly overclock as well and even if you did manage a good stable OC or not OC at all, your memory would fail much faster, mine did. It was well known that the Micron modules were sub par/ran more hot, there are articles about it. Many AIB's designed their cooling solution around the Samsung modules, those didn't need active cooling, where the Micron modules did. Anyways, scummy things like that, I and many others are trying to avoid a repeat. At the time EVGA didn't even bother to put thermal pads on the modules, oh boy, was that a screwup, thankfully, they still covered the self-repair-application of pads sent to their users, even if you caused accidental damage in the application attempt.

Anyways, people fear a repeat, this is why I check many things on W1zzards reviews before buying certain models, if the card is not on the TPU for review, I do not buy them.


This is me, concerning nGreedia:

 

[x4it3n]

Depends on what they're even rated at. SK Hynix is pretty good at memory manufacturing currently (HBM3e, HBM4 and DDR5).
They showed 40Gbps speeds a while back. Also 3GB modules. Laptops may use the 3GB modules but i doubt we'll see 40Gbps before the mandatory Super refresh.

40Gbps will not happen this generation (RTX 50s), RTX 60s are a lot more likely to have some though.

What's this memory OC people speak of:

AFAIK this is true for any GDDR7 equipped 50 series card, unless they changed something since reviews dropped.

It's locked on Nvidia App but not the other apps if I remember well. Also when we look at the 5090 FE review from @W1zzard we can see that the Memory temps are already at 94°C so I wouldn't push them too much unless you wanna put a Liquid Cooler on it lol.

Not just a personal experience, but this happened to me with my GTX1070 and with many others, they switched from Samsung to Micron, you couldn't nearly overclock as well and even if you did manage a good stable OC or not OC at all, your memory would fail much faster, mine did. It was well known that the Micron modules were sub par/ran more hot, there are articles about it. Many AIB's designed their cooling solution around the Samsung modules, those didn't need active cooling, where the Micron modules did. Anyways, scummy things like that, I and many others are trying to avoid a repeat. At the time EVGA didn't even bother to put thermal pads on the modules, oh boy, was that a screwup, thankfully, they still covered the self-repair-application of pads sent to their users, even if you caused accidental damage in the application attempt.

This whole generation is a joke anyway so I don't have much hope anymore lol. Maybe RTX 60s on TSMC 3nm will be a better upgrade...

 

[Visible Noise]

So original

If these achieve the same clocks, long term, I don't see an issue out of it, as the products will be identical. If they're worse somehow, then it's problematic.

Trolls aren’t really known to be intelligent enough to come up with original quips. Usually they see something cute, and repeat it ad-nauseam until it just becomes boorish, and not even realizing it.

They are the perfect candidates for feeding your ignore list.

 

[wolf]

I won't stop calling them that though, get used to it.

I'd recommend looking into the etymology of name calling, how it affects debate, and think about your position on it long term. I won't hold my breath of course, but I hope so for your own sake.

Yes.

However, I want to add; let's not pretend that these "influencers" are not influenced by nGrEeDiA, I am pretty sure calls were made, especially since said influencers made follow up videos bringing the OC headroom to the forefront when people were extremely disappointed with the performance uplift compared to the previous generation, and they heavily stressed that you can OC it one tier up, when that simply isn't the case, it's closer, but not the case.

The average user doesn't OC, they buy a card and they expect next generation uplift in performance. nGrEeDiA most likely influenced in this manner, otherwards the backlash would have been even greater, and they sure didn't want to un-launch another xx80 card and slot it lower, but it sure as heck deserves the same treatment.

I'll take your assumptions made here with a pinch of salt. You being pretty sure, sure isn't good enough for me.

Not just a personal experience, but this happened to me with my GTX1070 and with many others, they switched from Samsung to Micron, you couldn't nearly overclock as well and even if you did manage a good stable OC or not OC at all, your memory would fail much faster, mine did. It was well known that the Micron modules were sub par/ran more hot, there are articles about it. Many AIB's designed their cooling solution around the Samsung modules, those didn't need active cooling, where the Micron modules did. Anyways, scummy things like that, I and many others are trying to avoid a repeat. At the time EVGA didn't even bother to put thermal pads on the modules, oh boy, was that a screwup, thankfully, they still covered the self-repair-application of pads sent to their users, even if you caused accidental damage in the application attempt.

Sounds like it's isolated to a single card you can name, and an AIB partner mistake when cooling memory, hardly a case of Nvidia usually doing this after a launch with the same results you experienced. This makes your original statement asserting that as if it were some sort of overarching truth misleading.

Anyways, people fear a repeat

Do they? or do you? Looking through comments on various other sites and platforms, I am not seeing your sentiment echoed.

They are the perfect candidates for feeding your ignore list.

I disagree, while some have made the ignore list, I prefer to call this kind of sillyness out when I see it so everyone else can see how absurd it is. What tends to happen is they put me on their ignore list which is honestly the best case scenario, because then my rebuttal stands with no counter, not that I've yet seen a counter to name calling be rooted in sound reasoning mind you.

 

[Tomorrow]

40Gbps will not happen this generation (RTX 50s), RTX 60s are a lot more likely to have some though.

I would not be so sure. Current 30Gbps on RTX 5080 can be OC'ed to 36Gbps (2250*16 up from 1875*16).
It's actually rated at 32Gbps, but runs at 30Gbps. I imagine if Nvidia removes the +375 limit (unlikely) or it's bypassed (we can hope) it would be possible to already get very close to 40Gbps.