Intel strikes back

[LiamC]

~Gack! I'm a maroon! That would either be 12:00 US Central, 4:00PM Aus EST or 12:00AM US East Coast, 3:00PM Aus EST

 

[LiamC]

The King is dead! Long live the king. For those who thought the "leaked" Core Dua 2 benchmarks were rigged:

http://www.techreport.com/reviews/2006q3/core2/index.x?pg=18

A quick synopsis. An Intel E6600 Core Duo 2 (2.4 GHz) slated for a US$316 price point, in most tests beats the AMD FX-62. When it doesn't, it is still faster than a 4800+. If AMD were to compete on price/performance, the FX-62 would have to come down to the $300~$350 dollar range.

Long live competition.

Now, a new Folding box...

 

[Pradeep]

http://anandtech.com/cpuchipsets/showdoc.aspx?i=2795&p=1

"The 2.4GHz E6600, which outperformed the FX-62 in most benchmarks at stock speed costs $316, and overclocked to 4Ghz with excellent air cooling."

Certainly seems to be the new processor of choice. Caveats being availability, and limited chipset selection for the next few weeks (if SLI matters to you).

Tony, quick question: Will you be selling any of the new Intel chippies in your shop, considering the stunning performance/price?

 

[Mercutio]

If the rumored AMD X2 price cuts hit, I'm still going to have a hard time finding the will to buy Conroe chips; AMD chips are a known quantity, as are AMD motherboards, and AMD boards tend to be much cheaper anyway.
Hopefully there will be a round of price cuts by the fall; a $120 or $150 chip is a lot more interesing than a $190 one.

On the other hand, the better-in-every-way qualities of Conroe will, I hope, lead at least a couple of my contracting clients to spend some money to finally dump aging P4 desktops.

 

[ddrueding]

That's what I'm most looking forward to. I have several clients who are AMD-fanboys like no other. They insist on getting Intel chips beacuse they are "proven technology". I haven't had the heart to tell them that all their servers have been AMD for years.

 

[Bozo]

Do you think Steve Jobs knew this was comming from Intel?? Maybe got some inside info from Intel??


Bozo :joker:

 

[Mercutio]

I'm positive of it. Even still, Apple couldn't seriously consider AMD chips; AMD has enough problems manufacturing to meet demand as it is, and chip-shortages were a big reason Apple dropped Power in the first place.

A Merome-based Thinkpad... Mmmm. That WOULD be nice.

 

[LiamC]

As expected, AMD's price cuts:

http://www.dailytech.com/article.aspx?newsid=2800

A sample:

Processor Now Price 7/24 Price
X2 5000+ $696 $403

X2 4600+ $558 $301

X2 4200+ $365 $240

X2 3800+ $303 $169

I still think the high end chips are on the expensive side, but...

 

[Tannin]

http://anandtech.com/cpuchipsets/showdoc.aspx?i=2795&p=1
Tony, quick question: Will you be selling any of the new Intel chippies in your shop, considering the stunning performance/price?

Pradeep, quick answer. I am a price-performance junkie. If they really do deliver a better product at a lower price, absolutely. Three things have to happen though:

• Reliable supply in volume. That's reliable supply: if the chips are subject to frequent out-of-stocks, that's a downcheck.
• Mainboards. I'm not a big fan of Intel chipsets, but some of them are certainly OK (i.e., looking at the one I have used in the past as a rough guide to what this new lot might be like). Mainboard price is a significant factor too: Intel-based mainboards have, in years gone by, often been dealbreakers, as even when you get a good price on the CPU you can often find that the extra cost of the mainboard makes the system less competitive than you hoped.
• AMD's response. My expectation - again based on history - is that AMD will adjust their pricing such that there is no particular advantage to switching, but this is another wait-and-see situation. Bear it in mind that I'm not interested in the high end parts - practically no-one buys them. Right now the things that sell well, in order, are:

• Sempron 3000 - would be the excellent 3100 but out-of-stocks are a problem with that part and we have pretty much given up on the 3100 for that reason.
• Athlon 64 3200 - picking up and might be #1 soon
• Sempron 2800 - still a regular
• Athlon 64 3000 - fading fast now
• Athlon 64 4200 - still very slow but picking up

 

[LunarMist]

Interesting. Does anyone know how well this new CPU will convert RAW files? Will there be many speed bumps soon?

