Revisiting Intel’s New Pentium III at 1.13 GHz — THG.RU

bdfd229a7e16fe4f6ceceee816447239
bdfd229a7e16fe4f6ceceee816447239







Introduction

After the publication of the Pentium III 1.13 GHz article on Monday I was facing harsh criticism from several people who wrote off my moaning about the instability of Intel’s new processors as ‘obvious bias’. Now while I am tired of replying to those kinds of pathetic accusations I do understand that it is hard to believe that Intel would indeed ship a seriously flawed processor product after being world renowned as the provider of the most reliable and compatible CPUs out there. At the same time I would also appreciate however, if the sensible ones of you remember that I don’t make waves of this amplitude light heartedly. My tests showed a serious instability of Intel’s new Pentium III and that is what I reported.

It Was Not The Micro Code Update

In the article from Monday I assumed that the instability issues which kept me from supplying you with benchmark data were caused by the lack of the latest micro code update for Intel’s new processor. I also read reviews of this processor from other publications and I was surprised to see that e.g. my respected colleague Anand Lal Shimpi from www.anandtech.com had indeed been able to run his 1.13 GHz Pentium III on different platforms without any major problems. I still guessed that Anand must have had the micro code update and so I waited patiently until this morning when I finally received several new BIOS releases from Asus that included this processor software patch.

I flashed the new BIOS onto the motherboards and started the 1.13 GHz system to finally finish my benchmark runs. The boards booted properly and this time without the error message complaining about the missing micro code update. I assumed that now my testing should be done smoothly, but it turned out that I had been wrong.

Again, each platform that I used crashed at several occasions with the new Intel processor. Often I was not even able to run Quake 3, or the system wouldn’t even boot to the Windows desktop altogether. Raising the core voltage had the effect that the system wouldn’t even reach POST. At the same time each system ran flawlessly with my Pentium III at 1 GHz.

It was obvious. The micro code update had never been the reason for my problems. Instead, my Pentium III 1.13 GHz sample was clearly a faulty part.

A Faulty Production Sample Pentium III 1.13 GHz

Last night I had received a very interesting email from Kyle Bennett, the man behind www.HardOCP.com. Surprisingly enough Kyle had experienced instability issues with his 1.13 GHz Pentium III as well, although he had even used the mysterious VC820 motherboard from Intel that was specially designed for this processor.

Here’s Kyle’s statement:

We had no severe problems with the Intel 1.13GHz and VC820 (supplied by Intel with 128megs of RAMBUS) but it would NOT run Prime95 even for one minute. I do not know if you use this program, but it has been an old standby for us when it comes to testing 100% stability. We could also run low resolution benchmarks inside of Quake3 but at higher resolutions where more bandwidth is utilized, we ran into problems with the program crashing.

As per my experience it was NOT due to a Video Card problem, but rather pointed to the CPU. Overall, we were not happy with the results on the i820 mainboard and moved to another platform.

As for the BX boards we tried, an ABIT BF6 and ABIT BE6-II, we could not even get Win98SE to load at the default 1133MHz clock. Mind you, all of this was being done with components proven to run flawlessly at the 133MHz bus specification. We did get the OS to load effortlessly at 850MHz (8.5x100MHz FSB). Once stable, we raised the FSB to the 133MHz rating of the Intel CPU and was met with a flurry of BSODs.

Heat was our first thought as to an easy culprit to fix, but the HUGE Intel Stock HSF was doing a very goof job at dissipating the heat put off by the CPU. We recorded temps under the 95 degrees F mark while under load. This seemed totally acceptable to me. Tweaking SDRam settings to minimal performance to alleviate stresses from the board in other areas proved to do nothing for the problem. The fact of the matter is that we could not get the 1.13GHz Intel PIII to run stable no matter what we tried.

We made the assumption that we had a «bad CPU» or we were making mistakes somewhere in our testing. The CPU we had was marked as a 1.13GHz PRODUCTION CPU, not an engineering sample. We made a decision NOT to publish our findings as we were unsure as to whether or not our data was correct. This is a consideration we offer EVERY manufacturer. We committed to Intel to not report our findings on the Hard|OCP until we have concluded the problems we were having were «REAL» and not simply a fluke or due to improper testing on our part. As it stands now, the part is to be sent back to Intel and they committed to report back to us on the above mentioned problems.

