Previewing the Performance of the Intel Conroe

pirate rtt writes "bit-tech has spent some time with an Intel Conroe system and has published a preview of its performance as compared to the current Intel flagship chip - the Presler 965. From the article: 'Core 2 Duo is clearly a very capable processor. We found that it was faster than the current 965 processor in most situations on the desktop, and far more proficient at gaming - an area where Intel has traditionally been weak. The added memory bandwidth that will come from having faster RAM enabled on the Core 2 Extreme chips will be an extra bonus for those looking to Conroe as a gaming platform.'"

Comments

Wait for v2

[MarkByers]

Core 2 Duo is clearly a very capable processor

Version 1 of anything always sucks. I'm waiting for the Core 2 Duo II v2 Second Edition.

Re:Wait for v2

[Crussy]

This has been brought up many times. Every chip has a sheet of bugs and the core duo is no exception. Though if you had actually read the sheet you'd notice that almost all of the bugs were found during intel testing and most involved extremely rare scenarios. The amount of bugs reported show only that intel is conducting serious testing. As for not fixing the bugs, do you think AMD is fixing all of the bugs its processors have? http://www.amd.com/us-en/assets/content_type/white _papers_and_tech_docs/25759.pdf There are a number of bugs still in amd processors and even a few are not planned to be fixed. However this is really negligible concern to anyone.

Re:Wait for v2

[Tough Love]

"All modern cpus have bugs. It is common practice to work around them in the compiler rather than retape the chip, an expensive and time consuming process."

Nonsense. For compilers do not work around bugs in general purpose chips. If a chip bug can't be worked around by microcode or bios settings, or (in rare circumstances) the operating system, the chip will be binned. Compatibility is king in the general purpose CPU market. Nobody can sell a CPU that crashes on some programs that used to run perfectly well.

Conroe vs. FX-62

[Anonymous Coward]

And now for Conroe vs. AMD's Athlon FX-62 (and presler).

Yes, the FX-62 does lose... badly in several cases..

Re:Conroe vs. FX-62

[Brian Stretch]

1) Hexus used Intel's compilers with their synthetic benchmarks. Intel has been known to rig their compilers to ignore post-i486 instructions (SSE, etc) on non-Intel CPUs. This is suppoesd to have been corrected in later Intel compiler releases, but...

2) Some of those benchmarks, like Pifast, likely fit inside the Core 2 Duo's massive L2 cache. Intel uses all that expensive cache to compensate for their lack of on-board memory controllers and HyperTransport.

3) Curious how they chose much lower latency memory for the Intel machine than the AMD. I'm not sure that the higher bandwidth of the AMD PC's memory overcomes its higher latency.

4) Why use 1024x768 res for the FarCry benchmark and 1600x1200 with AA and AF cranked up for theother two games? Games are GPU-limited at hires, so if you wanted to spike the results where AMD is superior...

5) Despite all of that, the AMD FX62 still won the Cryptography benchmark.

6) Why are nearly all of these reviews showing up on websites outside of America? Could it be that Intel wants to keep these reviews out of reach of AMD's American lawyers?

It sure looks like Intel's playing dirty (again). Wake me up when we get reviews done outside of Intel-controlled environments.

Re:Conroe vs. FX-62

[Anonymous Coward]

1) Hexus used Intel's compilers with their synthetic benchmarks. Intel has been known to rig their compilers to ignore post-i486 instructions (SSE, etc) on non-Intel CPUs. This is suppoesd to have been corrected in later Intel compiler releases, but...

For which benchmarks do you claim they used special Intel compilers? The only reference in the article is that they used an "Intel HT compiler" for their "HEXUS.in-house MP3 Encoding Benchmark".

2) Some of those benchmarks, like Pifast, likely fit inside the Core 2 Duo's massive L2 cache. Intel uses all that expensive cache to compensate for their lack of on-board memory controllers and HyperTransport.

Noone's interested in Pifast scores. Conroe beating the FX-62 by almost 60% in Far Cry is much more impressing. And using a 65nm process Intel can sell Conroes with 4MB cache for very competitive prices. If you believe the leaked documents on the web a Conroe E6600 will sell for a little over $300 (in quantities of thousand) and should be on a performance level of the best AMD CPU.