 

[LunarMist]

speed bumps

Erm, I meant increases in speed.

 

[Bozo]

I don't think clock speed is all that important any more. It's more about getting more done with what you have. Efficient use of power. But there may be 'speed bumps' next year, and there is always overclocking.

BTW, congrats on 1000 post :beer:


Bozo :joker:

 

[LunarMist]

I don't think clock speed is all that important any more.

But shouldn't a moderate clock speed within a CPU family yield a nearly proportional increase? For example, if it takes 12 hours to process a batch of RAW files with a given processor, a 20% faster processor should do the job in a little over ten hours (some loss to HDs, etc.).

 

[P5-133XL]

But shouldn't a moderate clock speed within a CPU family yield a nearly proportional increase? For example, if it takes 12 hours to process a batch of RAW files with a given processor, a 20% faster processor should do the job in a little over ten hours (some loss to HDs, etc.).

Sure in a perfect world without any other limiting chokepoints like RAM bandwidth, I/O limiters like HD speeds, or even graphics processing for gaming. In the real world, you only go as fast as the slowest component (very often, the human operator) but in a batch processing setup, that would most likely be the hard drive. It's been a long time (years) since we've been CPU bound (with a reasonable computer) for most computer work.

I would really like it if the HD manufacturers would actually offer high speed average access times found in SCSI drives instead of increases in capacity for consumer drives. At least they have started offering those high speed business-class drives some good capacities but you do have to pay $$$ to get both.

 

[Tannin]

In general, I agree with Mark, but I suspect that raw conversion is one of the exceptions. Have to test it to be sure either way.

 

[Pradeep]

RAW conversion is definitely CPU bound. I haven't seen any benchmarks of Core 2 Duo with RAW conversion yet. It would be hard to imagine that even the mid level Duos wouldn't outpace the 3.6 and 3.8 GHz Pee4s. And with a little overclocking, you should have a nice performance gap from yesterdays state of the art.

 

[LunarMist]

In my ancient system the CPU is at 100% for about 15 seconds with no diks activity. Prior to that the RAW file is read in less than a second from one disk and it takes a second or two to write the TIF file to a separate disk after conversion.

 

[Splash]

Do you think Steve Jobs knew this was comming from Intel?? Maybe got some inside info from Intel??

Steve Jobs certainly knew about Core and Core2 more than 2 years ago (Itanium goings on as well). He was very likely "shown the lab" at Intel before he made any decisions. He has regularly been wined and dined over the years by Intel people. One of the people that helped start Apple (the money man) was ex-Intel. Apple's headquarters is near Intel's headquarters. Apple has used Intel 960 RISC controllers and X-Scale processors in their laser printers and storage arrays, so, they already were an account holder.

If, like me, you are primarily interested in price-performance, and only mildly interested in raw performance, and not at all interested in the performance of anything that costs so much that no-one will buy it, it has been a very long time.

If it's strictly price, yes, you will be sticking with AMD for at least the next year, because Intel will not have any "bargains" other than end-of-production markdown "Prescott" P4 processors for a while.

However, I'm guessing by October of next year, that tune will likely change because Intel should be pumping out zippy little cool running Core Solo (i.e. -- 32-bit Core"1") desktop chips for absolute dirt cheap prices, with the Core Duo at a whopping 10% price premium. By October of 2008, 32-bit Core processors will probably still be around but fabricated at 45 nm and costing 50% less than before and capable of outrunning a 3.0 GHz Pentium 4 (at that point, I'm thinking 2 GHz for US$20). But, as far as Intel goes, the real action from here on out will *definitely* be processors based on Core2 (64-bit).

