Apple releases dual 90 nm (66 mm2 die size) 64-bit G5 2.0 GHz Xserve 1U with ECC support *Pix*

[hungrygoose]

hey
drag.....question....do you not seem to realize that processor power does make a difference?....the main need for SMP can be overshadowed by CPU clock and bandwidth capabilities.....you seem to have all your ideas based on what apple says is the best on paper.....if you have a dual 1000mhz "imaginary smp cpu" and you compare it to a 2000mhz athlon.....which do you think will perform the tasks you mentioned(video encoding/decoding) faster?.....the smp will not matter if the computing power of the other processor is that much better.....sure, if u compare one cpu to dual cpu of the same type....then yea, dual matters......but that's not the case.....ur comparing dual 2ghz to an almost 4ghz p4 running HT at 800mhz fsb....using the same speed ram as the g5.....how the hell do see that the g5 will outperform in these type applications?????......paper ideas lead to zealots.....logic leads to the truth........basically, it doesn't HAVE to be good just b/c apple says it is!

 

[jhu]

Dual Opterons running on linux would obliviate a G5... Apple has been touting the G5 as THE fastest for months.. They are great at paper launches. If it wasnt for AMD and IBM, Apple wouldnt even HAVE the technology for the G5.

Intel Vs AMD zealotry can be dealt with, but when Apple zealots appear, it just makes the whole discussion in a whole, assanine. Go back under your rock and get off Steve Job's nutsack.

obliviate? that's not even a valid word. apple touting g5 as the fastest? no way! a company trying to promote their own product as better than someone elses? that's blasphemy! look at how intel is trying to promote their latest p4ee as the fastest. but on closer inspection their itanium 1.5ghz is much faster.

how does amd figure into apple getting a g5? in the beginning, powerpc was designed by a collaboration of 3 major companies: ibm, motorola, and apple. as a corollary, if it wasn't for texas instruments, sun wouldn't have their ultrasparc iii.

intel vs amd zealotry can be dealt with? give me a break. why don't you get off craig barrett's and hector de ruiz's nut sac?

 

[HokieESM]

Originally posted by: Eug

BTW, today Apple released Xgrid 1.0. (Clustering app, for their new G5 Xserves.)

Eug, I wonder where Xgrid comes from? <big grin> I don't think its actually Apple.... oh yeah, its been running for the past few months at a school in the Appalachians. From what I hear, Xgrid is just a "flowery" version of what VT is running on System X (I'm SO upset that McDonalds denied them from calling it "Big Mac"), albeit slightly less optimized.

For the Mac haters.... I'm not a Mac fan. I don't currently own one (although I briefly did own a PowerBook). But they ARE a legitimate alternative now. The G5 is a nice little machine--albeit slightly slower than the Athlon64 and P4/Prescott. We should all welcome competition--not to mention, maybe some enterprising company will take it upon themselves to make an OSX clone for x86... or x86-64. OSX is BY FAR the easiest to use Linux/Unix variant I've touched.

 

[NFactor]

The G5 has been nowhere for benchmarks and whenever they pop up, like the PCWorld and Macworld ones, you Mac zealot fanboys are quick to jump on the train calling the benchmarks BS. Apple sits there probably crippling boxes and putting useless most likely fake benchmarks up on their site which no one is able to reproduce and uses them to call the G5's the fastest computers in the world.

I will be the first to tell you that i like OS X. In fact I like it a lot. But I will never use it because Apple needs to get their heads out of their butts and realize that they own something like 2% total marketshare and that number is not going up. The G5 is a plenty fast computer, agreed, it is significantly better than that load of sh*t they called a G4. But it is nowhere near as fast as any high end PC. A single AMD64 chip could take it, the P4 3.2 and EE could take it, and of course, the Opteron, Zeon, and Itanium's would destory it. The only benchmarks to the contrary are on Apple's site. When Apple gets the balls to send a G5 or XServe out to a proper hardware review site and have them detail their benchmarks then I will admit that I have been wrong.

 

[Ardan]

That looks like a pretty nice server! I don't own Apple Computers myself (I don't hate them, I just simply don't have one) but I always enjoyed using them. Where I am going for more advanced Networking classes and other things, like programming classes and database design/management, they have those G5's in the Library (along with 2.4Ghz Pentium 4s from Dell) and I always use them. I would say it seems more responsive, and I just simply enjoy looking at the OSX interface more than Windows XP's interface. They are both very good platforms...its a shame people blindly knock one down (and obviously forgetting how horrible Windows is and was) without having used it.

