Wow .. that is impressive
Wow .. this is going to be a killer chip. Should beat Nehalem by quite a large margin.
IBM has divulged some specs of its forthcoming Power7 chips and their related Power Systems servers, throwing down the gauntlet to its peers. The company confirmed last fall that the Power7 chip would span up to eight cores and would use a 45 nanometer manufacturing process developed at IBM's East Fishkill, New York foundry. …
Perhaps it was a typo, or perhaps it is a breakthrough. Both are possible. But if we're really talking offchip DDR bandwidth of 100GBytes/s from TWO DDR3 controllers, let's see what that would mean.
There may be 1600MHz DDR3 widgets out there. (1.6 G transfers per second.)
Each port contributes half the bandwidth, or 50GB/s. 50/1.6 is 32 bytes per transfer, more or less.
That means that each DDR3 port is 32 bytes wide. Hmmmm... That's something like 500+ pins devoted to offchip DRAM data path.
Quite within the realm of possibilities. Though one should note that "sustained bandwidth" is often far less than "maximum bandwidth". In this case, few DDR3 complexes get sustained bandwidth more than 70% of peak. (Take a look at the Nehalem stream triad scores.)
100GB/s is a good target. Achievable with heroic efforts. And the Power-X series has always been about heroic and extremely competent efforts. Please tell me it isn't a typo. ;)
"IBM is one of the few companies left that makes its own chips and servers that wrap around them, and unless something radical happens, Big Blue will continue to do so into the foreseeable future"
If I'm not mistaken, hardware is one IBM's least profitable areas? With IBM's focus on Software and Services, not to mention their current fanaticism over expense reduction, who want's to wager IBM will jettison this unit at some point or at the very least offshore the R&D?
Reworking the pipeline on the Power 6 had "less than expected results"?
The STRETCH, or IBM 7030, from back in 1961, was the first IBM machine to offer what was then called "superpipelining", where the execution stage of instruction processing was broken into multiple parts: the 7090 could already overlap fetch, decode, and execute. Because this didn't provide as much performance improvement as expected, IBM stopped making that computer, and gave a reduced price to those who had it on order.
Much later, IBM came out with two higher-end machines for its System/360 line, the Model 85, which had cache memory, and the Model 91, which was pipelined. Cache memory turned out to offer a better than expected performance improvement, so the significantly cheaper Model 85 was almost as good as the model 91. In 1969, IBM added cache to the model 91, yielding the model 195, something of a precursor to the Pentium, which also had both cache and pipelining (the 5-volt Pentium had a cache that was only half as large as that of the 195, but it ran at 66 MHz instead of 16 MHz).
And now, the Power 6. IBM seems to be destined to be forever haunted with a curse on its pipelining efforts.
Looks like the world will be left with Nehalem and Power7. Nahalem with the low end 4 socket infrastructure business (8 sockets requires the disastrous two hop) and Power7 for the scale up data centric business.
The world is getting pretty simple....cheers from the UK
Ah yes lets all get excited about a new fast chip (for which we only have IBMs word that its fast in anything we care about) - cos of course "raw CPU performance" is where we have all our headaches in servers these days isn't it? The weird thing is if the server market was like the auomobile market we'd all be making our purchasing decisions for our next car based on the Formula 1 teams scoring...
IIRC an "in-box" upgrade from P5 to P6 meant you kept the power switch from your old P5 box (cos thats where the serial number was) - everything else got exchanged - so before commiting to a P6 box that IBM tell you is upgradeable to P7 ask IBM about the details of the process and what sort of outrage you can expect (plus how much new kit you'll have to buy)
of course what I meant was:
"if the automobile market was like the server market we'd all be making our purchasing decisions for our next car based on the Formula 1 teams scoring..."
not the other way round,
And I was really trying to type "ask IBM what sort out outage you can expect" not "what sort of outrage you can expect" - although myabe the typo will be closer to the truth!
To upgrade the p595 (up to 32 socket / 64 core system / 128 threads) you simply use partition mobility to move applications to other servers without any outage, then you pull out the books and push in the new p7 books. You then use partition mobility to move the applications back to the p7 based p595.
Now you have a system with up to 256 cores and simultaneous 1024 threads.
If you had read the article, you would have noted the "expected" arrival of the first Power7 servers as below:
"....Power7 machines are just around the corner in early 2010...."
