Slashdot Mirror
Apple's Dev. Tools Hint @ Dual-core G5 & Quad Mac
Eug writes "Apple just released a new version of its CHUD tools, which provides clues about dual-core G5s and quad core Macs (dual dual-core).
The clues include a reference to the 970MP, which is dual-core G5 with increased L2 cache. Also, there is now support for 4 CPUs, whereas previous versions of these Mac OS X tools only supported 2.
This likely means we'll see dual-core Macs by WWDC, and possibly quad Macs based off these dual-core chips by then too."
I can't help thinking that this is bad timing on Think-Secret's part. To raise your profile by doing (again!)
The reason is that Nick Ciarelli is making beaucoup cash from Think Secret. It has been a cash cow for him.
On the other hand, there's the case that if he's not doing anything wrong, why not continue doing exactly that.
The blog world is having a field day with this case but the reality is that Apple is not trying to limit what blogs report, and they do not consider Think Secret to be a blog site. It is a rumor site that has broken the law by soliciting confidential information and compensating those who choose to divulge confidential proprietary information. Good write up on the reality here.
Visit Jonesblog and say hello.
CHUD = Computer Hardware Understanding Developer, a suite of optimization tools for Mac developers.
970MP = new G5 processor with dual cores (mentioned in TFS no less)
WWDC = Apple's Worldwide Developer Conference
-mkb
Theory is that the dual core chips will run cooler than two single core chips and there's less real estate on the motherboard required (one socket and less glue circuitry).
John.
Would two dual-core CPUs be considered 4 discrete processors by the OS? It seems to me that 4-processor support in OSX probably means two physical dual-core CPUs
One CPU is always one CPU. A dual core die is two CPUs both in software in hardware. In this context, a CPU is a processor is a core. So they are talking about discovering the ability of the software to handle four processors. That could mean a single quade-core (unlikely), two dual-core CPUs, or four CPUs on four dies (also unlikely).
When you mix hyperthreading in, that gets tricky because then it is a single CPU represented as two CPUs to the OS. The pipeline of a G5 is not designed in a way that would benefit much from hyperthreading, so that will probably be left to Pentium IV processors for now.
There actually was a four-processor Mac once- The DayStar Genesis clone. It had 2-4 604e processors, and required a special asymmetric multiprocessing library and software written for it (this was back in the days of System 7.5).
Just to rephrase the last part of your statement: The latest Pentium IV processors have such extremely long pipelines that "bubbles" in the flow cause huge performance penalties. Hyperthreading is an attempt to kill this penalty by allowing the processor to do unrelated work during this bubble, rather than effectively sitting idle during that time.
The G5 does have the longest pipeline of any PowerPC processor, but it is still short enough that "bubbles" are rarer and not the enormous penalty that they are on the Pentium. Therefore Hyperthreading is neither needed, nor would it be beneficial.
Saying that the G5 is not designed to benefit from pipelining is misleading.
Add to this that mutli-core chips usually have better memory interconnects with each other, improving speed to a small degree...not to mention that either there are more memory coherence interconnects for all chips involved (assuming a point-to-point design similar to AMD aka Hypertransport) or a shared memory bus like Intel (and reduced memory throughput per chip)
In short 2xDual is better than 4xSingle.
Make sure everyone's vote counts: Verified Voting
Right now we are suffering from a bottle neck in rendering. We typically make new hardware purchases every major Apple upgrade typically about every 6 months. The former top of the line machines get intergraded as render nodes, with older machines going off the render farm for general business use like for the secratary. (G3 400'sand G4 500's still run MS office just fine).
We were seriously considering purchasing some Mac Mini's and adding to the group for about $600. (Base model with 512MB of Ram). Why? Well for the base price of one entry level G5 we could purchase about 4 Mac Minis. 4 versus 1 machines, so even if one fails we still are at 75%. Typically we figure that if the Mac Minis took 1 hour each per frame and a G5 Tower could do 2.5 we'd still be rendering 4 frames per hour.
We currrently have 6 Mac Mini's on order to test them out. If all else fails, the the none Graphics people will be getting nice new upgrades on their desks.
Having this kind of information though is nice because it plays a role in determining when we make major hardware additions (we're talking spending $50k+ at a time). The thing is, we don't typically purchase first generation Apple anything. We purchased 12 new Dual G4's two weeks after the release of the first G5's just to make sure the bugs were worked out. We just figured we'd play a hunch and buying $3,000 worth of new computers isn't a major purchase to us (2 copies of Lightwave in our business).
"The problem with socialism is eventually you run out of other people's money" - Thatcher.
Hyper-Threading is a marketing term for a perfectly legitimate microarchitectural feature more generally known as simultaneous multithreading (SMT).
SMT has nothing to do with "dual input pipes" and absolutely requires more than "a single execution unit." The essential features include storage for more than one processor state (data and status registers, program counter, etc.), the ability to mix instructions from multiple threads within the CPU, and multiple execution units.
The Pentium 4 processor with Hyper-Threading has all of those things, including SEVEN execution units that can operate in parallel (but very rarely all at the same time). HT is a simple version of SMT, but it was also fairly inexpensive to add, giving a 20%-30% performance boost for just a few percent more die area.
Yes, a true dual-core processor without SMT can achieve much better throughput on some code than a single-core SMT processor. But that dual-core processor will be almost twice the size of a single-core SMT processor. Dollar for dollar, SMT is a much better way to get more performance than a multi-core design.
The best designs will use both-- SMT within the core because it's such a big win in cost and power efficiency, and multiple cores because that's the most practical way to take advantage of the high transistor counts available in current semiconductor manufacturing processes.
. png