Intel Dual Core Xeon Benchmarked

An anonymous reader writes "A few weeks back, Intel launched a new dual core chip with little applause. It appears we know now why, as the chip has been benchmarked by the chaps at GamePC. In tests against the dual core AMD Opteron processor, Intel's new chip gets thoroughly thrashed, losing out in terms of raw performance while eating a lot more power. "

Comments

AMD's dual cores are great

[jarich]

I just bought my wife a dual core (3800 model) and it's just as responsive as my dual Opteron. I'm seriously considering selling my dual CPU box and getting a dual core myself just to have fewer fans in the box and generate less heat.

I had been considering an Intel dual core but it sounds like I need to aim for an AMD instead.

Re:AMD's dual cores are great

[DrSkwid]

A *real* geek would get a dual dual core ;)

Re:AMD's dual cores are great

[Jherek Carnelian]

They aren't real dual-cores hence the abysmal performance

No, they are honest-to-goodness real dual cores. Two fully functional cpus in a single socket.

The problem is that the socket only has enough memory bandwidth for one cpu's worth of work. So, even if you double the number of cpus, you still can't shovel the data in and out fast enough to keep up with the work being done. Thus one of the two cores is almost always stalled out waiting on memory.

The AMD chips have got more memory bandwidth, so they can keep both cpus fed with data reasonably well.

Re:AMD's dual cores are great

[Eukariote]

The AMD chips have got more memory bandwidth, so they can keep both cpus fed with data reasonably well.

Not just that. The AMD dualcore chips have an on-chip connection between the cores: both cores share a crossbar fronting the memory controllers and have the on-chip equivalent of a coherent HyperTransport connection. So, you see, the AMD design is in fact a real dual-core design. The current Intel dual-cores, on the other hand, share nothing on-chip.

Re:AMD's dual cores are great

[quarkzone]

The problem is that the socket only has enough memory bandwidth for one cpu's worth of work.

This is exactly right. It is really surprising that Intel has focussed so completely, almost obessively, and for so long, on the problem of supplying the maximum number of work-cycles per unit of time (GHZ, Pipelining, Itanium's EPIC design) while seemingly paying so little attention to supply-of-work-to-do (FSB speed and architecture)

AMD has paid quite a bit of attention to the work-supply and has a much more efficiently balanced work-cycle-supply/ data-for-work design. http://www.hypertransport.org/ gives AMD a big leg-up over Intel.

If Intel fails to do something spectacular to FSB speeds, AMD is sure to continue to pull away from Intel. The more cores and threads per CPU, the greater AMD's lead over Intel will become (at least from a performance point of view), until Intel addresses this problem.

Stupid pre-retail release

[dada21]

This release seems dumb for Intel. No optimized motherboards, outrageous power requirements and a really inefficient core? It isn't even alpha-release worthy. Why would Intel release a product that is just waiting for a poor review? Is the high end market that hungry?

The article didn't need 15 pages to explain Intel's mistakes. Intel will lose more customers to AMD than if they had waited until they had a viable and competitive product.

400W while idling? For sub-standard performance? Yay.

Re:Stupid pre-retail release

[Ihlosi]

Is the high end market that hungry?

No, the high-end market is waiting for something that has "Intel" and "Dual Core" written all over it. Everything else is irrelevant.

Power Consumption

[matr0x_x]

This is the 21st century in North America, since when do we care how much power a CPU uses. *Drives away in Hummer*

Bah

[Red Flayer]

Got through several pages of the benchmarking before it appeared /.ed.

First concern is that though the chip has been released, motherboards configured for it aren't close to release yet. I'd rather see it benchmarked as distributed, since that's what really matters to the end user.

Second concern is power usage and heat production. If you can't make a chip as powerful as your competitors, you better make sure it is not as expensive to operate. Really, why would someone choose to use a chip that is less powerful, intrinsically costs more to operate, and costs more to cool? Chips are cheap enough that the operating costs are often now more expensive than initial cost.

Is GamePC really Intel's target here?

[hal2814]

Is GamePC the best place to read benchmarks on a dual core Intel Xeon chip? The article appears to be /.ed already (or just REAAALY slow at my end) so I can't read the results, but I can't help but think somewhere called GamePC isn't exactly Intel's target audience here.