Speaking of Windows, anyone remember the Windows 95 commercial with that Rolling Stones song in it? I always remember something I read in a PC World magazine a long time ago, asking why they didn't include the lyrics "you make a grown man cry" in it .

 

[drag]

Originally posted by: hungrygoose
hey
drag.....question....do you not seem to realize that processor power does make a difference?....the main need for SMP can be overshadowed by CPU clock and bandwidth capabilities.....you seem to have all your ideas based on what apple says is the best on paper.....if you have a dual 1000mhz "imaginary smp cpu" and you compare it to a 2000mhz athlon.....which do you think will perform the tasks you mentioned(video encoding/decoding) faster?.....the smp will not matter if the computing power of the other processor is that much better.....sure, if u compare one cpu to dual cpu of the same type....then yea, dual matters......but that's not the case.....ur comparing dual 2ghz to an almost 4ghz p4 running HT at 800mhz fsb....using the same speed ram as the g5.....how the hell do see that the g5 will outperform in these type applications?????......paper ideas lead to zealots.....logic leads to the truth........basically, it doesn't HAVE to be good just b/c apple says it is!

It's not good because Apple says so. it's good because it's good. The dual PowerG5 setup is perfectly capable of keeping up with any computer aviable from any OEM manufacturer in the same price level.

Now If I compare Apple's current offerings to CPU and stuff that just got released and reviewed this month, sure they are going to loose out.

That's just how the computer industry works. When Apple was saying that the power g5 was the fastest computer aviable, IT REALY WAS THE FASTEST COMPUTER AVIABLE AT THAT TIME. Technology moves on. The fastest computer this month isn't going to be the fastest computer 3-6 months from now.

But how long is the AMD64 3400+ going to be fastest affordable computer chip from AMD.

I can say conformatbly that a computer using AMD64 3400+ with all the niceties is the fastest computer you can get right now for under 2500-3500 dollars. Now a few months from now when some guy compares a pentium 4.0ghz to it and shows that the pentium is faster is the AMD64 now a POS and I was a lier!?

Look. Motorola's G4 got raped in proccessing power from x86 running at the same price level.

But the G4 has little to do with the current Power970's aviable from IBM. They actually keep up with Pentium4's and Athlon chips quite nicely and have no performance deficit. In fact in some cases they are actually quite a bit faster then Intels/AMD's. Just like AMD is faster then Intel in some cases and Intel is faster then AMD in others.

The point of a SMP setup is that it WILL be faster for multitasking then any comparable single proccessor setup. The Dual G5 is designed specificly to be consumer level SMP machine, which ironicly is mostly unavaible from any other OEM computer builder for consumer level stuff.

We aren't comparing 2 1000mhz cpu to 1 2ghz cpu. (To compared straight ghz is retarded anyways). We are comparing 2 2ghz IBM cpu's vs 1 1.8-2.2ghz AMD/IBM cpu vs one 3.0+ghz Pentium CPU. In that light the G5 does fairly well.

PLus even though proccessing power is important, it isn't everything. The power g5 is whisper quiet and stable as a rock. It has all the newest technology from everybody. (64-bit PCI slots, Sata drives (even raid setups), optical high-end audio, USB 2, both firewire 400/800 gigabyte network, etc etc) And in a technical sense it is legacy-free. Unlike the x86 setup it is pretty much a new design from the ground up.


Anyways. I haven't yet decided weither my next computer is going to be a dual g5 setup or a dual Opteron setup. Very little of the time I need raw proccessing power, What I want is the ability to multitask with the best performance. (My favored OS runs well on BOTH x86 and PowerPC setups.)

And the next gen of 3.0+ powerg5's is going to be a pretty atractive computer. Hopefully IBM will have some 4 or 8-way Power970 workstations to choose from.

Now if I was a gamer the choice would be obvious. Pentium 4 all the way, Windows XP is king. But that not what I want.

