Would you prefer AMD wait until early 2015 to release new cards (with 20nm or lower?)

T

tybert7

2[H]4U
Joined
Aug 23, 2007
Messages
2,763
I ask because it seems like nvidia is going ahead and releasing a new series of cards without bothering to wait for a node drop. They have gone longer in the cycle without releasing their next generation gpus, but why not wait until the chip fabs are ready to release lower process node gpus?
 
From a consumer stand point I'd prefer them to release as close to Nvidia as possible. This way you know basically what each side is offering and what price enabling you to make the best decision.

In regards to moving to 20nm in general, according to TSMC own material (so these are likely more optimistic numbers) -
30 percent higher speed, 1.9 times the density, or 25 percent less power than its 28nm technology.
Click to expand...

1.9 times density is largest change and sounds amazing but if you factor in higher wafer costs, demand vs capacity, and yield (less mature node) it may not.

But unless you actually have deep industry information it's impossible, other than wild speculation, what the actual pros/cons to this type of decision is.

You'd also need to factor in reasons other than technically related to the end product, for example getting something out for the holiday shopping season.
 
the rumours of a full tahiti chip with 3072 shaders would suggest a stop-gap product ready for the arrivaal of big-maxwell, but equally suggests that it will be a short lived bridge productwaiting for a new 20nm GPU early next year.
 
tybert7 said:
I ask because it seems like nvidia is going ahead and releasing a new series of cards without bothering to wait for a node drop. They have gone longer in the cycle without releasing their next generation gpus, but why not wait until the chip fabs are ready to release lower process node gpus?
Click to expand...

Fiji is 28nm, expected Q1 '15 and ~550mm2.

limitedaccess said:
In regards to moving to 20nm in general, according to TSMC own material (so these are likely more optimistic numbers) -

1.9 times density is largest change and sounds amazing but if you factor in higher wafer costs, demand vs capacity, and yield (less mature node) it may not.

But unless you actually have deep industry information it's impossible, other than wild speculation, what the actual pros/cons to this type of decision is.

You'd also need to factor in reasons other than technically related to the end product, for example getting something out for the holiday shopping season.
Click to expand...
That was specifically for 20nm vs 28HP.
20nm benefits vs 28HPM, other than density, are much smaller.

The issue is, the 20nm process didn't ramp quick enough to have enough volume to go around. Throw in having to do double patterning for 20nm and below, 16FinFet being right around the corner and the easy portability of designs from 20nm to 16FinFet and you would need to have crazy volume to even think of using 20nm.
 
I'm sure there are financial pressures to push AMD (and other GPU manufactures) to release incremental stop-gap products with little tweaks. Power usage is the most obvious.

With nothing worthwhile worth the upgrade on the horizon, waiting on node shrink & 5K screens upgrade combo.
 
I'd prefer AMD wait only because I don't think it'll be much longer. I plan to get 980s because my GTX 680s are no longer cutting it for me on some games because of my resolution and 120hz goal but AMD may replace those cards(My GTX 980s) if they come out with 20 nm cards right when the tech is available. I don't expect the Nvidia GTX 10xx (or whatever they will be called) until Q4 2015, which I anticipate will be a fully unlocked Big Maxwell in all it's glory @ 20nm.

If AMD responds to GTX 9x0 with a 28 nm response, I'll be less interested in that.
 
Last edited:
If AMD waits to release the R9 285 follow on, it will lose sales.

Black Friday is a bit of a watershed moment. Between there and Christmas, a lot of sales get made. If AMD has nothing high-end to compete with the 970/980 launch, they will miss. Period.

I imagine they're waiting for tomorrow's pending 979/980 NDA lift to see how their upper/high end cards need to price/perform vs. Nvidia. I'm guessing they're sitting on a 285x or 295 and either tweaking with bios (unlikely) or holding off on MSRP.

My .02
 
c3k said:
If AMD waits to release the R9 285 follow on, it will lose sales.

Black Friday is a bit of a watershed moment. Between there and Christmas, a lot of sales get made. If AMD has nothing high-end to compete with the 970/980 launch, they will miss. Period.

I imagine they're waiting for tomorrow's pending 979/980 NDA lift to see how their upper/high end cards need to price/perform vs. Nvidia. I'm guessing they're sitting on a 285x or 295 and either tweaking with bios (unlikely) or holding off on MSRP.

My .02
Click to expand...