The Pentium 4 has never been attractive. The current ones are red-hot and dog-slow...

You'd be happy to know that the last Pentium 4 was recently released (65 nm fabrication). I don't know why they really even bothered with the damned thing. It is only a bit faster than the previous generation "Prescott." (PS: Prescott, Arizona: I've been through there, and it's plenty hot. Good code name for that processor!)

Then there was the legendary Celeron 300A. A damn good chip. A brilliant chip if you were into overclocking, maybe the best overclocker of all time...

The Celery 300A is now the second best Intel overclocker.

The Core2 "Conroe" processor has tremendous overclockability. There have been numerous reports of overclocking well into the 4 GHz range with Core2 processors. There was a very recent major stepping (B0) of the Core2 core (for Duo and Solo processor packages, both Desktop and Xeonic flavours, and probably mobiles). There has been a few reports of people getting an astounding 5+ GHz out of these along with an equally astounding 2+ GHz front side bus speed!


Below is a handly little table that I've not had a chance to update in a while:

• 
  
  SERVER
  
  Core2 Duo Xeon "Woodcrest" (64-bit dual-core Core2 Xeon)
  Model# Clock GHz FSB MHz Watts L2 Cache
  5160 3.00 1333 80 4MB
  5150 2.66 1333 65 4MB
  5148(LV) 2.33 1333 40 4MB
  5140 2.33 1333 65 4MB
  5130 2.00 1333 65 4MB
  5120 1.86 1066 40 4MB
  5110 1.60 1066 40 4MB
  
  
  ------------------------------------------------------------
  
  Pentium-D Xeon "Dempsey" (64-bit dual-core Pentium 4 Xeon)
  Model# Clock GHz FSB MHz Watts L2 Cache
  5080 3.73 1066 130 2+2 MB
  5070 3.46 1066 130 2+2 MB
  5063(LV) 3.20 1066 95 2+2 MB
  5060 3.20 667 130 2+2 MB
  5050 3.00 667 95 2+2 MB
  5040 2.83 667 95 2+2 MB
  5030 2.66 667 95 2+2 MB
  5020 2.50 667 95 2+2 MB
  
  
  ------------------------------------------------------------
  
  Core2 Solo Xeon "Kaylo" (64-bit single-core Core2 Xeon)
  Model# Clock GHz FSB MHz Watts L2 Cache
  31xx 3.00 1600 4MB
  
  
  ------------------------------------------------------------
  
  Core2 Quadro Xeon "Cloverton" (64-bit quad-core Core2 Xeon)
  Model# Clock GHz FSB MHz Watts L2 Cache
  71xx 2.67 1066 110 4+4 MB
  71xx 1.86 1066 80 4+4 MB
  
  
  ------------------------------------------------------------
  
  Core Duo Xeon "Sosaman" (32-bit dual-core low-voltage Core"1" Xeon)
  Model# Clock GHz FSB MHz Watts L2 Cache
  ??? 2.00 667 31 2MB
  ??? 1.66 667 31 2MB
  
  
  
  DESKTOP
  
  Core2 Duo "Conroe" (64-bit dual-core desktop Core2)
  Model# Clock GHz FSB MHz Watts L2 Cache
  X6900 3.20 1066 80 4MB
  X6800 2.93 1066 80 4MB
  E6700 2.66 1066 65 4MB
  E6600 2.40 1066 65 4MB
  E6400 2.13 1066 65 2MB
  E6300 1.86 1066 ??? 2MB
  E4200 1.60 800 ??? 2MB
  
  ------------------------------------------------------------
  
  Core2 Quadro "Kentsfield" (64-bit quad-core desktop Core2)
  Model# Clock GHz FSB MHz Watts L2 Cache
  ??? 2.40 1066 95 4+4 MB
  