Please note that's "expected" as IBM have not released a firm date yet, so "early 2010" could translate to early adopter kit by June 2010, with general availability a lot later, and some models not until 2011 or later. It also only mentions the first Power7 servers, which leads me to suspect we'll see the same laughs as we saw with Power6, where IBM released one server model with the new CPUs and then tried to claim it had a Power6 server range.....
Whilst Rock is almost certainly a dodo, Tukzilla is still on course for 2010 according to people at HP I trust, so the idea that Power7 will allow IBM to sow up the market before Tukzilla gets in is, frankly, just whimsicial. Having said that, the spec for Power7 is definately very appealling, and I especially like the fat pipes out to memory. It's not hard to see why Rock got canned if Sun had intelligence on what Tukzilla and Power7 are bringing to the party!
Matt, don't try to fool yourself...
El reg readers are way too exhausted of your misunderstanding of the news.
If you still believe what HP says about Itanium, fine, but keep it to yourself.
After three or four delays on Tukzilla, nobody believes it will ever get to market.
Intel 'expected' Tukwila on 2007, 2008, 2009 and now 2010.
Even if it gets to market on 2010, I believe POWER7 and Nehalem will leave no space to EPIC boxes.
Let them come, and then you can talk on real facts.
In the enterprise space it is Power and Itanium. Unfortunately, for HP and subsequently Matt, Intel does not seem to give a rats ass about Itanium. The odds are Power7 will come out before Tukwila. It is also disheartening that there will not be core performance improvement. Intel is only claiming to have twice the performance per chip which has twice the cores. Not very exciting for anyone paying for software on a core basis and very scary for anyone paying for a Oracle product as they will increase the core multiplier for Tukwila to .75. 50% more cost for the same performance per software license.
It also blows my mind that Intel will bring out Tukwila as a 65nm product when everyone else moving from 65nm, including Intel, to 45nm. Intel did a huge press release event about Tukwila being the first 2B+ transistor chip....will they have to do retraction since Nehlem EX will be first in the market with 2B+?
The other thing that gets me is how HP claims to have exited the chip business but also claim they have the best chipset for Itanium with the SX2000 and the future Windjammer. It seems every other vendor is dropping Tukwila support because they dont want to compete with proprietary chipsets.
I wont buy HP products anymore since they put chips in their ink supplies to prevent refills. HP may claim to be Green, but look at the landfills littered with perfectly usable cartridges which HP chips have disabled.
Matt...appreciate the positive Power remarks...come on over the water is fine :-)
Cheers from the UK
Oh dear, are you saying that HP Itanium boxes have a "proprietary" chipset? Hmmm lets have a look at what's in a Power or SPARC box (and sun guys don't tell me just cos some of the specs are in public domain its not proprietary in every other sense!)
Whilst we're at it, why don't we take a look at the chipsets in a IBM x-series box (you know, those x86 systems that are supposed to be standards based?)... pot - kettle -black.
>> Not very exciting for anyone paying for software on a core basis and very scary for anyone paying for a Oracle product as they will increase the core multiplier for Tukwila to .75. 50% more cost for the same performance per software license
Care to substantiate this claim? With that kind of FUD you kind of give away where you work!
>> I wont buy HP products anymore since they put chips in their ink supplies to prevent refills.
You are of course entitled to your opinions, but if you gave that reason to *my* CFO or CIO for not considering a product range, you wouldn't stay in your job very long.... I may as well drag up all that "IBM helped the Nazis" cr*p as a reason for not buying IBM.
"Specifically, a top-end Power7 server will have 32 sockets and 360GB/sec of SMP bandwidth per chip linking them together into a shared-memory system."
Actually I wouldnt trust these numbers at all. Earlier when IBM claimed that the Power6 had 240GB/sec bandwidth or so - it turned out that IBM added all bandwidth in the CPU! That is clearly wrong and uneducated to do. If there is a bottleneck at 10GB/sec, then the CPU's bandwidth will never be higher than 10GB/sec. It is funny people here dont realize it. Maybe they are uneducated or drink the kool aid without asking?
Also, the Niagara T2 is four times as fast as a Power6+ in certain server benchmarks. This Power7 will apparantely be four times faster than a Power6+. Then it should match one Niagara T2. And then, IBM will not need three Power servers with 12 Power6 CPUs to get half the performance of one Sun T5440 box! That is good.