+1

;)
 
I dunno folks, the GPU industry has a long history of releasing new architectures on tried-and-true processes, then only later moving them to finer processes for more speed and lower build cost. Just a few off the top of my head:

Radeon 9700 Pro, done on the same 15nm process they used for the 8500, which could have been majorly delayed if they'd gottten fixated on 130nm like Nvidia did with the FX series. Didn't see a 130nm shrink until the x800.

Nvidia went with the positively ancient 90nm process for the legendary G80, and stuck with that for a whole year before moving along when the 65nm process became stable. Meanwhile, ATI got stuck with a piss-poor 80nm process forthe 2900 XT, which mad them late and power-hungry.

There is no wrong time to release a new architecture in relation to process node (more to do with seasons and new releases from the competition). Also, if you release a new architecture on an older process, it's more likely to be ready on -time, and gives you an easy way to add more performance/efficiency a year down the line.
 
Last edited:
The problems are in general with the new process are that it can be expensive. And today it seems that dropping to a certain nano meter does not hold as much value as people think it does. Mature process can be better option.

If you followed SSD production it seems that they moved up a node to produce the new 3d V-nand?
 
What are people's expectations with 20nm? Keep in mind the earlier numbers mentioned and that they were compared to the older 28nm HP process as opposed to the newest 28nm HPM process.

One thing to keep in mind is the last couple of process shrinks have been coupled with large architectural changes from both sides. So you cannot attribute the gains to one factor over the other.

Pieter3dnow said:
If you followed SSD production it seems that they moved up a node to produce the new 3d V-nand?
Click to expand...

Samsung V-nand is 30nm-39mm (wasn't specified I believe) whereas they were shipping 20nm nand drives. There are competitor drives shipping with 16nm and 15nm (although bear in mind these numbers aren't necessarily standardized in terms of reporting/classification).
 
limitedaccess said:
What are people's expectations with 20nm? Keep in mind the earlier numbers mentioned and that they were compared to the older 28nm HP process as opposed to the newest 28nm HPM process.

One thing to keep in mind is the last couple of process shrinks have been coupled with large architectural changes from both sides. So you cannot attribute the gains to one factor over the other.
Click to expand...

Unless something has changed in the last ~6months, I don't think we will even see any GPUs built on TSMC's 20SOC.
 
Yes it does seem like the current focus is TSMC 16nm FinFet process which purports larger gains in terms of efficiency and performance. Although density is not supposed to change as much (since supposedly the process is more accurately described as 20nm FinFet).

But the question is still there. I assume if one has some preference in regards to process node transition they have some sort of expectation on what that would bring consumer side.
 
Don't care. Gpu scaling has stalled IMHO. And titles coming out demanding more power also have too. We're at a stalemate currently, and I don't mind, since my next purchase is a nice projector, which won't require anything more than 1440p. Currently, 290x with a hefty oc and vmod under water at 50c is keeping me sated.

If the rift comes out with dual 1080 screens or a single screen capable of resolving the screen door effect (likely requiring one of those ultra wide aspect ratios) then perhaps I'll throw a new PSU and 290x in for crossfire... Even so, I still have no true need for the sub 28nm parts...

I believe the maxwell refresh will hurt sales due to awesome power efficiency. But not cannabalize them entirely due to up front cost.
 
16nm maybe but 28nm is old now.
I have no rush to upgrade my card yet and I wait for the shrink to happen.
 
R3MF said:
the rumours of a full tahiti chip with 3072 shaders would suggest a stop-gap product ready for the arrivaal of big-maxwell, but equally suggests that it will be a short lived bridge productwaiting for a new 20nm GPU early next year.
Click to expand...

This rumor has been denied by AMD. R9 290X is the full Hawaii chip.
 
Relayer said:
This rumor has been denied by AMD. R9 290X is the full Hawaii chip.
Click to expand...

ah well.

another thought:

shifting process is an expensive business in its own right.
with maxwell's amazing power efficiency nvidia could provide a competitive performance product without needing to resort to the cost of bleeding edge process transitions.

amd don't have a 'maxwell' in the bag, therefore they need 20nm more more than nvidia do.
 
Yeah,I was very disappointed with R285. I was hoping for Bonaire class efficiency gains, especially when they cut 1/3 the memory bus. But it was not to be.

Maxwell really sucks for AMD because it means they will lose big in gaming notebooks. Most enthusiasts don't care about desktop power, but efficiency is what wins you notebook designs.
 
I wouldnt count Fiji out quite yet...

the 970/980 while impressive efficieny gains, are certainly out of the ballpark performance wise...
 
While the decrease in power consumption is nice and all, I was expecting more performance even if we are still stuck at 28nm; to be worth a new product release.