Of course the findings that you posted yesterday prompted me to write this mail. While we have certainly not agreed with you in the past on certain issues, I did not want to see you getting «hung out to dry» on something I think you are correct reporting on.

To sum it up, I don’t think you are off base on your report about the problems with the 1.13GHz Intel PIII.

…
Kyle Bennett
WebMonger @ Hard|OCP
Purveyor of Smoothness @ Ratpadz
Hosting Ho @ Gamers|Hardware

Kyle might send his processor back to Intel, but it seems obvious that he was having serious problems as well. The one platform that performed acceptable in Kyle’s testing was the specially modified VC820 board that I still haven’t even received. However, even this platform did not show the complete stability one would expect from a system with ‘Intel Inside’.

Intel Pentium III 1133 MHz Markings

It seemed obvious that I was not the only one with a bad 1.13 GHz Pentium III sample. Both processors were official production samples with the ‘SL4HH’ coding, not some pre-production exemplars. One bad sample may have been just about acceptable, but two faulty samples raise some serious questions. Could it be that Intel is shipping a major amount of buggy Pentium III 1.13 GHz to their OEM customers right now?

Contacting Intel

I had been surprised that Intel hadn’t got back to me after my ‘nasty’ Monday article, which accused the Satan Clara based company of producing unstable 1.13 GHz processors. After realizing that my processor was obviously faulty and after finding out that another respected hardware tester had experienced similar problems as well, I considered it to be about time to get in touch with Intel. Thus I sent of the following email:

Dear Intel PR department,

as displeased as you may be with my recent review of the P3 1133 processor, as much you all know that I don’t publish my findings of a highly unstable CPU light heartedly.

Unfortunately the micro code update that I just received from Asus in form of several new BIOSes for my different motherboards didn’t do the trick for me whatsoever. My 1133 MHz sample is utterly unable to finish any benchmark, while my 1 GHz part finishes each test completely reliably in the same platform. I am occurring system hangs on a highly regular basis, completely regardless which software I am running or which platform I am using, although I am even using a heavy duty full copper super duper heat sink.

The other strange thing about this processor is the fact that it seems to request 1.8 V through its VID pins, not the official 1.75 V.

Altogether my Pentium III 1.13 GHz sample seems to be a clearly faulty part. While this may explain my negative findings and while others seem to have samples that might perform reliably, I know of at least one other person that is also having instability issues with his P3 1133. This is obviously raising a serious question. Is there a major flaw in the validation process of the Pentium III 1133? Doesn’t it seem likely that several or even many of those processors are just as faulty as mine, and currently shipping to OEMs? On top of that, I regard it as highly unprofessional to ship a faulty part to a reviewer in the first place. I would have expected that you (Intel PR) double check the reliability of any test sample you ship.

I am sure you are aware of the gravity of this issue. It might cost Intel a major face loss if I should not be the only person with a bad 1.13 GHz processor.

I expect that you start some kind of investigation and give a rather urgent statement regarding the issue described above. I doubt that you will get away with claiming that Thomas Pabst is unable to test a processor properly. Ignorance of this issue will also most likely backfire at you. I apologize for expressing this warning, but so far I am very displeased with the lack of response from Intel in regards to my negative findings.

Sincerely,

Dr. Thomas Pabst
CEO & Editor in Chief
Tom’s Hardware Guide Inc.
San Jose, CA, USA
www.tomshardware.com
CEO
TG Publishing AG
Munich, Germany
www.tomshardware.de

I copied that email to several editors of other hardware publications. The only meaningful response came from Nico Ernst, editor of www.tecchannel.de. He told me that his 1.13 GHz Pentium III production sample, while running reliably, was also requesting 1.8 V instead of the 1.75 V that it is supposed to request from the motherboard. I consider this as rather disturbing. Isn’t Intel claiming that the new CPU only needs 1.75 V? Why does the processor in reality run at 1.8 V? What was going on?