 

[Eug]

Originally posted by: NFactor
The G5 has been nowhere for benchmarks and whenever they pop up, like the PCWorld and Macworld ones, you Mac zealot fanboys are quick to jump on the train calling the benchmarks BS. Apple sits there probably crippling boxes and putting useless most likely fake benchmarks up on their site which no one is able to reproduce and uses them to call the G5's the fastest computers in the world.

I will be the first to tell you that i like OS X. In fact I like it a lot. But I will never use it because Apple needs to get their heads out of their butts and realize that they own something like 2% total marketshare and that number is not going up. The G5 is a plenty fast computer, agreed, it is significantly better than that load of sh*t they called a G4. But it is nowhere near as fast as any high end PC. A single AMD64 chip could take it, the P4 3.2 and EE could take it, and of course, the Opteron, Zeon, and Itanium's would destory it. The only benchmarks to the contrary are on Apple's site. When Apple gets the balls to send a G5 or XServe out to a proper hardware review site and have them detail their benchmarks then I will admit that I have been wrong.

Itanium? Yeah probably, assuming can find software to run on it.

Opteron? Depends on the app.

Xeon? Depends on the app.

I think a dual G5 2.0 is about on par with a dual Xeon 2.8-3.0 GHz, depending on the situation. For gaming a G5 will be destroyed by the Xeons, but for other apps it really depends. I've already seen lots of benches out there supporting this claim, and I've already posted the benchmarks here. I doubt Apple will send their computers to AT for review however, since quite frankly, AT customers generally aren't interested. And Mac users don't generally read AT either.

"Nobody will ever need more than 640k RAM! (Bill Gates, 1981)"

AFAIK, Bill Gates never said that.

 

[MadRat]

Originally posted by: Eug
New G5
Die size: 66 mm2
Transistors: 58 million
Process: 90 nm, Silicon on insulator
L1 instruction cache: 64 KB
L1 data cache 32 KB
L2 cache: 512 KB

Eug, why don't the i- and d- caches in the L1 match in size? What effect does it have on performance when using the non-symmetrical sizes like this?

 

[Eug]

Originally posted by: MadRat

Originally posted by: Eug
New G5
Die size: 66 mm2
Transistors: 58 million
Process: 90 nm, Silicon on insulator
L1 instruction cache: 64 KB
L1 data cache 32 KB
L2 cache: 512 KB

Eug, why don't the i- and d- caches in the L1 match in size? What effect does it have on performance when using the non-symmetrical sizes like this?

<Bones' voice>

"Dammit Jim! I'm a geek, not an engineer."

Maybe someone else can answer this question...

By the way, there is a discrepancy on the http://www.apple.ca website. It says 2.3 GHz. Last time this happened, it was "premature specification", and the thing was announced shortly afterwards.

 

[addragyn]

Originally posted by: Schadenfroh

Originally posted by: Eug

Originally posted by: dexvx
Where does it say its 90nm?

Check Page 15 of the PowerPC G5 January 2004 White Paper or else the "What's New?" section on Page 3 of the G5 Xserve Tech Overview.

how? that process is not mature

Via's decision to partner with IBM was based on the company's ground-breaking silicon manufacturing technologies, such as copper interconnects, silicon-on-insulator (SOI) and low-k dielectric insulation, together with its advanced 90-nanometer (nm) process.
link

 

[drag]

Ya the Power4/4+/5 series IBM proccessors are some sh*t.

Every new technology you see in the G5 and the AMD64 first appeared in the power# series proccessors (such as SOI transistors that AMD touts so much). Interesting weird stuff like dual core proccessors and SMT support (HT to you pentium people. )

Right now the Power5 will be O.13nm manufacturing proccess. They are working on fabbing for the 0.09nm versions. And are working on developing the design for a 0.065nm design proccess!

That's pretty hardcore for a proccessor with a 276million transistors and is designed to run reliably for 24/7 operation for 5+ years (easy)!! (opteron is lik 106million transistors.)

One of the cool things about the .09mn power970 proccessors is that it could point to a desire on Apple's part to use these proccessors in their laptops and 3+ ghz dual proccessors power g5's.

(that's one of the reason's like linux so much. I can run on itanium, AMD64 (in native 64bit mode even right now), 32bit x86 proccesors, Motorola g4's, IBM power970. Even sun sparcs, IBM mainframes, and alphas are open to me. It may require more work to run on non-x86 hardware (since x86 the most popular), but it can be well worth it.