I think Fiji will be in very good shape when it releases.
 
Yakk said:
While the decrease in power consumption is nice and all, I was expecting more performance even if we are still stuck at 28nm; to be worth a new product release.

I think Fiji will be in very good shape when it releases.
Click to expand...

If it misses the holiday season as others pointed out, it will miss a large chunk of potential sales. And if AMD releases a bigger version of the hawaii boards that are more powerful, they may take the performance crown, but in the same way that fermi took the crown vs AMD way back when.

The chips ran much hotter, and used much more energy, and were more expensive to produce for a given performance level. AMD needs to get to their next generation gpus asap. I guess we'll see what they announce. But the 970 will cause a blood bad for amd gpu sales this holiday. People already break towards nvidia 2:1 when nvidia has a higher cost/performance metric. Now that nvidia has a lower cost/performance metric, it will get much worse.

AMD lost gpu share again this last quarter did they not? So even when they give better performance per dollar, they cannot make headway. I don't know what they can do to get ahead. Perhaps around a third of the market is all they can hope for.
 
I want an amd gpu for an htpc/gaming rig. It'll feed an AVR for sound and pass the 1080p signal through to the screen.

Low heat, moderate performance, hdmi 2: r9 285 is just a bit shy for future proofing. Wonder what amd will release to compete with 970?
 
I wouldn't mind if AMD released a refresh of the 290x with HDMI 2.0 support or even make an overclocked card that would match the GTX 980 would be awesome, I'll buy one too.

Whatever keeps the competition going, price wars is good.
 
I'd love to see the 300 series now, with power consumption at least that low as Nvidia has. Selling my 29)X and breaking even for 970, just for the sole reason of having 200W pulled less from wall.
 
You should wait for new gen amd , they are going to be way powerful as always than nvidia.
 
tybert7 said:
I ask because it seems like nvidia is going ahead and releasing a new series of cards without bothering to wait for a node drop. They have gone longer in the cycle without releasing their next generation gpus, but why not wait until the chip fabs are ready to release lower process node gpus?
Click to expand...

I always prefer ruthless kick to the balls competition between companies. Always works out best for the consumer, always drives innovation.

So yes, AMD should release sooner to compete with NVIDIA. They will either have to compete by price war or continue innovating.
 
Irvine said:
I always prefer ruthless kick to the balls competition between companies. Always works out best for the consumer, always drives innovation.

So yes, AMD should release sooner to compete with NVIDIA. They will either have to compete by price war or continue innovating.
Click to expand...

They need to compete more on price now, the problem is that their 290 series chips are more expensive to produce are they not? More expensive memory interface? I am not sure on the details here but if they are more expensive, then that leaves them less room to maneuver on cost vs nvidia.

Oh well.
 
I would prefer AMD not depend on node shrinks for efficiency! NV can reduce power, and increase performance while on the same process. So waiting for 20nm sounds like a cop out :p unless they drop 290x prices considerably.
 
Yeah, that's what really amazes me about Maxwell: the performance increased, power consumption dropped AND they also dropped the die size (both compared to GK110). The fact that they managed to do that opened-up more die and power space for big Maxwell to be released soon.

GM204 is smaller than GK110 AND Hawaii, and outperforms both it's predecessors in absolute performance, and performance/watt. That's absolutely amazing!

I gave AMD a pass when they released Hawaii, because I stated that they were optimizing their design for die size and performance first, and power efficiency second. But now with the release of mid-sized Maxwell, I have to reverse that opinion. And I don't foresee AMD changing their approach anytime soon, since Tonga was such a disappointment (same die size, power and performance as Tahiti).
 
Last edited:
Tonga and the new Nvidia cards seem to be following the same path. I'm no more impressed with Tonga as I am with the new Nvidia cards as I am certain that both releases are mid to low range releases. I always get the 2nd to highest tier card when I upgrade. Nvidia priced themselves at $549 and $329. Even if the 980 was the high end card there is a $220 difference in price. That tells me that there will be a card in the $450 range at some point where I usually buy at.

http://www.digit.in/pc-components/amds-david-bennett-on-gaming-great-apus-the-competition-23989.html

"Let me put it this way, when I started out 2 years ago, two companies would keep me awake at night- Intel and Nvidia. Today, only one company does that- Intel. We are 100% committed to graphics and gaming, we’re not going to be distracted."

AMD seems confident that they have an answer in waiting for Nvidia.
 
You must log in or register to reply here.
Back
Top