  ------------------------------------------------------------
  
  Pentium-D "" (64-bit dual-core desktop Pentium 4)
  Model# Clock GHz FSB MHz Watts L2 Cache
  965 3.73 1066 130 2+2 MB
  960 3.60 800 130 2+2 MB
  955 3.46 1066 130 2+2 MB
  950 3.40 800 130 2+2 MB
  940 3.20 800 130 2+2 MB
  930 3.00 800 95 2+2 MB
  920 2.80 800 95 2+2 MB
  840 3.20 800 130 1+1 MB
  830 3.00 800 130 1+1 MB
  820 2.80 800 95 1+1 MB
  805 2.66 533 95 1+1 MB
  
  
  
  MOBILE
  
  Core2 Duo "Merom" (64-bit dual-core mobile Core2)
  Model# Clock GHz FSB MHz Watts L2 Cache
  T7700 ??? ??? ??? 4MB
  T7600 2.33 667 35 4MB
  T7400 2.16 667 35 4MB
  T7300 ??? ??? ??? 4MB
  T7200 2.00 667 35 4MB
  T7000 ??? ??? ??? 2MB
  T5600 1.83 667 35 2MB
  T5500 1.66 667 35 2MB
  L7500 ??? ??? ??? 2MB
  L7400 ??? ??? ??? 2MB
  U7300 ??? ??? ??? 2MB
  U7200 ??? ??? ??? 2MB
  U7500 1.06 533 15 2MB
  
  ------------------------------------------------------------
  
  Core Duo "Yonah" (32-bit dual-core mobile Core"1")
  Model# Clock GHz FSB MHz Watts L2 Cache
  T2700 2.33 667 31 2MB
  T2600 2.16 667 31 2MB
  T2500 2.00 667 31 2MB
  T2400 1.83 667 31 2MB
  T2300 1.66 667 31 2MB
  L2400 1.66 667 15 2MB
  L2300 1.50 667 15 2MB
  U2500 1.20 533 9 2MB
  
  ------------------------------------------------------------
  
  Core Solo "Yonah" (32-bit single-core mobile Core"1")
  Model# Clock GHz FSB MHz Watts L2 Cache
  T1400 1.83 667 27 2MB
  T1300 1.66 667 27 2MB
  U1400 1.20 533 5 2MB
  U1300 1.06 533 5 2MB
  

 

[LiamC]

Looks like the price cuts will be even deeper than expected. DailyTech has a new distributor price list up, so add US$15 to $25 dollars for US retail

http://www.dailytech.com/article.aspx?newsid=3361

Processor Today $ 24 July $
5000+ 649 282
4600+ 522 224
4200+ 339 175
3800+ 277 149
Semp 3400+ 92 69
Semp 3100+ 78 EOL

 

[Gilbo]

Intel's new chips aren't perfect. Apparently, they lose some functionality in 64-bit mode. Macro-ops fusion has to be shut off with 64-bit instructions. Also the instruction fetch from L1 cannot, in some cases, fetch 4 instructions per cycle in 64-bit mode, rendering Core 2's widened execution core starved / useless under those conditions.

At the moment most code isn't available compiled for AMD64, let alone optimized for it. I wonder to what extent if any, these limitations will reduce Core 2's performance on the more optimized code that should start to appear around the launch of Vista. X-Bit Labs has a very limited investigation here. At the moment the Core 2 does scale to 64-bits worse than the K8, but it's nothing major in most cases. Mostly because the potential performance increase seems limited. ScienceMark is an example of a well optimized 64-bit program. In one test the K8 scales significantly better. In the other the two scale the same. The former, unfortunately, is the one where 64-bit turns out to be much more useful...

From what I understand, the loss of Macro-ops Fusion isn't that big a deal. I'm mostly concerned that Core 2's extra execution units won't be utilized exactly when the greatest performance benefits should be extracted --when the 64-bit execution units and the extra registers actually could provide big boosts. I guess there's no such thing as a free lunch. The L1 fetch has always been a limiting factor in clock rate scaling in x86 CPUs. It's invariably a compromise between extracting parallelism & increasing the clockrate. Considering the prevalence of 32-bit code on Windows machines at the moment they probably made the better choice for most users and applications.