3) Curious how they chose much lower latency memory for the Intel machine than the AMD. I'm not sure that the higher bandwidth of the AMD PC's memory overcomes its higher latency.

Even if there was a 5% difference, it wouldn't change the big picture.

4) Why use 1024x768 res for the FarCry benchmark and 1600x1200 with AA and AF cranked up for theother two games? Games are GPU-limited at hires, so if you wanted to spike the results where AMD is superior...

That's an interesting question indeed. But all other Conroe previews suggest that its gaming performance will be stellar.

5) Despite all of that, the AMD FX62 still won the Cryptography benchmark.

The average user doesn't use his CPU for cryptography very much.

6) Why are nearly all of these reviews showing up on websites outside of America? Could it be that Intel wants to keep these reviews out of reach of AMD's American lawyers?

Maybe because Intel makes most of its revenue outside the US?

It sure looks like Intel's playing dirty (again). Wake me up when we get reviews done outside of Intel-controlled environments.

July 23rd is rumored to be the launch date. Until then I'd definitely wait before buying a new CPU. Even if you want to buy an AMD processor, they're will be huge price drops.

Are These Reviews Significant?

[buckhead_buddy]

The article reads:

The Core 2 Duo benchmarks we ran were not completed in our own labs and we have used some unfamiliar tests in order to establish how well the new Core architecture performs. This was because we were not allowed to tweak the system or install our own benchmarks - the machine was built and configured by Intel engineers.

So if Intel provided hardware, chips, and tests themselves, isn't this more of a write-it-yourself press release from Intel than a real independent review? If they provided "some unfamiliar tests" then that would seem to indicate Intel doesn't know what common and familiar tests should be run, or (more cynically) that they didn't want potentially bad or uncontrolled results polluting their positive reviews.

Are reviews like this of any real significance?

And what makes this review different?

[LIGC]

There have been Conroe previews several sites, all of them using intel preconfigured boxes. The only way to make sure of Conroe's performance is to benchmark it once its bought. I'm sorry if I'm a bit of a sceptic, but I don't trust these boxes made by Intel to market Conroe.

Sorry but

[opusman]

I'm waiting for the 400 page Tom's Hardware review!

gaming

[ltwally]

"...far more proficient at gaming - an area where Intel has traditionally been weak..."

Define "traditionally." Normally that word is reserved for a long-term scope, in which case Intel has been better at gaming than the competition. Up until the Athlon, no x86 clone could compete with Intel when it game to games (2d or 3d). Think back to the K5 & K6 -- neither was good at 3D (not even the K6-3 could compete with a similarly clocked P2/P3. And the processors from previous generations of AMD, Cyrix, and IBM were much the same. The Athlon was the very first x86 clone that was better at gaming than an Intel flagship processor.

So, this isn't so much as Intel stealing the crown as re-claiming the thrown.

<shrugs>No big deal .. I just get tired of the Intel bashing crowd. "OMG INTEL IS TEH DEVIL, AMD IS OUR SAVIOR!!"

Re:About Time

[ruiner13]

Why does a laptop need 64-bits? Are you addressing more than 4GB memory? I haven't seen a laptop yet that can support more physical memory than 32-bit chips can address, nor can I see someone doing heavily scientific work on a laptop as they tend to have slower, smaller hard drives. Extended memory and scientific precision are the only valid reasons I can think of needing 64-bit architecture, neither of which apply directly to laptops.

Re:About Time

[TheRaven64]

I write code on my PowerBook that is going to be deployed on either a 64-CPU IRIX box, a 40-node UltraSPARC cluster, or a 64-node dual-Xeon cluster. Some of this code mmaps large blocks of data. It doesn't matter that I don't have that much physical memory, because not all of it will be loaded at once.

Remember, it is not uncommon to use more address space than you have physical memory. This is why we have swap.

Technical Prowess of Reviewer?

[macentric]

When I was studying Computer Science in College I took a Parrellel Programming class. One of the first things my professor taught my class was that not all tasks are well suited for parrallel processing. In many cases it is more difficult to break the task up into smaller tasks and hand off to other processors or cores. In many cases performance will be significantly degraded by parallelizing a task that is not well suited to multiprocessing.