The problem for IBM is that the T3 is on the way, and it will smoke the T2, hence it will kill the Power7. The same relation between Power6 and T2 today, we will see between Power7 and T3 soon. The T3 will probably be four times as fast as the Power7 on certain server workloads. It is funny IBM needs 32 Power7 to reach 1024 threads, whereas SUN needs 4 Niagara T3 to get the same amount of threads. I bet a 4 Niagara box will be much cheaper than a 32 Power7 box. Anyone willing to bet against me? Could the IBM box cost 50 times as much as the SUN box? Or 100 times? What do you think?
Man, do you realized what you just wrote?
Don't compare a low-end processor with a high-end one.
If T2 or T3 goes beyond 4 sockets than you can compare them with POWER.
It is fair easy to create an architecture with limited scalability eg. Nehalem, the real challenge is to go scale beyond.
If you want a fair comparison apple to apples, compare you so called King of the Hills T3 with Nehalem EX when they GA next year.
By the way, if you love so much benchmarks try to compare current T2 with Nehalem EP which scales about the same.
If the T2+ is that good then why does Sun still try to sell us M-class boxes with Fujitsu chips for any DB or I/O workload?
I would ask if those T3 threads are simultaneous or if they are round robin KISS ass treads.
SPARC and Itanium are dead. Squeezed out between Nehalem and Power.
As I recall Pony tail boy claimed the T2 was 9.6GHz because he tried to multiply 1.2 * 8 cores.
Pot/Kettle/Black....join us or die ugly
Cheers from the UK
do you realize what you just wrote?
"....Don't compare a low-end processor with a high-end one...." Why not? The high-end Niagara is several times as fast as the lowend Power6, while Niagara uses 1.6GHz and the Power6+ uses 4.7GHz! If that is not low-end performance from the Power6+, then I dont know. Admit it, the Power6+ is slow, uses lots of energy and is really expensive. This IS true, and nothing you say can change it. The Niagara is several times faster in certain server-client benchmarks - noone can deny that. The Niagara uses 1.6GHz, and the Power6+ uses 5Ghz, noone can deny that. It is more expensive, no one can deny that. Admit it.
"....If T2 or T3 goes beyond 4 sockets than you can compare them with POWER...." Whoa. Let me tell you, your logic is FAIL. Can you explain the point of T2 must go beyond 4 sockets? No? Then dont talk about this. (The point of using lots of sockets, is to get more performance. But Niagara packs plenty of performance already, 4 of the Niagaras can match 16 Power6+. If one CPU is as fast as 32+ Power6 then I dont see why that CPU is worthless because it only uses one sockel. 4 of the Niagara provides more performance than 16 Power6+, hence you dont need more than 4 of those. FAIL LOGIC)
"...It is fair easy to create an architecture with limited scalability eg. Nehalem, the real challenge is to go scale beyond...." True. But if that architecture already kills all performance with one socket, then I dont deem it worthless.
"...By the way, if you love so much benchmarks try to compare current T2 with Nehalem EP which scales about the same...." Why dont you want me to compare T2 against the slow Power6 instead? Is it because you know that the Power6 will loose big time? (Which it does)
"....If the T2+ is that good then why does Sun still try to sell us M-class boxes with Fujitsu chips for any DB or I/O workload?..." The T2+ is that good. On certain work loads. It is not suited for all work loads. This is no secret and everyone knows the T2 sucks on some work loads.
"...I would ask if those T3 threads are simultaneous or if they are round robin KISS ass treads...." Take a wild guess. Is it possible to achieve extreme through put only with round robin?
"...As I recall Pony tail boy claimed the T2 was 9.6GHz because he tried to multiply 1.2 * 8 cores...." Now, if that were true, then the T2 would be several times as fast as the highest clocked CPU, right? But... how can the T2 be several times as fast as Power6+ - which it is? Hmmm... Schwartz must be correct in some regard, or how can you explain the extreme through put of the T2? If Schwartz lied, then the T2 would be really dog slow.
Now tell me, is the T2 dog slow, or is it several times faster than the highest clocked CPU? Is Scwartz correct, or is he lying? (Hint: study the architecture of the Niagara and you will see why Schwartz said so. Then you will understand why the T2 doesnt need a large cache. And why it is revolutionizing the old desktop CPU model with high clock speed and large cache).