Now I only hope that the intaniums and current Intel offerings can keep up the pressure and the performance for x86. Competition gives me fast proccessors for cheap.

 

[Sunner]

Originally posted by: MadRat

Originally posted by: Eug
New G5
Die size: 66 mm2
Transistors: 58 million
Process: 90 nm, Silicon on insulator
L1 instruction cache: 64 KB
L1 data cache 32 KB
L2 cache: 512 KB

Eug, why don't the i- and d- caches in the L1 match in size? What effect does it have on performance when using the non-symmetrical sizes like this?

That seems to be more of an x86 thing really.
The Crusoe for example has something like 64/32, as does the UltraSPARC III.
PA-RISC's tend to have huge L1's, on the order of over one MB, and IIRC they're not split evenly either.

By the way, Eug IIRC Prescott is a bit over 100M transistors, more like 115 or 125, and a bit over 100 mm2.

 

[Eug]

Originally posted by: Sunner
By the way, Eug IIRC Prescott is a bit over 100M transistors, more like 115 or 125, and a bit over 100 mm2.

You sure? I saw that too on a few sites, but then again I saw the 109 million and 81 mm2 elsewhere.

Anyways, if true, it does seem like the G5 90 nm is gonna be cheap. The L2 is 512 KB only though. How much space would another 512 KB take?

 

[drag]

Originally posted by: Sunner

Originally posted by: MadRat

Originally posted by: Eug
New G5
Die size: 66 mm2
Transistors: 58 million
Process: 90 nm, Silicon on insulator
L1 instruction cache: 64 KB
L1 data cache 32 KB
L2 cache: 512 KB

Eug, why don't the i- and d- caches in the L1 match in size? What effect does it have on performance when using the non-symmetrical sizes like this?

That seems to be more of an x86 thing really.
The Crusoe for example has something like 64/32, as does the UltraSPARC III.
PA-RISC's tend to have huge L1's, on the order of over one MB, and IIRC they're not split evenly either.

By the way, Eug IIRC Prescott is a bit over 100M transistors, more like 115 or 125, and a bit over 100 mm2.

Maybe the smaller data cache to instruction cache is a trade off. Instead of trying to even them out or increase the costs to match them.

Because if you think about it for most tasks you are going to be repeating the same tasks over and over again, however your going to be applying the instructions to continously changing data, so you get better performance from caching the instructions more then you'd get from caching the data.

And plus maybe a smaller cache is quicker to clean out and replace with new info?

I remember that with the playstation2 vs a PC that in the PC you have big caches and large amounts of memory everywere. Like in the video card you have a large amount of memory to store textures and stuff like that, because to read the textures all the way from main memory or the harddrive is a slow and high-lantancy thing. Playstation2 on the other hand is designed to get information in and out quickly, you don't need large amounts of video ram (other then the fact your running a 640x480 resolution) "very large pipes and very small buckets vs small pipes and large buckets". So you just move information around faster, that way you get much better 3-d performance/effeciency vs a PC which has to be usefull for a wider range of tasks. (the downside is that you have to make textures fit with in a certian size of cache/ram or you end up having big performance drop)

So since that the cache would be designed a lot like that... Instructions get used repeatedly, so you want to make that as big as possible. Data needs to be moved in and out fast so that you don't need the big cache...


Just a Idea.

 

[drag]

Oh ya and the other half of my crappy theory.

The reason of the x86 why you want a semetrical data/instruction data set is because of the ISA.

ISA not a hardware thing so much, but more of a standard on how to interatact with hardware. When they designed the x86 ISA it dictated how the machine was expected to intereact the software.

Now just as long as designers make the hardware APPEAR to the software be x86 ISA-compliant you can make the hardware anything you want. You add instructions, change things, change how the memory works, how many cpu's you have. You just have provide the extra compatability layer...

Like the transmeta cursuoe stuff. It's instruction set is incredably different from a pentium setup. However they have a extra code morphing chip (or whatever) that translates and cache's the the x86 instructions and optimizes them to work on the radicly different chip.

(code morphong seems like it would be slow and painfull, but often it actually increases the performace because you can tailor and optimize the instructions to suite specific hardware and then you cache that info and just re-use each time that same byte of code is used)

Now a pentium chip has to be x86 ISA compilant, right?

