AMD launches K10 dual core

[LiamC]

AMD launches K10 dual core--say hello to the Athlon 7750 and brethren

http://www.xbitlabs.com/articles/cpu/display/athlon-x2-7750.html

http://www.bit-tech.net/hardware/2008/12/15/amd-athlon-x2-7750-7550-cpus-review/1

http://www.hexus.net/content/item.php?item=16614

Sometimes better, sometimes worse, but overall an improvement

 

[Stereodude]

*yawn*

So, that's going to help them how?

 

[ddrueding]

That was my thought. I have never been this uninterested in a new chip.

 

[P5-133XL]

Exactly my sentiments. It certainly isn't an i7 killer and that is what AMD needs.

 

[Chewy509]

*yawn*
So, that's going to help them how?

By offering better performance per dollar... until Intel does a price drop to match a few days later.

I'm coming up to looking at a new box (since my current one is 4yr old now), and seriously, nothing is looking that promising from either Intel or AMD. Heck, I might as well as spend the $$$ on new toys for my 10mth old, and keep my current dual Opt system for now.

While the i7 has jaw dropping performance, Intel is about to start playing another round on socket/chipset games, where they'll be changing sockets in the new future. (LGA 1366 from what I've read will only be for the current i7's, and a new socket for mainstream systems will replace it 1Q09). Not to mention that X58 boards are US$200+ (which doesn't equate too well when the Australia dollar is rather low against the greenback at the moment).

And AMD doesn't have the performance that a new system upgrade should warrant, esp someone who'll be moving from a dual socket workstation to a quad-core single socket system (since my workload and credit card can't justify a dual quad-core system).

 

[Tannin]

I am totally unimpressed by the CPUs (of either brand) on the market today. Jaw-droppingly mediochre performance improvements on offer from both sides. Sorry kiddies, if it doesn't run my apps faster, I'm not interested in spending money to replace things that work.

The problem, of course, is this mania for adding more and more slow cores, instead of making something that actually goes faster in a meaningful way. What's the point in going to quad core when your workload isn't maxing out the second core?

AMD's announcement offers us less performance than the ancient A64 6000 and draws more power. Why would I want one of those? Why would anyone want one?

The underlying problem, of course, is that the CPU manufacturers simply don't know how they are going to get a meaningful performance improvement out of their products. They can't see any way forward except adding more and more useless extra cores. I doubt that anyone else an see light at the end of the tunnel either. With current technology, it isn't possible.

All of this leaves us dependant on the software writers to find efficiency, seeing as the hardware guys can't find any more brute power.

And we will see a return to fast, efficient software. Trust me on this. It will happen for sure. Happen the same day that hell freezes over.

(Yes, Adobe I mean you!)

 

[sechs]

I think that we already have plenty of processing power. When I switched my computer from two 2GHz processors to two dual-core 1.8GHz processors, I could just notice the loss in single-core computing power.

I don't really need much more than this. The limitations on my system are my video card and the storage subsystem. I may upgrade the video, but I'm not yet prepared to spend $800 to unleash storage.

We don't really need faster processors. And, I'd suggest that most people don't need more than two cores. So, all that AMD is doing, is filling a market segment.

I doubt any of us here are in that group.

 

[MaxBurn]

I thought the i7 was showing gradual single threaded improvement when compared on a clock for clock speed to the Core2 chips? Not a lot but it's there, somewhat due to all the caching and memory controller improvements. If I understand correctly these things have hyperthreading too so the OS sees eight processors and can split things up even more.

When have we ever really had a drastic speed increase due to different architecture? Aside from the step backwards with the PIII to P4 and the correspondingly large jump to core. I am trying to set clock speeds aside in consideration for this point.

 

[ddrueding]

i7 does seem to be quicker, and my 3.6Ghz quad core does run circles around my previous chips Mhz-wise as well as any other way. CS4 can also easily hit 80% sustained across all the cores doing batch operations.

 

[timwhit]

Is hyperthreading just the ability run more than one thread at a time with a single core? Is this any different than what the UltraSPARC T2 can do?

 

[MaxBurn]

Is hyperthreading just the ability run more than one thread at a time with a single core?

Yup, supposed to utilize the idle times that the CPU has waiting on data. Also seems to force things to "share" better when loaded at a hardware level. The OS actually sees two CPU for each hyperthreaded core.

 

[Mercutio]

On most of the systems I build, there's no subjective difference between, say, a 2GHz E1200 and a 2.2GHz (I think) E7300 and a 2.4GHz Q6600, unless I start using them at what is for me a normal workload - pirating a DVD while doing image-intensive word processing, with 50 or 100 open browser tabs open, several large PDFs (in Foxit, 'cause Adobe has teh suck) open, multiple remote sessions, possibly a VMware Window and in all likelihood Paint.NET or Gimp will be running.

With an average system these days having 2GB RAM, the dual core Celeron will keep up just fine with all of the above until I throw the VM into the mix. With the Quad Core systems I can go ahead and throw in a City of Heroes session and not notice a slow down anywhere except the MPEG4 recode which, gosh, will complete in 30 minutes instead of 15.

CPUs seem plenty fast to me.

 

[sechs]

So, all we really need CPUs to do is get cheaper and more efficient, and for more crap to get crammed into the package.

When are on-die GPUs supposed to be here?

 

[Mercutio]

When are on-die GPUs supposed to be here?

I've read that AMD is working on integrating an ATI GPU in future generations of its processors, as a replacement for one of the general-purpose cores found on current chips.

I'm not sure that's the right way to go. I have to think that an add-in GPU is always going to be a lot more advanced.

The other part of the problem is that it's not CPUs that actually need to get faster. RAM is what needs to get faster. Right now we have in our best case scenario, 3000Mb/s Hypertransport point-to-point links between each CPU core and RAM, which I think is still the best case scenario for I/O (Supercomputers are still mostly made with Opterons). Even decoupled from the FSB speed, there's still going to be a massive number of idle cycles on each core of any CPU. The goal isn't just improving parallel execution - that makes the CPUs work better - but also keeping CPUs fed with the data they need to operate, which has been a problem since Intel started clock multiplying 486s.

 

[Fushigi]

Since you mentioned supercomputers, MB/s and GB/s are no longer the measure. Try 1.3TB/s.

The Power 595 supports up to 4 TB of memory per server, twice as much as the HP Superdome and Sun SPARC. Continuing the IBM tradition of innovation with the Power Architecture® systems design, the Power 595 also supports four memory operations per cycle and an aggregate memory bandwidth of more than 1.3 TB/sec (terabytes per second), which is enough to transfer in each second the amount of information printed on the paper made from 50,000 trees.

 

[Mercutio]

That's an aggregate number. I'm sure each system in the supercomputer is closer to the scale I mentioned.

 

[Fushigi]

Divide it by the 64 cores and you're still at 20+GB/s/core. And not only does it run IBM i (current name for OS/400) and AIX, it runs both Power Linux and via VM it can run x86 Linux.

 

[ddrueding]

I found another thing that clobbers my 3.6Ghz Quad core. The terrain mod TileProxy for Flight Simulator X. It basically replaces the terrain textures in the game with high resolution ones streamed and converted in real-time from one of the online mapping sites (Microsoft, Yahoo, etc). It will keep all cores pegged for 10 minutes before I start flying, and will use any available cycles after that to try and keep up with my flight!