Slashdot Mirror
AMD Launches New Processor Socket Despite Poor Economy
arcticstoat writes to tell us that despite a poor economic climate, AMD is moving forward with a new processor socket launch, although they are trying to make it as upgrade-friendly as possible. "As you probably already know from the AM3 motherboards that have already been announced, AM3 is AMD's first foray into DDR3 memory support. As Phenom CPUs have integrated memory controllers, it's more accurate to say that it's the new range of Phenom II CPUs (see below) that are DDR3-compatible. However, the new DDR3-compatible Phenom II range is also compatible with DDR2 memory. As the new CPUs and the new AM3 socket are pin-compatible with the current AM2+ socket, you can put a new AM3-compatible CPU into an existing AM2+ motherboard. This means that you can upgrade your CPU now without needing to change your motherboard or buy pricey new DDR3 memory."
The latency is generally lower than DDR2, measured in wall-clock time. The advertised latency appears worse only because of the faster clock.
Or a lot of small-budget husbands :P
You may be able to put a am3 processor in a am2+ motherboard, but the Register says that am2+ processor in a am3 motherboard will not work. (http://www.reghardware.co.uk/2009/02/09/review_cpu_amd_phenom_ii_am3/page2.html)
To quote: ..
"makes life horribly confusing as the Phenom X4 920 and 925 and the X4 940 and 945 will be identical apart from the processor socket. This means that there is the possibility that some poor so-and-so will buy an AM2+ CPU and an AM3 motherboard when ne'er the twain shall meet."
careful what you buy out there
This is the sort of thing that gets us out of a poor economy.
That's bullshit, CL in periods * period length = latency, and since they are clocked higher the latency will probably be around the same, I won't calculate it for you.
And that latency is how long it takes before you actually start to read any bits, but as soon as you have started each bit will come faster from the higher clocked memory.
If you don't get a speed increase it's because either of:
1) Processor not fast enough to take benefit of additional bandwidth.
2) Cache system smart enough to not take benefit of additional bandwidth.
3) Application not using memory in a fashion where it will take benefit of additional bandwidth.
Most likely the later one ..
All higher end graphic cards come with faster memory, it may not be a huge deal always but it probably add some benefit, rather stupid if it didn't.
AMD said they would skip DDR2 and go directly to DDR3 earlier because there was no benefit when actually in use but I guess they "had to" when Intel was using DDR2 just because people see the numbers and wonder why one is bigger than the other.
Though first AM DDR2 chips vs 939 DDR chips showed no increase in speed in benchmarks.
Anyway, DDR3 is faster than DDR2, will you notice it? I have no idea.
AMD is competitive at the low and middle end as long as you don't overclock the middle end CPUs.
(If you include the price of the motherboard and don't care about overclocking a low- or midrange AMD system will be cheaper.)
AMD don't have as high end CPUs as Intel and the ones which are closest don't overclock as good or use as little power.
Though then I'd say you shouldn't overclock anyway and AMD chipsets have used less power making the two when used in a complete system rather comparable.
Also AMD used to have an advantage in memory bandwidth and when using multiple CPUs.
Information may be slightly outdated but all of it is probably true, Intel may have catched up in memory bandwidth performance with their latest CPUs since they have put the memory controller within the CPU themself to.
The 3-fold clocking scheme will only really help on interleaved burst reads. The memory cells don't charge the output buffers any faster just because you clock them at a higher rate. This is why the nCLK latencies scale with the number of folds in DDR scheme. The only things that will make the cells charge faster are a) higher voltage or b) smaller process or c) a more conductive semiconductor chemistry that lowers resistances and increases currents on the wafer.
If you can have 3 banks of DDR3 interleaved by 1 clock then you can probably see some significant gains on sequential (aka burst) reads. In real life, this doesn't happen very much, especially in a multithreaded environment where almost all s/w is written using high-level foundation classes with very little machine optimization.
This is probably something not many people care about, but...It's a hell of a lot easier and cheaper to find an Athlon64 motherboard that supports(and actually does ECC) ECC memory. Think $50-$100 for an Athlon64 motherboard that does this, versus $200-$300+(original Asus Maximus Formula, Asus P5E WS Pro) for a Core 2 motherboard(has to have an X38 northbridge, unless you want to give up PCIe x16 with a server chipset). I don't think the currently released Core i7 processors with built-in memory controllers support ECC *at all*.
(PS to trolls: Unbuffered ECC memory is only marginally more expensive than unbuffered non-ECC, though it usually has a small latency penalty. Registered/FB-DIMMs ECC on the other hand are Quite Expensive)
AMD has the Geode LX and NX lines.
Geode LX is very low powered and the highest clock speed (I've seen) is 566Mhz.
Geode NX is targeted directly at the Atom. Although I have yet to see any of these out in the wild.
I've only ever found a Geode in the wild clocked as high as 500Mhz (see the ALIX boards)
Actually the Geode is a dead end processor, AMD already has stated they are disconinuing it.
AMD recently announced a new processor "Conesus" that is intended for netbooks and UMPC.
http://gizmodo.com/5086703/amds-upcoming-conesus-netbook-chip-wont-stoop-to-mid-levels