Intel’s Response

I actually called my reliable and very helpful Intel PR contacts in Germany before they could call me. First of all they played the issue down as a minor incident and they blamed my test motherboards and cooling. However, after I asked them to find out if there wasn’t another person who might have had similar problems they got back to me and admitted that Kyle Bennett had reported problems although Kyle had been equipped with the very VC820 motherboard that was supposed to be the perfect platform for the new Intel processor. They started to admit that it might well be that my sample was faulty. Still they raised the following issues:

  • The Pentium III at 1.13 GHz is only shipping to OEMs. Those OEMs are either supposed to use the specially modified VC820 motherboard or to modify any other motherboard they prefer according to the specs of the 1.13 GHz processor. The processor is not available in retail, which is why it wouldn’t have to run on any mainstream platform in the first place.
  • It is not that uncommon that Intel processors might be faulty. For end customers this would not be an issue with the Pentium III at 1.13 GHz however, since this processors would ship to OEMs only, who would have to validate their systems prior to shipping anyway.
  • Intel considers it as exaggeration if I make it sound as if a major amount of the new Pentium III at 1.13 GHz would be faulty. After all that wasn’t proven yet. However, Intel admits that it is of course possible that there is a serious problem, because the opposite hasn’t been proven yet either.
  • Intel requests me to send back my sample so that it can be diagnosed. I responded that I wouldn’t give away my only proof. They said that they unfortunately don’t have another sample that they could send me right now, but I would get one as soon as possible.

Basically I can understand Intel. Of course they want the part back so they can try to find out what’s wrong with it. However, I prefer to hang on to this buggy little bugger, since it is the only thing to back up my story. Of course Intel wants to find the reason for the system crashes before they make a statement, make a new stepping, write a new micro code update or even start retracting their great new product. Still I doubt that any other journalist in my situation would have hesitated to report on this issue.

Bottom Line

Two days ago Intel has released a new high-end processor and already now two samples seem to be faulty. One of them is either still in Kyle’s possession or on its way back to Satan Clara, the other one is in my possession. I doubt that much more than 20-30 test samples were shipped to the press. Two of them makes 7-10% of the lot. If the percentage of faulty parts is just as high in the shipments to the OEMs Intel has got a major disaster at hands.

I just wonder, was that really worth it? Couldn’t the world have done just fine without the new Pentium III 1.13 GHz? Couldn’t have Intel done without another little scandal? Let’s see, maybe Intel is right and Kyle’s as well as my Pentium III 1.13 GHz sample were indeed the only faulty ones. It’s up to you to decide if you believe that and it’s up to Intel to find out what’s the reason behind those buggy parts.

The one fact remains:

Something is wrong with Intel’s new Pentium III 1.13 GHz processor.

Please follow-up by reading Latest Update On Intel’s 1.13 GHz Pentium III.







Вам также может понравиться

Добавить комментарий

Ваш адрес email не будет опубликован. Обязательные поля помечены *

Leading by Speeding: Solano at 166 MHz — THG.RU

e9d99bcd2e7d05b779fab46410348609
e9d99bcd2e7d05b779fab46410348609







BX Outclocked: Solano Shows Its Potential

Intro

The most significant characteristic of high performance computers is a fast system speed. This external clock speed (currently 133 MHz) is crucial for best overall performance still today. Some months ago, we had a VIA 694X and a Intel 440BX platform running at 150 MHz system speed and circumstanciated the performance gains thanks to the higher memory bandwidth. The results clearly indicate that raising the system speed is often more effective than obtaining a faster clocked processor.

Naturally I wanted to determine the performance benefit at 166 MHz Front Side Bus. So far there is no 815-166 (see picture). However, modern mainboards like the Asus CUSL2 support several clock speeds beyond the specified 133 MHz.

Maybe you are disappointed to hear that speeds faster than 166 MHz are not really possible today. Even with highest quality memory modules it is virtually impossible to run the system bus at any higher clock speed.