During several of the tests, the author of the article ran single threaded and multithreaded tests. In some of these tests, the performance of both the Conroe and Pressler chips decreased. The author incorrectly states that the multitasking performance of the Conroe chip is lower than the Pressler chips. He is incorrect becuase his own graphs reference multi-threaded performance. These are two entirely different things. While the multi-threaded perfomance of Conroe is slower in some cases, the single threaded performance was faster in most cases. These tasks obviously are not tasks well suited for parallel processing, and as such should be coded to run as a single thread to keep performace high.

While the rest of the benchmark seems solid, his analyis should be brought into question as he doesn't seem to have a solid grasp on his technology vocabulary. That or his editors don't know what they are reading either. If that is the case their reviews should not be showcased.

Re:Technical Prowess of Reviewer?

[macentric]

Hyper-Threading is a load of crap. It fooled the computer into believing that the computer has an additional processor. The reason that they removed Hyper-Threading from the dual core chips is that it is redundant and not needed. You are not going to get any greater performance gains from multithreading on a single chip than you are from a dual core chip.

On another note, it is up to the developer of the optimize his/her program for the best performance. Individual tasks inside a program can be made to run single threaded or multithreaded based on how well that part of the program performs. The programmer must make these decisions or leave them to the compiler. In many cases the programmers will understand that the compiler is building generic code that may not be suitable for their program.

These benchmarks were run as single threaded or multithreaded. This has absolutely NOTHING to do with multitasking. Multitasking is running several programs or processes simultaneously. The individual programs in these benchmarks may not have been that optimized for these tasks and may have left significant headroom for other processes to run efficiently. An example of this is that many programs that I use on a dual processor G5 will never take full utilization of the available processing power. It may take running several instances of a process to fully load up the CPU utilization on the system. This is why virtualization is being looked at as a holy grail of saving money in the data-center as it will allow those data-centers to make better use of their investments.

Next time you are encoding audio or video go check your task manager or activity monitor and look to see what your CPU utilization is. Go load your web browser and surf around a while. You will not notice the encoding stop, it may slow for a few cycles every time you load a new page. This is your computer multitasking. Multithreading is allowing an individual process or program to do more things at once instead of waiting for an invidiudal task to complete before moving on to another task. You do not need multiple processors to take advantage of multithreading, nor do you need that crap called Hyper-Threading.

I thought this was a place for geeks and nerds who knew something.

"Traditionally been weak" ?

[drsmithy]

People who have only been observing the industry for a handul of years shouldn't be talking about "tradition". The Athlon was the first product AMD made that could even offer a match to intel's CPUs, let alone exceed them.

I'm sure there's more than enough people here who remember how intel poorly comparbadly the K6 ran Doom, for example.

Not to mention the atrocious record of motherboard chipsets for >K6 AMD processors that, alone, contributed more to slowing their uptake by the market than any other factor (it astounds me that VIA has managed to stay in business).

The Conroe myth gets busted a little bit every day

[Visaris]

We found that it was faster than the current flagship Pentium Extreme Edition 965 processor in nearly every single-threaded scenario, but there were times where Conroe fell behind in multi-tasking scenarios.

I think that's a very funny quote. This is exactly what I was expecting all along. The reason most people have been running Super-PI and other toy benchmarks is because they are single-threaded, and that is the one area where Conroe really shines.

If the Conroe can't beat the Pentium Extreme Edition 965 how is it going "own" or "destroy" an Athlon 64 FX-64? The Conroe myth gets busted a little bit every day.

Re:About Time

[Brian Stretch]

Why does a laptop need 64-bits? Are you addressing more than 4GB memory?

You get twice as many general registers in AMD64 mode, providing a nice performance boost independent of how much memory you have. Java, cryptography, and codecs react particularly well to AMD64 environments.

2GB RAM is already pretty standard for power users. Throw in virtual memory and, voila, you're at the 4GB barrier. Being able to run the same 64-bit binaries on your notebook as on your quad processor, 8 core 64GB RAM server is kinda nice too.