I'm a little concerned about this because I've had all 64-bit applications for years now, thanks to the GNU/Linux community. Now I might have to move back to 32-bit for best performance. What about image editing? 64-bit math is useful in these apps. The extra memory is useful in these apps. Unfortunately there aren't nearly enough good tests available at the present time to see what the consequences of these limitations will really be. I'll probably end up buying one eventually anyway, so I guess I'll see.

 

[Mercutio]

My take is that if you're on a new-ish processor, you're almost always going to get better performance than on an old one. You're fretting about a performance differential that's probably going to be too minute to notice, and you're worrying about performance on a top-end chip at that. The real-world difference between optimal and "sub-optimal" Core 2 Duo performance will probably be told only with very close benchmarking; I doubt there will be any subjective difference at all. That's how things seem to be these days.

 

[Gilbo]

I mostly agree with you Merc. OTOH it does make a difference for some photography applications. Batch converting RAW files is still a long way from being the instantaneous process I'd like it to be and the X2's could well be noticeably better values than the Core 2 depending on the effects of these limitations. Of course, I haven't bought anything but chips at or near the bottom of the food chain for years now for exactly the reasons you're talking about Merc.

Mostly I'd just like to know more for knowledge's sake. Concerned in a curious way shall we say.

 

[Splash]

A while back, I was "allowed" to play around with a Woodcrest-based server and a Dempsey-based server side-by-side, both of which were running Windows Server 2003 R2.

The seat-of-the-pants test pretty much revealed that the Woodcrest server was clearly faster at everything compared to the Pentium-4-based Dempsey. Both had 4 GB of RAM, but the Dempsey was running at 3.73 GHz and the Woodcrest was running at a measily 2.33 GHz!

 

[Splash]

> measily 2.33 GHz!


measly 2.33 GHz.

me silly

 

[ddrueding]

> measily 2.33 GHz!


measly 2.33 GHz.

me silly

Oh yes, that's much faster :crap:

 

[Adcadet]

what's the verdict on 64-bit vs. 32-bit software? Other than having room to address more memory, does it make a difference?

 

[sechs]

On a general basis, I'd say, not enough that you could tell.

A well-defined group of applications can gain greatly.

 

[Bozo]

A test computer that I use is 64 bit capable. Putting a clean install of XP 32bit on it with all the patches, loading mundain software (office etc) using it for a few hours then loading XP 64bit with the same mundain software, you notice that the 64bit version is snappier (faster??). But, the XP 64bit install has the new SP2 for 64bit installed. (still in beta) SP2 for 64bit made a huge difference.

Bozo :joker:

 

[Mercutio]

XP64 uses the same kernel as Server 2003, a newer one than 32-bit XP uses. I've stated before that Server 2003 feels about 10% faster than XP, and that's probably what you're seeing, Bozo.

I've had 64-bit Linux for a while. Side by side (well, co-mingled X sessions, anyway) with 32 bit Linux, I couldn't tell the two apart.

 

[Bozo]

Merc,
I ran the pre-SP2 64bit XP for a few weeks. There is a difference with SP2. It's hard to describe.
Even Server 2003 seems better with SP2 beta.

Maybe it's just me :-D


Bozo :joker:

 

[Pradeep]

Performance of FB-DIMMS with Woodcrest Xeons:

http://anandtech.com/mac/showdoc.aspx?i=2816&p=11

I'm glad AMD stuck with normal DDR2.

 

[Vlad The Impaler]

Well, I have to say that these Core 2 Duos are really something else.... I just did a straight swop (retaining the XP installation) for my old 945/P4 3.4 to a 965 and a Core 2 1.83Ghz. The difference is unlike anything I have seen before. It seems in general boot up speeds and Windows about 50% faster. I have swopped out many different boards into my good old Supermicro case in my time, starting with a K62-350 on an Epox board. This is by far the biggest noticeable leap I have ever experienced. And all on a £100 CPU, the bottom of the range! Unbelievable!