Coral cache link

[Freggy]

http://www.gamepc.com.nyud.net:8090/labs/view_cont ent.asp?id=paxville&page=1

Seems it's slashdotted already after 8 posts. Finally when will all slashdot-links be coralized automatically?

Stock market

[13bPower]

Hello, Stock market? Please read slashdot. I need to sell my AMD stock and buy a new amp.

strange.

[CDPatten]

You would think with all their resources intel could start to make a chip to compete with AMD.

Its really surprising to think AMD blind-sided intel this badly (multi-core/x64), but I guess they really did. Good for them, and great for us. Once again supply and demand in the free market prevails.

Dude, you're getting whatever we sell you!

[Wesley Felter]

Dell is locked into Intel and they really needed dual core, so there it is.

Re:Dude, you're getting whatever we sell you!

[cbreaker]

Yes, I think this is exactly why. Other big vendors - IBM, HP, Sun - they all have Opteron/Athlon machines in their line-ups. When I asked a Dell rep why Dell had zero, and no intentions to ever have any AMD, he said it was because AMD wouldn't be able to supply them enough CPU's. I call bullshit. AMD has a great deal of production capacity, and adding more all the time. Dell wouldn't have to all of a sudden convert 100% of it's line up with AMD. But, therein may lie the problem. They very-well might have to, or lose some insane deals with Intel. I think that's why they stay Intel - and mention it on every single Dell ad.

If I could upgrade my existing 2P dell servers to even inefficient dual cores that run too hot, I'd do it. But I doubt my existing servers would be able to cool them, so it's probably not going to happen anywyas. If we could get 2x dual-Opteron servers, we'd jump on it for all our ESX servers - especially with ESX3 and native x64 memory support. SWEET! But no, we'll be stuck with Xeon "EMT64" bastardized x64 CPU's because we're locked into Dell.

Intel is all about the Mhz

[mahdi13]

Intel has been really slacking more and more since AMD beat them in the 1GHz race. After that Intel has seemingly been focusing on making the 'fastest processor' and not improving on the design much.

It seems to me that Intel procs these days are more of the same but overclocked; while AMD has been making their procs more efficient, by running cooler and streamlining the instructions.

Faster isn't better these days and Intel needs to realize this before it's too late.
I just picked up a +3200 AMD Sempron which is clocked at 1.8Ghz and compre that to the AthlonXP +1600 at 1.4Ghz I had before, it has well over double the perfomance in almost every application. From doubling the fps in Doom3 to cutting compile times down by half. For a 400Mhz difference there is a lot more going on then just 'speed'

Re:Intel is all about the Mhz

[dpilot]

Both parent post and grandparent post are making the mistake of assuming Intel's current problems are technical, and the result of engineering. That's utterly wrong, the problem is higher up - a product of corporate culture, management, and myopia.

Since you brought up IA64...
While I'd be one of the last to argue against the ugliness of the x86 instruction set, that does not mean that IA64 is necessarily better. From an instruction architecture point of view, IMHO IA64 looks like an academic exercise got sold to the executives well before it was really ready for prime time. Look at the sheer amount of money Intel and HP have dumped into IA64, Sure, they can get some impressive results, but I suspect that given THAT amount of cash, time, and engineering, x86, Alpha, Power, etc all could have reached at least the same performance level.

It's necessary to realize that the number 1 problem it was designed for had nothing whatsoever to do with performance. IMHO, the prime purpose of IA64 was to prevent cloning. Neither Intel nor HP hold any of the IP on IA64 - it's all held by a third company, and Intel and HP are the licensees. That's because both Intel and HP are extensively cross-licensed with others in the business, including AMD. Had Intel and HP owned the IP for IA64, it would have come in under those cross-licensing agreements. By setting up the third company, there is no cross-licensing involved, and ONLY Intel and HP can product IA64.

So IA64 is a product of "corporate myopia," of Intel being more concerned about it's internal problem of cloning than customers' external problems of power and performance, and I once heard it was another attempt by Carly Fiorina to "reset the clock" on her tenure by announcing a grand new strategy that needed her at the helm. Both are driven by internal politics, not the marketplace. It's a classic problem of big companies

Re:Yet strangly...

[PygmySurfer]

Intel's sales will again beat AMD's by several fold. The reason seems to be that most PC and server purchases are not intended for games, beyond Solitare of course, and people prefer the reliability, power savings and lower temperatures of the Intel chips.

Did you miss the part where they said this chip consumes more power and runs hotter than Opterons?

