The P4 4.1GHZ Article: discussions here

eden

Champion
I am surprised no one is discussing this!

Now, type of article-wise, I felt this was something some had been asking for, enthusiast articles. It really felt like they went extreme, and the video just proves it. I loved how they pushed it all including CAS 2.

Now, professional-wise, it all started well until the benchmarks. I never had felt that it was THIS bad!
Both from the benchmark results, AND the professionalism.
First the benchmarks: No, this is not a 4.1GHZ performing CPU! This absolutly blows. I think we are witnessing the same thing as XP3400+ OCed AthlonXPs are getting, physical limits of the components. I can't believe THG were beleiving them and praising the results which are so horrid.

Professionally wise I was finally let down completly, the way they just praised and used words of might on each triumph. I think bbaeyens should do another spoof SOON!

--
<A HREF="http://forumz.tomshardware.com/modules.php?name=Forums&file=faq&notfound=1&code=1" target="_new">The THGC Photo Album, send your pics and see others'!</A>
 

tnadrev

Distinguished
Jan 6, 2002
269
0
18,780
i must admit, i did find it odd that they said "rambus is dead" then a couple pages later.. "this cpu won out in the benchmark because of rambus" ... i am not a big fan of rdram... but still... doesn't seem to make sense

(bb || !bb) - Shakespeare
 

ritesh_laud

Distinguished
Nov 16, 2001
456
1
18,780
No, this is not a 4.1GHZ performing CPU! This absolutly blows.
Umm, it actually didn't do badly at all. Performance compared to 3.06 GHz w/HT:

Q3 640x480x16 Demo001...+15%
Q3 640x480x16 NV15Demo..+14%
3DMark 2001 SE..........+8%
Comanche 4..............+21%
UT 2003.................+10%
Lame....................+21%
mp3 Maker...............+24%
VirtualDub..............+16%
Pinnacle Studio.........+21%
Sysmark 2002 overall....+14%
WinACE..................+0%
Lightwave...............+23%
Cinema 4D...............+21%
3D Studio Max...........+22%
SPEC Viewperf DRV08.....+8%
..............DX07......+19%
..............LIGHT05...+20%
..............PROE01....+11%
..............UGS01.....+17%

Not too shabby at all! It scales pretty much like you'd expect. I mean, the maximum theoretical gain would be ~33%, but that never happens because there are always other system bottlenecks: memory, video, FSB, etc.

Ritesh
 

lhgpoobaa

Illustrious
Dec 31, 2007
14,462
1
40,780
I liked the article. Lots of detail.
Good to see the granite bay stretching its legs.
DC PC3200 will rock.

The only thing i missed was the comparison to the XP2600/2700/2800 with the 166fsb.

I still find it mildly amusing though that 4.1Ghz <b>still</b> cannot beat AMD's 2Ghz in a few benchmarks.

<b><font color=purple>[Rik_]</font color=purple> I wonder how many people have made their own phasechange system?
<font color=blue>[LHGPooBaa]</font color=blue> I get phasechange whenever i eat a hot chillie :lol: </b>
 

FUGGER

Distinguished
Dec 31, 2007
2,490
0
19,780
AMD has no answer to this: <A HREF="http://fugger.netfirms.com/sandra.JPG" target="_new">breaching PR8600</A> on a daily basis. Granite bay scores to come.

You are limited to what your mind can perceive.
 

lhgpoobaa

Illustrious
Dec 31, 2007
14,462
1
40,780
hehehe :smile:
I knew you couldn't resist, stickin that knife in deep :wink:

Good to see with toms and your overclock that you DONT use liquid nitrogen, and thus your rigs are actually useful in an everyday sense.

Those N2(l) people are crazy!

