AMD Enters Desktop Memory Market 65
siliconbits writes "AMD has quietly released a new range of memory products and recycled the Radeon brand, which moves from graphics processing units to memory modules. According to the product page, AMD Radeon for systems are 'ideally' suited for the company's APU and CPU solutions and have been 'tested to the highest industry standards on AMD platforms.' Three different categories are currently on offer, roughly matching AMD's APU/CPU product categories; Entertainment, Ultra Pro and Enterprise. Oddly enough, the company is offering only 2GB modules with data rates at 1333.33MT/s and 1600MT/s, with 9-9-9 and 11-11-11 timings for the first two product ranges respectively."
Better idea! (Score:2)
Maybe they could take whatever they're smoking and sell that instead. I'd pay good money for it!
Re: (Score:2)
They get their stuff from me.
AMD has too many products (Score:1)
Isn't AMD's single biggest weakness a lack of fabrication facilities? And they're introducing a line of memory modules for some heretofore-unseen vertical integration on the motherboard... Using the Radeon name. Poor ATI.
Re: (Score:2)
Re: (Score:2)
Re: (Score:3)
But wouldn't it seem likely that AMD still owns some equity in GlobalFoundaries? It seems like AMD's move was to better utilize its facilities. Given that the company isn't public, this kind of information is hard to find but I can't imagine AMD would just spin it off if they would take a loss.
I do believe they kept 51%. They spun it off because they just didn't have the financial depth to keep up with the massive, massive investments involved in next-gen processing tech. Their competition is Intel and TSMC - which is also becoming a bigger and bigger behemoth. They simply had to scale up or get out, and this was their way of scaling up.
Re: (Score:3)
Re: (Score:2)
It stinks and Intel should have been busted for antitrust.
They kinda were. [nationaljournal.com]
Re: (Score:3)
Re: (Score:2)
GlobalFoundries is used by AMD exclusively for their CPUs. I expect this memory venture is likely resale of someone else's DRAM.
Re: (Score:1)
Isn't AMD's single biggest weakness a lack of fabrication facilities?
No:
http://en.wikipedia.org/wiki/GlobalFoundries [wikipedia.org]
AMD used to own their own FAB.
A few years ago they spun off Global Foundries into a separate company. They said the usual bullshit about being to be more agile, more focused, whatever.
It was really a move to hide losses and to tell the investors "Hey, we're doing SOMETHING!" (in response to getting their asses handed to them during the Athlon II generation.)
Now AMD has to deal with an external company to get things fabbed. That same fab has to entertain orders from companies other than AMD (such as Intel)
Re: (Score:1)
Gee, if they had their own FAB they would have been able to have much more control and would have been able to react to problems much more quickly.
Or, perhaps, if they had their own FAB and it was fucking up on yield, they would own it and be stuck with it. They have a degree of freedom to shop around that they wouldn't in their 'own' FAB.
Re: (Score:3)
GF has already been spined off and it is an independent entity.
Re: (Score:2)
Re: (Score:1)
Re: (Score:2)
No. Why make up shit? It is not Apple biggest weakness, or Nvidia, or Samsung, or anyone else. There are plenty of fabs going around to do this sort of work.
However, AMD's biggest weakness may be morons who make shit up.
Re: (Score:2)
Why would that be their biggest weakness?
They had manufacturing facilities, they spun them off into a different company and I'm sure they still have contracts in place to have first priority for their use.
I find it hard to imagine whoever thought up splitting off GlobalFoundries from AMD did that on a whim and is now sitting in some office muttering "Oh shit, I accidentally the whole thing".
On die RAM... (Score:2)
Re: (Score:2)
In some embedded applications(smartphones and friends, most notably) "Package on Package" designs with a RAM die packaged on top of the CPU are pretty popular; but that is largely about board space savings, the two dice aren't actually coupled much more closely(which allows them to be test
Re: (Score:2)
Cache almost always uses SRAM, which uses a whopping six transistors per bit instead of one (although it's MUCH faster, latency-wise). Thus you get much higher memory densities from DRAM, with the corresponding price advantage (which is further inflated by economies-of-scale, as DRAM is a much higher-volume product than SRAM). So that's not really what he's saying.
