AMD Details Bristol Ridge And Stony Ridge A-Series APUs At Computex 2016

Status
Not open for further replies.
I think the decline in the desktop market is mostly due to need. I look at my desktop that I built in Dec. 2010 and the only thing I have needed to upgrade was the GPU (6870 to R9 390). The CPU runs everything I throw at it and does so in stride. Intel is on the 6th generation Core CPU's and I haven't found a need to replace my 1st generation i7-950, plus, I haven't even overclocked it yet. It doesn't help that the last big increase in CPU performance was from gen 1 to gen 2.

Supply and demand, but with the CPU market as it is, there isn't much need driving demand like there was before the Core series. Maybe when software demands more power from the CPU, things will change. The GPU market seems fairly healthy right now as the software increases the demand.

There seems to be a lull in the CPU market that is driven by a lull in CPU performance increases.
 

apazeus

Honorable
Jul 17, 2015
11
1
10,515
In my humble opinion, part of the reason the PC industry seems to be plateauing in terms of growth is because, over the past few years, hardware has become good enough that it can all handle the average user's needs without needing to be replaced for a long time. It's interesting to see a company like AMD focus so heavily on all-in-one chips. This segment has a legitimate shot to become THE segment within the no too distant future.
 

alextheblue

Distinguished
Replacing the aging Jaguar quad cores in the entry-level segment with a pair of Excavator cores was a great move. Overall performance might be up, but the biggest gain is single threaded performance! That should make cheaper machines feel more responsive in daily tasks.

For Bristol Ridge, I'm glad they segregated more by TDP. That helps tighten the range and allows consumers to genuinely configure a custom machine with the TDP they want (or closer to it). We'll have to see if any OEMs release a single-channel-only solution this time around for Bristol Ridge - the TDP issue is under control, so I hope we don't see a repeat of the memory debacle too.
 

none12345

Distinguished
Apr 27, 2013
431
2
18,785
So, i dont get it.

Nvidia announces 1080, and no mention of paper launch(which was correct at the time). They said cards will be avaliable on xxxx date, cards were available on xxx but sold out in minutes because there was extremely short supply. Thats a paper launch, and i didnt see mention of the term until it actually happened on this site. Which is the way it should have been.

AMD says something about polaris, and does NOT say a date that it will be released, and its immediately branded a paper launch? WTF.

Its only a paper launch if they promise it on a certain date, and dont deliever, or deliver very few units for weeks or months.
 

bit_user

Polypheme
Ambassador
FFS, another round of 28 nm! Enough!

AMD could kill Intel on the VR-ready PC front, if they'd pack a Zen core with their Radeon 480 and 16 GB of HBM2 into a single APU. It'd be an expensive product, but would probably enable sub-$800 VR-ready PCs, with well above the minimum specs.

But I've no doubt Intel will beat them to the punch.
 

InvalidError

Titan
Moderator

4GB of HBM2 on an APU would be more than enough when there is 16GB of much cheaper DDR4-2400 next to it. The IGP only needs enough HBM to cache actively used data and from Intel's Broadwell, we know that even a meager 128MB of local buffer already goes a long way towards covering that.
 

eklipz330

Distinguished
Jul 7, 2008
3,033
19
20,795
FFS, another round of 28 nm! Enough!

AMD could kill Intel on the VR-ready PC front, if they'd pack a Zen core with their Radeon 480 and 16 GB of HBM2 into a single APU. It'd be an expensive product, but would probably enable sub-$800 VR-ready PCs, with well above the minimum specs.

But I've no doubt Intel will beat them to the punch.
if they did that and it was priced cheaper than those parts if purchased individually, home theater market would explode. I would be willing to put down $500 on a chip like that, and all i would need is a case, mobo, psu and ssd... that form factor would be sleek
 

mavikt

Distinguished
Jun 8, 2011
173
0
18,680
They don't make it easy for themselves, 25 times improved energy efficiency, on 28nm... Wonder if that was intentional or are they planing on a die-shrink before 2020?
 

somebodyspecial

Honorable
Sep 20, 2012
1,459
0
11,310
ROFL. 28nm in mid 2016. FAIL. If you own AMD stock you may want to sell before the next quarterly report ;) I don't expect them to make money unless ZEN is larger than 200mm^2 and really, I mean won't make much unless it's >250mm^2 (Note AMD has done 300mm^2 before). AMD keeps aiming for low-end garbage that has no pricing power. A HUGE ZEN core could be priced above Intel chips at every level. If ZEN is not at least 50% larger (cpu die to cpu die) vs. Intel's current gen cpu, they won't be able to price for profit. If it is anywhere near equal, even if it wins in this or that (while losing some crap to intel in other stuff) Intel will just price it to death until they win everything again.

Dirk M. was correct. You NEED a KING CPU before bothering with anything else. Same story for GPU, ask NV how those profits are on their top 3-5 cards (pro included).

