Why is nVIDIA always last on the wagon of GPU memory?

[Enderman]

11 minutes ago, stconquest said:

No.  Probably more of an availability issue.  Too much risk for little reward.  HBM2 is coming to Nvidia GPUs soon, why would they adopt the faster/broader RAM?

nvidia GPUs already have HBM2

look at the P100

 

its not on consumer cards because its too expensive and unnecessary

 

maybe we should be saying "why is nvidia always the first on the HBM2 bandwagon and AMD always last"

lol

[stconquest]

1 hour ago, Enderman said:

nvidia GPUs already have HBM2

look at the P100  <<NO 

 

its not on consumer cards because its too expensive and unnecessary

 

maybe we should be saying "why is nvidia always the first on the HBM2 bandwagon and AMD always last"

lol

What I am saying is that if Nvidia did not see a need for HBM(2) in the current marketplace, they would not be designing new cards for it.

[Maxxtraxx]

I'll see if I can add some of my own thoughts to what is being said here.

 

That is: AMD seems to adopt some technologies faster when it comes to memory and also seems to have larger memory buffers compared to Nvidia.

 

I have to say up front, for the last few years I have preferred Nvidia's offerings to AMD's but I am hoping that Polaris is a sales and performance hit that pushes Nvidia's priceing down, also I'm hoping that Vega is a real star.

 

Thought #1: AMD tends to look 1 generation ahead when it comes to VRAM capacity and Nvidia seems to look at the current generation and decide what best fits without going overboard. I can argue for both sides on this:

- AMD's approach gives you potentially longer lifespan with the card due to next gen VRAM size.

- Nvidia's approach to VRam size seems to fit well what the card is capable of, it may be nice to have 8gb VRAM but if the card is not powerful enough to take advantage of such a large frame buffer then you'll just have 4GB of never used VRAM...

- Nvidia has been the first on the most recent Memory standards GDDR5X and HBM2

 

 

Thought #2: AMD is more forward looking on some of the hardware features but tends to be more of a follower of Nvidia when it comes to some other things.

For example: G-sync/Freesync: freesync was a very creative(and better IMO) kneejerk reaction to the appearance of Gsync.

- Geforce Experience came first followed by Crimson, also Nvidia tends to be much more active in GPU driver releases(despite some recent problems)

- Nvidia seems to be really pushing the VR scene forward with great rendering techniques to reduce GPU load, increase image responsiveness(by making adjustments to the image at the last possible moment before outputting to the screen)

- Nvidia has great color compression algorithms that do wonders for lowering needed memory bandwidth

- Nvidia has GameWorks/VRworks to assist developers

- Nvidia seems to be more aware of making the experience more seamless, for example Frame time latency on Nvidia cards tends to be very good, and really put AMD to shame about 6-8 years? ago and even still with AMD's HBM1 Fury cards that seem to have very noticeable frame time latency and pacing issues.

 

Thought #3: Nvidia is more like Apple and AMD is more like Microsoft(from the early windows era) or Linux

Explanation: Nvidia is a more closed ecosystem that also tends to be more expensive but also more polished and a bit more... arrogant/unfriendly/applely sometimes?

AMD is more open source (because the need to be to stay competitive in ways IMO) and also more friendly and an advocate of the common PC user?

 

Just some thoughts of my own. I'm trying not to be a fanboy (I did say I have Nvidia hardware) here but to provide some counter points to what was said.

[Mew]

It probably has to do with yields. Because of AMD's low market share, they're able to get away with using cutting edge technology without running into issues with backorders. 

[Mew]

3 hours ago, Maxxtraxx said:

\

 

Thought #3: Nvidia is more like Apple and AMD is more like Microsoft(from the early windows era) or Linux

Explanation: Nvidia is a more closed ecosystem that also tends to be more expensive but also more polished and a bit more... arrogant/unfriendly/applely sometimes?

AMD is more open source (because the need to be to stay competitive in ways IMO) and also more friendly and an advocate of the common PC user?

 

\\

Replace Microsoft with Google and it will be spot on. 

[Big-G]

5 hours ago, hex4 said:

For example ATi went with DDR4 which did not last long, then they were the first to use GDDR5 with 4870, then nVIDIA went GDDR5 with GTX 400 series.

AMD again go to HBM first, nVIDIA uses GDDR5X but it's still basically GDDR5 just higher clocked.

 

Never made sense since nVIDIA have a huge amount of money vs AMD.

1. HBM1 was only available to AMD and not Nvidia... blame AMD.

2. GDDR5X is an alternative available before HBM2 becomes mainstream. Nvidia released the first production HBM2 (P100/GP100) and GDDR5X (GP104) boards... where's AMD? blame AMD.