It's still not really a good idea, though. RAM is generally one of the cheapest parts in a computer, and is often the only thing upgraded before
Re: (Score:2)
We are approaching the point where it does not make sense to upgrade RAM. Right now you often only get 256MB per memory chip, and it takes 4 chips to fill a typical bus (x16), so the minimum amount of memory you can sensibly install is 1GB. Obviously you can go with older 128MB or lower chips, but they are not much cheaper. Already you can do 512MB per chip, and with DDR4 we will likely hit 1GB per chip. We are not many chip generations away from being able to satisfy the needs of most users with a few sold
Re: (Score:2)
Except that most of the latency is inherent to DRAM - in the past decade, average latencies have dropped from 10-20ns to 6-7ns, while total bandwidth has skyrocketed from 1600 MB/s per channel to 10666MB/s per channel and capacities have surged from average systems having 128MB total to average systems reaching 2GB or 4GB.
Basically, DRAM latency isn't a problem that can be fixed by moving it on-die. Nobody's found a solution yet, other than piling on more and more cache and hoping your branch predictor work
very very stupid (Score:2)
considering that RAM business has repeatedly saw the bottom fell out from under it over the years (news flash - most RAM chip manufacturers have been losing money on the product most of the time) this is ONE more way AMD is going to accelerate its demise. Whichever MBA bone head that came up with this one ought to be drag out and shot in the base of his/her neck.
Re: (Score:3)
Not necessarily. If they're working on a way of reducing the bandwidth bottle neck the way that I think they're going about it, this is likely to prove to be quite the wise idea.
As they've been moving more and more stuff on die, it's getting to the point where they really need a more tightly tied supply of RAM and I would be surprised if they aren't looking at how to get the RAM closer to the processor. They've been doing it for years and this would be one potential step to the logical conclusion.
Re: (Score:3)
As they've been moving more and more stuff on die, it's getting to the point where they really need a more tightly tied supply of RAM and I would be surprised if they aren't looking at how to get the RAM closer to the processor.
I just looked at my motherboard and the RAM is already so close to the CPU that it almost touches the heatsink. I'm not sure how much closer you think they can move it.
More seriously, if you're talking about building RAM into the CPU, we already have that: it's called cache. And since the amount of RAM you could add to a CPU die is never going to be enough outside of the low-end market, it would have to operate as a cache rather than a replacement for RAM on the motherboard.
Re: (Score:3)
As mentioned elsewhere in this thread, CACHE is usually SRAM not slower DRAM. Yes, it is RAM, but it is much faster than DRAM. IF they are pushing the proximity of DRAM to the chip package it would make sense. Especially if they can widen the bus to the DRAM and remove (or minimize) bottlenecks to the RAM.
Re: (Score:2)
I'm the town idiot. Is the "SoC" the parent referred to "system on a chip"?
Goddamn it. Can someone please tell me what this acronym means? "Byte order mark"? "Bill of materials"?
I'm on acronymfinder.com and I'm sick of trying to guess.
Re: (Score:2)
Re: (Score:1)
Re: (Score:1)
I know a guy that buys blank DVDs by brand name even when I clearly point out the industrially-packaged A-DATA disks for 1/3 the price on the top shelf, so I'm sure AMD will find a good following of sheeple for their memory.
How dare you call me a sheeple, you insensitive clod! I'll have you know... ~BAAA~ damn.
Re: (Score:1)
Different to the shepple that buy OCZ or Corsair how?. AFAIK they don't have fabs either. Being the exeption Patriot and Kingston.
I guess AMD user^n^n^sheeple it's the bunch more focused in bang for the buck, If AMD can't deliver good specs for a decent price people (AMD fanboys?) will not buy it. Kingston is good in that regard. If you're not in the "I can't have slow computor or people will realize my ineptitude" market the choices are suddenly huge.