HORUS

[Anonymous Coward]

These have been availible in Dell servers for a while now according to the online store. Intel are truly screwed for at least the next 6-12 months by the looks of things unless they are hiding something seriously good. I had thought that perhaps they had been based on Apple's decision to switch, it looks like they might just be pretending to be better than they really are though.

AMD looks like it's going to continue to be the winner on performance for the foreseeable future, especially with it's totally awesome HORUS chipset on the horizon which might just hail the beginning of commodity super computing.

For anyone wondering what HORUS is, it's an SMP system that can link 4 Opteron's together over HTT. The real killer is that it can it's self be linked to 4 other HORUS chips over InfiniBand. A HORUS SMP system appears as another Opteron chip to the other HORUS groups. AMDs current plans are for HORUS to scale to 32 CPUs in a hot swappable configuration. It's going to be great.

Re:Yet strangly...

[Glock27]

OK, troll...I'll bite.

Intel's sales will again beat AMD's by several fold.

Perhaps, although AMD has made impressive inroads into the server/enterprise marketplace and there's no sign of it slowing down.

The reason seems to be that most PC and server purchases are not intended for games, beyond Solitare of course,

Non sequitur, Opterons smoke Xeons at enterprise tasks like web serving, database hosting and so on, in almost every benchmark. Especially in the more enterprise-relevant 2-way and 4-way (4 or 8 core) configurations.

and people prefer the reliability, power savings and lower temperatures of the Intel chips.

RTFA. For several YEARS AMD's chips have been lower power and cooler than Intel's - a combination of doing more work at lower clock frequencies, and SOI. You're recalling something from the K6 days that is totally backwards today.

AMD should be happy they ran Cyrix out of the business but, they should have realized by now that they will not impact Intel sales no matter how vocal their fanboys might be.

AMD has already impacted Intel's sales in a big way. Did you hear about Intel's disappointing earnings today? Even worse for Intel, AMD has *creamed* the Itanium. Now 90% of what were potential Itanium customers (big bucks for Intel) are now going to do AMD64 instead...even if it happens to run on Intel silicon. Itanium is a financial and technical disaster for Intel.

Remember the days when AMD cloned Intel's instruction sets, not vice versa?

BTW, could I borrow your Opteron, I need to fry an egg for breakfast.

Wow, how...witty. At any rate, looks like Xeons are the hot ticket there... ;-)

Re: $700 - correction

[Glock27]

I'm seeing Intel dual-core processors appearing in ~$700 PCs. From that angle Intel absolutely devastates AMD, as somehow their dual-cores are far less expensive.

You're talking about Pentium-D of course, not Xeon...

At any rate, that is actually bad for Intel. AMD brought out enterprise-class dual core CPUs that have obvious applications on workstations/servers, which run lots of tasks and threads, and can always use more horsepower for higher throughput. Intel brought out, at about the same time, the Pentium-D for consumers. Not only is it clocked at about 1 GHz. slower than the fastest single-core Pentium, but desktop PCs don't typically run large thread and process workloads like servers. In fact, the Pentium D runs games substantially slower than cheaper, single-core Pentia. So, I expect a lot of consumers are out there scratching there heads over whether or not to buy Pentium-D.

AMD's dual core chips, on the other hand, only run 200 MHz. slower than the corresponding single core chip. Game performance suffers hardly at all. AMD will ramp up production of dual-core consumer chips once it feels it has a firm hold on the workstation/server side. Then we'll see the prices drop, and dual core will become mainstream. Maybe game developers will even start programming multithreaded games. ;-)

In summary, AMD is laughing all the way to the bank, while Intel has to content itself with low consumer product profit margins. It seems this new Xeon won't change that dynamic much.

Just specced out an AMD system for MHD modeling

[Dr. Zowie]

Running a Xeon dual-core is like mounting a Chevy big-block engine under a VW carburetor. The memory access just isn't there. Most of my stuff (modeling the solar corona) is RAM-bound anyway, so there's no win to be had at all by running the dual Intel cores. The Opterons have better RAM latency, which is a win -- but, more importantly, the two cores communicate cache-to-cache at the CPU clock speed, so dual-threaded processes run amazingly fast. If they're sharing memory, you effectively double the L2 cache size of both cores, which is a big win all around.

So, er, Xeon is teh 5uk and Opteron Pwns.