Console margins will never make AMD rich either, which is why NV passed. You need to spend R&D where you get the best bang for your buck (highest margins!). Then once richer (or in AMD's case actually making profit for a year solid), you can try crap like a console, apu, etc that make peanuts. Quarter after quarter AMD set sales records for cpu shipments (with apu), yet made loss after loss. Quit making crap that can't price like god's gift to mankind or get it over with and go bankrupt. 28nm...pfft.
 

RazberyBandit

Distinguished
Dec 25, 2008
2,303
0
19,960
Dear AMD,

Please align yourself with an OEM that will actually choose to use these in systems and NOT intentionally handicap them to the bare minimum TDP like all your OEM partners have been doing with the current and previous generations.

I've had my hands on several Carrizo-based systems over the last year or so and the performance always comes up short of what originally expected I these systems to deliver. Why? Because none of the OEMs will bother giving the chips more than the minimum 15W TDP.
 

bit_user

Polypheme
Ambassador
The thing is, if you're trying to optimize total system performance per dollar, it'd be better to put all the system memory in the CPU. That way, the CPU would have fewer pins and the motherboard would need fewer traces & no DIMM slots (until people start demanding NVDIMMs, but that's not yet the case).

Also, on a decent sized GPU (480 or greater), DDR4 is still going to be a major bottleneck, with only 4 GB of HBM. Otherwise, I think graphics cards would be mixing a few GB of HMB and 256-bit or 128-bit GDDR5, but they're not.

 

InvalidError

Titan
Moderator

There is no one-size-fits-all for RAM and HBM will cost at least twice as much as standard RAM, quickly eliminating the cost benefit of reducing pin count. People who only want an office PC will balk at having to pay $150 for 16GB of HBM2 they have no use for instead of $50 for the 8GB of DDR4 that they actually need. Higher-end users will also turn their noses at being forced to buy into a non-upgradable chip if they have any reason to think they may need more than 16GB in the foreseeable future - like me who has 32GB on my i5-3470.

As for 4GB vs 8GB on a RX-480 class GPU, wait for the reviews. I predict that the extra memory will make little to no difference in most games at playable details and resolutions, just as it made absolutely no difference in most cases on previous-gen GPUs - by the time the extra RAM makes a significant difference, you are usually deep into unplayable territory.
 

bit_user

Polypheme
Ambassador
What's this about an office PC? I was specifically talking about a hypothetical entry-level VR all-in-one.

For VR, you'll definitely need more than 4 GB between the CPU and GPU. 16 GB for both should be plenty.

It will happen, eventually. The benefits are just too great. And yes, upgrading your RAM will mean replacing the whole thing. But since swapping/paging to NVDIMMs won't be so bad (which will be standard, by then), people might just not bother.
 

InvalidError

Titan
Moderator

The way I look at it:
- APU with 16GB of HBM: $250+ (~$100 for the APU, ~$150 for 16GB HBM module)
- APU with 4GB of HBM + 16GB of DDR4: $200 ($110 for the CPU, larger socket, larger motherboard, DIMM slots, etc., $30 for the HBM, $60 for the DIMMs)

HBM is too expensive to use as system RAM. A bargain-basement office or non-gaming home PC would be better off with no HBM whatsoever.
 

bit_user

Polypheme
Ambassador
Wow, it's like you just skipped over half of my post! I'm specifically talking about a VR all-in-one! What a way to win - just change the argument to one where your position is correct!

An APU without HBM/HMC won't have enough memory bandwidth for VR.
 

InvalidError

Titan
Moderator

I doubt people going with APUs are going to be interested in VR any time soon and I would not be surprised if VR turned into a passing fad much like 3D TV.
 

bit_user

Polypheme
Ambassador
AMD has a stated strategy of building GPUs that put VR within reach of the average gamer, and making such an APU would be entirely in line with that. The main reason APUs are slow is because they lack the memory bandwidth to feed a more substantial GPU, which HBM2 would solve.

In fact, as if to underscore this point, there are leaks of a HPC-oriented version of Zen, which includes a substantial GPU, a large chunk of HBM2, and DDR4 support (because HPC workloads sometimes need a lot of memory).

But I get that you didn't win the argument, and couldn't change it, so now you're attacking its premise. Nice.
:p
 

InvalidError

Titan
Moderator

But putting 16GB of it on the APU would price it out of the market, which wouldn't do anyone any good. Entry-level graphics don't need 8GB of HBM (even 4GB may be somewhat of a stretch) and people who are looking for inexpensive APUs won't be buying high-end VR sets. That's assuming VR ever gets popular.
 

bit_user

Polypheme
Ambassador
You've got it backwards. It's people looking for entry-level VR that would be attracted to a high-end APU (if one could meet the minimum spec).

As an existence proof, I point to XBox One and PS4. Both are APUs, and XB1 uses integrated memory (much smaller, since HBM wasn't yet available). The fact that they went with APUs strengthens the argument that APUs are the most cost-effective solution. The challenge, then, is to build one that's up to the task.

The economics of putting HBM2 in an APU will sort themselves out. It's really a question of timing. I'm just waiting to see who'll get there first.
 
Status
Not open for further replies.