Corsair Sheeple (Score:2)
Not sure what you mean by Corsair sheeple -- Corsair prices very aggressively, particularly on a short-term sale basis. They are generally not marketed to appeal to the highend hobbyist market (which is what I assume you mean by "sheeple."). For that, you're looking at like G. Skill -- that's a straight fanboy brand.
I'm not much of a memory buyer -- as in, I only buy it every 3 or 4 years -- but Corsair has worked decently for me so far and if when I do a motherboard upgrade that needs DDR4, I will probably
Re: (Score:1)
Re: (Score:2)
Given the current quality of AMD's memory offerings I'm wondering how or why you consider people sheeple for buying them. AMD has been a solid memory manufacturer for a decade or more, it carried them through the hard times of the 486 until the first gen Athlon which came along and toppled Intel's performance crown.
They are hardly perfect but one of the things they definitely do well is making memory. I'll certainly give this ram a shot, challenging preconceptions is always a good idea afterall.
1600MHz is plenty fast for performance users (Score:2)
Three different categories are currently on offer, roughly matching AMD's APU/CPU product categories; Entertainment, Ultra Pro and Enterprise. Oddly enough, the company is offering only 2GB modules with data rates at 1333.33MT/s and 1600MT/s, with 9-9-9 and 11-11-11 timings for the first two product ranges respectively."
The suggestion that 1600MHz is too slow for what AMD is calling "Ultra Pro" (they presumably mean gamers) is just not substantiated by the data [tomshardware.com]:
We looked at different memory speeds for the LGA 1156-based Core i7-870 and chose to run DDR3-800, -1066, -1333, and -1600 at fast, as well as relaxed, timings. Although the differences were typically very small, there were a few applications that obviously benefited from faster memory. This wasn’t surprising, as we already did similar comparisons on most of the other popular platforms:
In all cases, we’ve seen significant performance differences when looking at the synthetic or low-level benchmarks. Memory bandwidth does increase considerably if you speed up the memory transfer rate, and tightening timings also improves performance by cutting latencies. However, only a marginal fraction of these benefits actually arrive at the application level. Even going for the fastest memory available will give you a performance boost that is probably smaller than the effect a faster processor speed bin would deliver.
Re: (Score:2)
Memory is memory; few applications are just paging (Score:2)
Memory is memory -- it doesn't have anything to do with the on-die memory controller. It's an application thing; most applications users use daily aren't doing constant memory pages, so tossing faster memory in is not going to do much.
Re: (Score:2)
So-called "UltraPro" users don't use onboard video (Score:2)
If AMD is hoping to make a killing selling performance memory to "Ultra Pro" users, it's not going to do it by pairing it with Fusion. No performance user (gamers, CAD, etc) that is willing to shell out extra for AMD-branded memory is going to be using onboard graphics -- they would be using a discrete graphics card. In such cases, the relationship between main memory speed and onboard graphics is completely irrelevant. See this review [tomshardware.com] breaking down Fusion's unsuitability for performance users:
When it comes to the desktop space, Llano’s prospects are decidedly less impressive in light of the competition. These APUs make for an ideal solution to replace entry-level PCs with crappy integrated graphics. And, they certainly could introduce a lot of graphics muscle to a segment historically light in that regard. If Llano catches a foothold there, the APU could impact peoples’ expectation of what a PC can do. Developers might start targeting a higher lowest common denominator in their games, and that’d of course be great news for PC gaming.
But once you reach outside of the budget basement and consider folks willing to use discrete graphics, the A-series’ utility is hamstrung. It’s easy to put an $80 Radeon HD 6670 in a cheap OEM box and walk away with something that easily trumps AMD’s product in both processing and graphics benchmarks.