In x86 ISA you have a grand total of 24 registers. (8 GPR, 8 FPR, 8 SIMD).

Now the pentium4 has 128 GPR registers alone!! (PowerPC ISA specs 32 GPR registers)

Registers are the part of the proccessor were data and instructions are fed into to get used...

So Pentium has 128 registers, however programs only see 8. So what it does is dynamicly reasigns the 8 registers and maps them to the different 128 to utilize the proccessing power. That way it can use it's long pipeline start proccessing 8 peices of information, remap the 8 ISA registers and then start proccessing on the next 8 peices at the same time and so on.

So I figure that the data in the data cache has to wait for the 8 ISA registers to get reasigned to the 128 registers that physically exist.

So you have a slight pause. That's why I figure that you want the data and the instruction caches to be the same size in x86. Not because it's great to have them the same per say. But because you need a larger data cache to make the proccessor more efficiant.


Of course this is all just speculation.

 

[Eug]

Originally posted by: Eug

Originally posted by: Sunner
By the way, Eug IIRC Prescott is a bit over 100M transistors, more like 115 or 125, and a bit over 100 mm2.

You sure? I saw that too on a few sites, but then again I saw the 109 million and 81 mm2 elsewhere.

Anyways, if true, it does seem like the G5 90 nm is gonna be cheap. The L2 is 512 KB only though. How much space would another 512 KB take?

Sunner, you were right. See this :camera:.

112 mm2 for Prescott, and 125 million transistors.

ie. Prescott is exactly twice the die size of the G5 90 nm, and has over twice as many transistors. I wonder what the comparative manufacturing costs and heat outputs are.

 

[BFG10K]

Still no 64 bit OS from Apple.

 

[Eug]

Originally posted by: BFG10K
Still no 64 bit OS from Apple.

True, but at lleast the OS will address up to 16 GB on current hardware (limited to 4 GB per app). There's a beta Linux 64-bit that runs on G5s, too.

However, that's a different market of course. If you're in need of true 64-bit enterprise hardware, G5 Xserves are not for you.

 

[Sunner]

Originally posted by: Eug

Originally posted by: Eug

Originally posted by: Sunner
By the way, Eug IIRC Prescott is a bit over 100M transistors, more like 115 or 125, and a bit over 100 mm2.

You sure? I saw that too on a few sites, but then again I saw the 109 million and 81 mm2 elsewhere.

Anyways, if true, it does seem like the G5 90 nm is gonna be cheap. The L2 is 512 KB only though. How much space would another 512 KB take?

Sunner, you were right. See this :camera:.

112 mm2 for Prescott, and 125 million transistors.

ie. Prescott is exactly twice the die size of the G5 90 nm, and has over twice as many transistors. I wonder what the comparative manufacturing costs and heat outputs are.

Damn, that Dothan is like 3/5 cache.

 

[Eug]

:camera: here.

Damn, that Dothan is like 3/5 cache.

I wonder how much bigger the G5 die size would be with 1 MB L2 cache.

 

[Baronz]

I wouldn't be first in line to get an Apple server.

 

[Eug]

Originally posted by: Baronz
I wouldn't be first in line to get an Apple server.

Hey neither would I, but it looks like a good piece of hardware.

 

[Go3iverson]

I've been waiting for the G5 XServe. We're considering placing a Dual 2.0GHz XServe with a 3.5TB XServe Raid. If we get it, it'll be sweet!

I've been very impressed with Panther and the increased compatability with the Windows servers we currently have.

 

[beachbreeze]

Originally posted by: nanyangview
rotten apple POS at work again
why can't they just die?

so is the IPOD (Icrap)
the Creatize Zen, thou bigger, beats it hand down. I went to see the ICRAP, have fun scratching the back, which is made of SHINY ALuminium.

Apple = POS everything

nanyangview your spittle is really going to ruin your keyboard.

 

[beachbreeze]

Originally posted by: zephyrprime

it's restrictive cpu bandwidth

What are you talking about dude? All the platforms have exactly 6.4GB/s of "cpu bandwith" because they all use the same sort of dual DDR 3200 memory subsytem.

I believe he is referring to the G5s 1GHz FSB