DDR3 Isn't Worth The Money - Yet

An anonymous reader writes "With Intel's motherboard chipsets supporting both DDR2 and DDR3 memory, the question now is whether DDR3 is worth all that extra cash. Trustedreviews has a lengthy article on the topic, and it looks like (for the moment) the answer is no: 'Not to be too gloomy about this, but the bottom line is that it can only be advised to steer clear of DDR3 at present, as in terms of performance, which is what it's all about, it's a waste of money. Even fast DDR2 is, as we have demonstrated clearly, only worthwhile if you are actually overclocking, as it enables you to raise the front-side bus, without your memory causing a bottleneck. DDR3 will of course come into its own as speeds increase still further, enabling even higher front-side bus speeds to be achieved. For now though, DDR2 does its job, just fine.'"

Comments

I need to get out more

[Gilatrout]

I read Intel supports Dance Dance Revolution 3.

Duh?

[ynososiduts]

Who in their right mind would pay so much for RAM? The only people I can think of are the middle - upper class teenagers with lots of money. The ones who run 8800Ultra's in SLI thinking that 2 cards = twice the performance when it's more like 30 - 50 % increase. Most educated system builders wont spend more money then they have to, and DDR 3 is just overpriced.

Why do these reviews only focus on one thing?

[downix]

Every time I see "the need isn't there" or "there's more than enough memory bandwidth" I check their figures, they're only measuring the CPU memory needs. Well, hate to break it to you, but there's more to a computer than just the CPU. Having that extra bandwidth means that those lovely PCI Bus Mastering devices (such as my SCSI 3 controller, and quad firewire card) aren't fighting with the CPU for memory access. Frankly, add in a game accelerator like the Phys-X and a high-end GPU fetching data from the main memory for local cache, and even DDR3 starts looking a bit narrow....

Re:Why do these reviews only focus on one thing?

[Slashcrap]

Having that extra bandwidth means that those lovely PCI Bus Mastering devices (such as my SCSI 3 controller, and quad firewire card) aren't fighting with the CPU for memory access.

With a SCSI 3 card and 4 port Firewire you'd be looking at about 360MB/s of bandwidth assuming that they reach their max theoretical speed (and of course PC hardware always reaches its maximum theoretical speed). Unless they're both on the PCI bus in which case 133MB/s max for both. Which is fairly minor compared to the 6GB/sec of memory bandwidth that I get with shitty DDR2 on a shitty motherboard.

Unless you can provide evidence to the contrary, I am going to go out on a limb and suggest that the performance increases you are expecting do not actually exist. Unless your primary workloads involve running memory benchmarks and Prime95 in which case I would point out that you accidentally posted to Slashdot instead of the Xtremesystems forums.

Re:Didn't this happen before?

[Zephiris]

Part of the reason that DDR2 was so much slower at most clockspeeds is because of the added latency. The lower speed DDR2 can have more than twice the tested latency of DDR400. The problem is that apparently both JEDIC, or whoever standardizes memory now, isn't thinking of what is the best direction for DDR to take. They're going in the same direction as the manufacturers, trying to sell higher "Megahertz" and "gigabytes per second" ratings, even when they're effectively meaningless now.

Does it exactly matter if your computer can do 6GB/s, or 12GB/s? 14GB/s? Where does it stop? And even then, that's mostly theorhetical, particularly in the case of DDR2. But a very important distinction is that so many memory accesses are of very small to small size. On basically all of those accesses, the memory request will be served in far less time than the latency will allow the command to return and allow another request.

Way back when, Intel motherboards tried out RDRAM for its 'higher end' boards, and the Nintendo 64 also started using it. Both were fairly large fiascos, in that sense, with more or less all technical reviews noting that the increased latency more than cancelled out the improved bandwidth. Now we're looking at DDR3, with far higher latencies than classic RDRAM for a relatively minor bandwidth improvement that only extremely large memory requests (such as applications that would theorhetically be done in an extremely large-scaled database and scientific research).

It reminds me acutely of the early 'Pentium 4s'. A 600Mhz Pentium 3 could beat up to a 1.7Ghz Pentium 4 in most applications and benchmarks, and the (rare and expensive) 1.4Ghz Pentium 3s were real monsters. But people kept trying to tailor benchmarks to hide that, so people would buy more product.

Overclocking has also generally demonstrated that overclocking regular 'old' DDR1, while a bit pricier (mostly due to the virtual elimination of production nowadays, though), scales better and also has far better numbers than DDR2 and the like. DDR600 equivalent is extraordinarily zippy, and (of course) real-world latency is also absurdly low.

It makes me feel like the 'governing bodies' here have really let people down. Instead of trying to standardize on and promote what's best for general computing, they're trying to push a greater volume of merchandise that has no meaningful improvement, and in fact usually a notable decline, over what we've already had for years. The bottom line for them is money, and that's just wrong to put their own pocketbooks over the long term well-being of computing technology and the needs of the consumer.