[Delicieuxz]

4 hours ago, Maxxtraxx said:

I'll see if I can add some of my own thoughts to what is being said here.

 

That is: AMD seems to adopt some technologies faster when it comes to memory and also seems to have larger memory buffers compared to Nvidia.

 

I have to say up front, for the last few years I have preferred Nvidia's offerings to AMD's but I am hoping that Polaris is a sales and performance hit that pushes Nvidia's priceing down, also I'm hoping that Vega is a real star.

 

Thought #1: AMD tends to look 1 generation ahead when it comes to VRAM capacity and Nvidia seems to look at the current generation and decide what best fits without going overboard. I can argue for both sides on this:

- AMD's approach gives you potentially longer lifespan with the card due to next gen VRAM size.

- Nvidia's approach to VRam size seems to fit well what the card is capable of, it may be nice to have 8gb VRAM but if the card is not powerful enough to take advantage of such a large frame buffer then you'll just have 4GB of never used VRAM...

- Nvidia has been the first on the most recent Memory standards GDDR5X and HBM2

 

 

Thought #2: AMD is more forward looking on some of the hardware features but tends to be more of a follower of Nvidia when it comes to some other things.

For example: G-sync/Freesync: freesync was a very creative(and better IMO) kneejerk reaction to the appearance of Gsync.

- Geforce Experience came first followed by Crimson, also Nvidia tends to be much more active in GPU driver releases(despite some recent problems)

- Nvidia seems to be really pushing the VR scene forward with great rendering techniques to reduce GPU load, increase image responsiveness(by making adjustments to the image at the last possible moment before outputting to the screen)

- Nvidia has great color compression algorithms that do wonders for lowering needed memory bandwidth

- Nvidia has GameWorks/VRworks to assist developers

- Nvidia seems to be more aware of making the experience more seamless, for example Frame time latency on Nvidia cards tends to be very good, and really put AMD to shame about 6-8 years? ago and even still with AMD's HBM1 Fury cards that seem to have very noticeable frame time latency and pacing issues.

 

Thought #3: Nvidia is more like Apple and AMD is more like Microsoft(from the early windows era) or Linux

Explanation: Nvidia is a more closed ecosystem that also tends to be more expensive but also more polished and a bit more... arrogant/unfriendly/applely sometimes?

AMD is more open source (because the need to be to stay competitive in ways IMO) and also more friendly and an advocate of the common PC user?

 

Just some thoughts of my own. I'm trying not to be a fanboy (I did say I have Nvidia hardware) here but to provide some counter points to what was said.

AMD is like Nicola Tesla - innovative, open, in the business of creativity, an underdog. Nvidia is like Thomas Edison - controlling and protective of IP, manipulating of public perceptions to capitalize on inferior tech, brute force methods.

 

I don't think AMD is more open for the sake of staying competitive, and I'd guess that being as open as AMD is has lessened their market leverage, over time - but it has improved the state of technology for everyone, and every person running a home PC today is running a bit of "team red."

 

Innovation isn't pulled out of a hat in response to being under pressure. An ingenious mindset is consistent in its forward progress and prolific creativity - and that mindset often holds the idealism of free and open information in high value.

[Schyken]

On the subject of cost... AMD announced their new Polaris based GPU with HBM memory and it was $199. So, if cost has anything to do with it, it's that NVidia is too greedy to put something new like than into place. They want to farm as much money as they possibly can BEFORE using AMD's innovations to better their own products. Granted, GDDR5X isn't really different from GDDR5, except for the 64-bit/32-bit. It's not new, just improved. There's a difference. However, AMD's Polaris architecture is on a 14nm FinFET process, which will most likely outperform NVidia's 16nm FinFET architecture. This is actually more beneficial to NVidia than it is to AMD. While AMD will compete with NVidia upon the release of Polaris GPUs, this gives NVidia time, not only to create a new, and smaller architecture (most likely on a 14nm or less process), this will also allow them to implement HBM(2) into their cards.

[ApolloX75]

Personally I've always found nVidia to be more of a "how can I make the best performing card while spending/doing as little as possible" style of company. AMD (and ATi before) have always been a bit more proactive with new tech but this often comes at a steep cost. A good example is the HD 2900 XT and it's 512bit RINGBUS, the design was a monster of memory bandwidth but in the end never performed as well as it should have. They also suffer shortages on almost every high profile release.

 

The huge leap of faith that nVidia took that has most stood out for me was the original GeForce 256. Both the SDR and later DDR versions were huge gambles and crippled a lot of competitors at the time. Typically in the past there was a trade-off when choosing between ATi and nVidia in that the former often had (this is debatable honestly, and was usually saved for 3dFX comparisons since they were stuck in the 16-bit color pit) much better visual quality while the latter was almost always much faster. 