Re: (Score:2)
Re: (Score:2)
Uh, you seem to be confused on your units. RAM latencies are measured in clock cycles - a CAS of 11 means 11 clock cycles for a content address strobe. The exact time will vary depending on the actual clock speed. It comes out to be 6.875 nanoseconds for 1600 mHz. Meanwhile, that CAS of 9 on the 1333 MT/s comes out to 6.75 nanoseconds - pretty much the same.
This is why "faster" RAM has higher latency numbers, and slower RAM has lower latency numbers. DDR3-2400 normally has latencies in the double-digits, de
Re: (Score:2)
I hate it when manufacturers don't print the tRAS number (which is at minimum 24 for the CAS-11 and 20 for the CAS-9 using the CAS + tRCD + 2 formula, where CAS is Column Access Strobe, RAS is Row Access Strobe, and tRCD is RAS-to-CAS delay). If that final number is ungainly high compared to the minimum, it is a often a sign of poorly designed memory (or so I recall from talking to my brother, who works as a RAM designer).
And for reference, as a general rule, RAM at half the speed and half the CAS number is
Anandtech is testing ONBOARD video performance (Score:3)
The story in anandtech [anandtech.com] that you're quoting finds that higher bandwidth memory pushes up gaming performance when you're using the onboard video. That's certainly true, but the performance user willing to pay for performance memory is using a stand-alone video card. With a stand-alone card, the tomshardware results that I linked to are the relevant benchmark -- not the anandtech one you posted.
Don't get me wrong, AMD's on-die graphics are head-and-shoulders above even Intel's Sandy Bridge HD3000 graphics -- b
Re: (Score:2)
Actually, there's only like three or four. But they don't sell direct to consumers: they just sell chips to other companies, who then package them and put their name on it.
Re: (Score:2)
That's not entirely true. Micron owns Crucial which does sell directly to customers. They've been my pick for years now and I've always found the quality to be good as well as the price. It's just really hard to justify buying from a middleman when you typically get a better warranty and price buying direct.
give us 8GB SIMM modules (Score:2)
Re: (Score:3)
http://en.wikipedia.org/wiki/SIMM
vs.
http://en.wikipedia.org/wiki/DIMM
And they do make 8Gb and even 16Gb DIMMS:
http://www.amazon.com/Kingston-ValueRAM-240-pin-PC3-8500-registered/dp/B003C015ZY
But you will likely find that many (if not most) motherboard chipsets do not support them. This is a chipset and bios coding issue more than anything else.
Re: (Score:2)
"And they do make 8Gb and even 16Gb DIMMS:
http://www.amazon.com/Kingston-ValueRAM-240-pin-PC3-8500-registered/dp/B003C015ZY [amazon.com]
But you will likely find that many (if not most) motherboard chipsets do not support them. This is a chipset and bios coding issue more than anything else."
Isn't that server RAM?
A few 8GB DDR3-SODIMMs have showed up (not really interested in anything other than laptops myself, these days), but Sandy Bridge memory controllers seem to be a bit finicky, with Intel only having specified one
Re: (Score:2)
Isn't that server RAM?
Indeed, ECC ("ChipKill") at $300/midule, ykes.
No profit in RAM (Score:2)
The margins on retail DRAM are really slim. Even in "performance" memory, where they charge more, there's still overhead in binning for the more aggressive timing numbers. Unless DRAM sticks were all you did, or your retail DRAM business was a front for a DRAM maker (as Crucial is for Micron), then I don't see how selling DRAM is going to add much to the bottom line of a large company like AMD. Are they going to charge more than Crucial does for the same stuff? Are they going to do something to make buy
Re: (Score:2)
Well, I'm guessing the final goal is not to sell RAM sticks separately but to integrate all the components they manufacture into one system-on-a-chip type thing as they're starting to do with CPU/GPU.
desperate move (Score:2)
Possible reason for this? (Score:2)
Re: (Score:2)
It's more possible that they can generate increased revenue by rebadging someone else's RAM with their well-known "RADEON" trade dress.