Demands: AGP

In contrast to the aged BX chipset, Intel’s 815 has been designed for 133 MHz operation. As a result it comprehends all dividers in order to run the AGP within the specs (66 MHz). Remember that both the PCI and the AGP clock are derivated from the system clock speed. As the BX is approved for 100 MHz max, it only knows the dividers 1/1 and 2/3 (for 66 resp. 100 MHz FSB). Overclocked to 133 MHz, the AGP will run at 89 MHz — too much for many graphics cards!

This issue it even more critical at 150 MHz system speed: In this case, your graphics card has to endure 100 MHz (!) AGP speed, which is merely tolerated by GeForce models so far.

The Solano does not suffer from such problems, as the AGP speed does «only» reach 83 MHz even at 166 MHz system clock speed (divider 1/2). According to our experience, this is no problem for the majority of today’s AGP graphics cards.

Demands: PCI and On-Board Components

The whole thing is a bit touchier for PCI components, as the bus will be running at 42 MHz. Timing critical add-on cards like SCSI or network controllers will easily be affected by high overclocking, thus you can almost be sure of running into trouble.

For performance freaks, there is only one way of solving this: Get rid of any hardware which you do not require. That will reduce the number of PCI cards to one or two (sound card and modem/network).

If you depend on specific components, I would stop thinking about 166 MHz FSB right now. The following hardware can cause problems:

  • SCSI adapters
  • IDE-RAID controllers
  • 100Base-TX- or ATM network adapters
  • Firewire controllers
  • High end audio devices
  • Any exotic hardware

Demands: The CPU

Every Pentium III will work reliable at 166 MHz as long as the core clock does not reach lunatic speeds. Using a 166 MHz setup, a Pentium III 650/100 would have to run at 1083 MHz, for example. You certainly do not need to have visionary power to realize that there’s no point of even trying it.

It’s best to use a 133 MHz model for overclocking to 166 (Pentium III 600EB, 667, 733, 800EB, etc). The next table discerns between realistic setups and desperate configurations:

CPU Specs Core speed
at 166 MHz FSB		 Chance of success Core voltage
Pentium III 550E 100 x 5.5 916 MHz Little 1.75 — 1.8 V
Pentium III 600E 100 x 6 1000 MHz Hardly any 1.75 — 1.85 V
Pentium III 600 EB 133 x 4.5 750 MHz Excellent 1.65 — 1.75 V
Pentium III 650 100 x 6.5 1083 MHz No chance —
Pentium III 667 133 x 5 833 MHz Good 1.65 — 1.75 V
Pentium III 700 100 x 7 1166 MHz No chance —
Pentium III 733 133 x 5.5 916 MHz Good 1.7 — 1.8 V
Pentium III 750 100 x 7.5 1250 MHz No chance —
Pentium III 800 E 100 x 8 1333 MHz No chance —
Pentium III 800 EB 133 x 6 1000 MHz Fair 1.75 — 1.85 V
Pentium III 850 100 x 8.5 1416 MHz No chance —
Pentium III 866 133 x 6.5 1083 MHz Hardly any 1.75 — 1.9 V
Pentium III 933 133 x 7 1166 MHz No chance —
Pentium III 1000 133 x 7.5 1250 MHz No chance —

Demands: SDRAM

SDRAM Dimm

Though most PC133 SDRAM modules run stable at this clock speed, they usually cannot be operated at a CAS latency of 2 clock cycles. Many types have to be run at CAS 3, which reduces the system performance by few per cent. At first sight, it does not sound much, but remember that the difference between two processor speed grades (e.g. PIII 650 and 700) is usually about the same!

We have always been using highest quality SDRAM memory from Micron/Crucial, Memory Solutions or Wichmann Workx (Germany), which can be operated at CAS 2. Even when «only» running 150 MHz you may be forced to reduce the CAS latency time from 2 to 3 clock cycles. At 166 MHz, not even one of our different memory modules runs stable at CAS 2.

Well, I think you understand what I am figuring out right now. If an average SDRAM module can only be run at 133 MHz using CAS 3, the timing would have to be increased even more at higher clock speeds, which of course is impossible. Thus forget CL3 SDRAM for this undertaking.