 

In the end they've all made some real crap from generation to generation. Be it the MX440, HD 3xxx or a leafblower FX 5800 they all make a comeback afterwards. Except 3dFX, no coming back from those external power bricks.

 

..... damn this makes me feel old.

 

#getoffmylawn

[DXMember]

7 hours ago, Bhav said:

No its not just that, its also 64 bit wide memory access, vs GDDR5 32 bit.

wait.,... now suddenly all the effective clocks and memory bandwidths make sense to me...

thanks

[Carclis]

8 hours ago, Zyndo said:

they are "last" because they know how little difference that makes in the performance of a GPU. AMD needs the new eye-popping or "WOW" technologies to get attention. Nvidia doesn't need that. Nvidia will make better GPU's without all that pizazz simply because it isn't needed. If you notice, AMD likes to pump a disproportionately high amount of VRAM onto its cards compared to Nvidia. So when you look at its specs and if two cards have similar stats, but one has more VRAM, you would be more inclined to go with that one, whereas Nvidia tends to better know how much VRAM they ACTUALLY need for any given card. For example the R9 390. 8 GB of VRAM. thats the kind of scale running into 4k. theres NO WAY a 390 could do 4k with any reasonable amount of efficiency (you could argue crossfire setups, but you should just buy a better card before you start crossifring or SLIing)

lets use a metaphor for how I feel about Nvidia vs AMD GPU's. Lets say you (Nvidia) take your little brother (AMD) out for a mountain hike. The goal is the top of the mountain. You understand this. That's where you're headed. but your little cousin doesn't have the same goal. His goal is to just be higher on this mountain than you are. He will sprint up to a big rock and climb that one and proclaim "HA! look how much higher I am than you!" but you just shrug that off and keep on trekking, after all, your goal is the top of the mountain not the top of the nearest rock. Your little brother has expended so much energy, reaching an irrelevant goal, that he needs to take a rest upon that rock. You catch up to and surpass him in no time at all, and at the end of the day, you climbed the mountain, whilst your brother lags behind because he was distracted by all the irrelevant rocks and slight plateaus that he thought we're the mountains.

To me, it feels like Nvidia goes where they want to go, does what they want to do, because they believe that's the best way to do it... and AMD is always just trying to stay one step ahead of them. You could argue that's simply because Nvidia has the market control right now and you may be (partially) right about that, but if your business model is simply to follow the competition and come out with gimmicks that sound better, but are irrelevant at the moment, it sounds very.... shady. In this comparison, HBM vs DDR5.

You are absolutely right that HBM is better than DDR5; BUT HBM is irrelevant. the performance difference of that innovation in current GPU's is almost nothing. Its just a badge for them to say "HA! look how much higher I am than you!"

That's just my two cents anyway... Not trying to fanboy, I'll buy whichever card is best, but Nvidia's dictation of what is needed vs what is flashy just seems to be better (at least for now)

I think the main differences are that AMD is hardware focused and Nvidia is more software oriented. This is why Nvidia has traditionally been ahead in driver software and optimisation and has such a heavy emphasis on Gameworks technology. AMD on the other hand, always tries to push the hardware innovation and as a result, is usually the first to include it. Much has changed lately with AMD lifting their software game with regards to Vulkan and DX12 and also driver updates as frequently as Nvidia.

 

I think you're looking at HBM the wrong way. HBM does not directly boost performance as the amount of bandwidth it can deliver is not required at this point in time. The main benefit from it is the small amount of space it requires and it's ability to be extremely power efficient at low frequencies while still remaining faster than GDDR5. These power savings indirectly result in a performance increase because it allows the GPU core itself to consume more power via more cores or higher clock speeds.

[KouenHasuki]

I would imagine its due to Nvidia wanting to play it safe or maybe they just want AMD to be a lab rat with the latest stuff and take it from there

[Acquire]

Nvidia is good at understanding when something is not worth using.  The only thing using HBM2 and adding a bunch of extra vram would achieve is higher prices and lower supply.  Their architecture and software features are their primary way of providing performance and value.

 

AMD seems cornered.  At this point they appear to be relying on just creating buzz with new technologies to make their products standout against their ever shrinking market share.

 

It's an interesting setup, because Nvidia then benefits from AMD's frantic attempt to create/adopt new technologies to standout by adopting second generation versions of those new technologies once they become feasible and beneficial.

 

The few times Nvidia has gone first, like with g-sync, have been when those technologies brought big value to consumers.