In order to have the system running stable, I had to reduce all SDRAM timings to the slowest options: CAS 2, RAS-to-CAS 3 and Precharge Time 3.

Asus CUSL2

I had two interesting motherboards available for this project: AOpen AX3S Pro and the Asus CUSL2. The Asus finally made it, because the AOpen board operates the main memory at only 140 MHz when running 166 MHz FSB due to some limitation of its clock generator. A comprehensive 815 motherboard round up is scheduled for the next weeks. Than we will also introduce the AX3S Pro in detail.

Asus CUSL2

The CUSL2 is a typical Asus motherboard as it gives you excellent quality teamed with performance and overclocking features. Making use of the ICH2 chip, the two IDE ports are UltraATA-100 compliant. Six PCI slots, AGP 4x and the AMR slot are provided.

The system did not want to run stable at 166 MHz FSB when I installed more than one memory module. Obviously the timing is already at the limit using one module.

Test configuration

Test system
CPU Several Intel Pentium III CPUs
Motherboard Asus CUSL2, Intel 815 Chipsatz
RAM 128 MB PC133 SDRAM, 7ns (Crucial/Micron or Wichmann Workx). CL3 at 166 MHz
Hard Drive Seagate Barracuda ATA ST320430A
20 GBytes, 7200 rpm
Intel UltraATA Drivers Ver. 6.0
Graphics card Asus V6600, nVIDIA GeForce 256
32 MByte SDRAM
nVIDIA Drivers 5.08 for Windows 98
OS Windows 98 SE 4.10.2222 A
Benchmarks and Setup
Office Application BAPCo SYSmark2000
OpenGL Game-Benchmark Quake III Arena
Retail Version
command line = +set cd_nocd 1 +set s_initsound 0
Graphics detail set to ‘Normal’, 640x480x16
Benchmark using ‘Q3DEMO1’
Direct3D Game-Benchmark Expendable
Downloadable Demo Version
command line = -timedemo
640x480x16
Screen Resolution 1024x768x85, 16 Bit
DirectX-Version 7.0a

As already mentioned above, I had to reduce all SDRAM timings to the slowest settings in order to ensure system stability. Again, I used the Asus V6600 GeForce graphics card with SDR memory, so that the benchmark results can still be compared to all the preceding articles:

Benchmarks: BAPCo SYSmark 2000

BAPCo SYSmark 2000

In spite of the slow memory timings I had to use at 166 MHz, SYSmark was still scoring better than at 133 MHz. Of course it’s quite senseless to overclock your system to 166 MHz only for office applications. Let’s take a look at the game benchmarks.

Benchmarks: Expendable Timedemo

Expendable Timedemo

Just as we expected, a 166 MHz system is ahead 133 MHz configurations at the same core clock speed. But the distance is not ample at all.

Benchmarks: Quake III Arena

Quake III Arena

First of all, this results proves again that Quake III is one of the very best allround benchmarks, as all computer components are important for it. Quake III will run more than one CPU speed grade faster on a 166 MHz system.

Conclusion

This time, Solano showed its real potential. Despite the restraining memory timings, the higher clock speed shows an impact. But to be honest, you really have to ask yourself if the effort is worth it, as most PCI cards likely have to be abandoned.

Your software definitely runs faster, but getting SDRAM which really is reliable at 166 MHz could become tough. Actually, overclocking to 166 MHz is likely to become an ordeal due to frequent hang ups caused by inadequate memory. In addition, your PCI components will have to tolerate 42 MHz bus clock.

Last but not least it could become necessary to increase the CPU core voltage in order to make the overclocked system more stable. As you can imagine, this measure does not contribute to system stability as long as you are still playing around.

On the other hand, there are high Quake III results which may tempt one to try it, as the performance gains are quite remarkable. At only 666 MHz core speed, the 166 MHz system performs just like a Pentium III 800/133. To achieve GHz-performance, you only need to overclock a Pentium III 733 to 917 MHz at 166 MHz FSB.







Вам также может понравиться

Добавить комментарий

Ваш адрес email не будет опубликован. Обязательные поля помечены *

Powered by