<b><font color=purple>[Rik_]</font color=purple> I wonder how many people have made their own phasechange system?
<font color=blue>[LHGPooBaa]</font color=blue> I get phasechange whenever i eat a hot chillie :lol: </b>
 

ZER0

Distinguished
Mar 9, 2001
265
0
18,780
<<<First the benchmarks: No, this is not a 4.1GHZ performing CPU! This absolutly blows. I think we are witnessing the same thing as XP3400+ OCed AthlonXPs are getting, physical limits of the components. I can't believe THG were beleiving them and praising the results which are so horrid.>>>

you must've seen a different article here's a link to the benchmarks.
http://www.tomshardware.com/cpu/20021216/p4_41-12.html
 

Quetzacoatl

Distinguished
Jan 30, 2002
1,790
0
19,780
You know, this is interesting in correlation to some of the rumors going around about the Pentium 4 and Intel's plans for the layout next year. There's been some news floating around that a number of overclocked (highly overvolted, like 1.65-1.75 volts) Pentium 4 systems haven't lasted a year, arguably due to CPU failure. It strikes me as strange that a similar annomaly wouldn't have been found in overvolted Athlon XP systems. Also, the trouble about the scaling, there is a noticeable drop in gain of performance. Perhaps Intel will need to step up to a stable 667 and 800 bus next year at stock voltage just to keep the scaling even. I'm interested to see the possible benefits of the 1mb L2 cache as well as the scaling of the .09um process on the Prescott. The Hammer...just seems less impressive everytime I see it unless the chipset manufacturers do some serious memory timings and improvements on the chipsets for the K8 series. The gap seeming is closing between the Barton and the Hammer, maybe a reason the Barton isn't planned for SOI. Strangely enough, it appears that many the mainstream playforms of the future (P4, Celeron, Athlon, Hammer) are going DC Ddr for the main performing memory setup.

Instead of Rdram, why not just merge 4 Sdram channels...
 

baldurga

Distinguished
Feb 14, 2002
727
0
18,980
Strangely enough, it appears that many the mainstream playforms of the future (P4, Celeron, Athlon, Hammer) are going DC Ddr for the main performing memory setup.
I would say luckly enough! I prefer all systems using the same memory so I can reuse modules if I change my system. Come on, that's good from the consumer point of view!

Still looking for a <b>good online retailer</b> in Spain :frown:
 

eden

Champion
You're really not seeing at all how much it is bad.
From the 3.06GHZ NON HT to the 2.5GHZ for example, in Quake 3 640*, thanks to zero's idiocy and hypocrisy by shooting himself in the foot, there is over 70FPS gained. From the 3.06GHZ NON HT to the 4.1GHZ, there is a mere 50FPS, for a full 1GHZ jump.
NOT TO MENTION, they had made sure the Granite Bay pushed the memory to 200MHZ DC, which is a full 6.4GB/sec, AT CAS2 I remind you, and even other memory timings that are aggressive. They then proceeded to use a very high performing video card, and under Quake III, if I am not mistaken, the programmability is made to make sure CPU and video card are used, and here the MAJOR addition of clock speed and bandwidth yeilded a mere difference. Both the bus and memory were insanely high and using aggressive timings. How on earth you are this blind and fail to see that it did miserably is beyond me.

This is just like the AthlonXP 3400+, it was losing performance in the end. I suspect the P4 and Athlon are reaching their 0.13m physical limits, OR that the paths inside the chip need some updating to handle the new clocks and transfers.
So again, I state, the article was extremly biased in terms of comments (P4 SWEEPS the competition, when it is about a few FPS higher, on a 400FPS scale which would mean few %.)

I of course know this is in no way representative of later performance, obviously, but the fact THG praised their setup's performance so much over the pride of the achievement of making the system is something that really annoyed me.

And I am still not convinced that clock speed scales with performance as well. I am not convinced yet that a 2GHZ CPU will be twice the performance of a 1GHZ CPU under same architecture, just as a 6GHZ CPU is needed to make twice the performance of a 3GHZ. I doubt it is proportional and I will always do, even under idea conditions, removing any bottlenecks.

EDIT: Wrote juin instead of zero, re-edited it, my bad Juin.
--
<A HREF="http://forumz.tomshardware.com/modules.php?name=Forums&file=faq&notfound=1&code=1" target="_new">The THGC Photo Album, send your pics and see others'!</A><P ID="edit"><FONT SIZE=-1><EM>Edited by Eden on 12/17/02 04:11 PM.</EM></FONT></P>
 

eden

Champion
I am also surprised about the voltage problem not happening to AMD so far.
I am surprised as well that THG seems to be so proud at 1.725V but seems to forget that their NW is a potential sudden death target.

--
<A HREF="http://forumz.tomshardware.com/modules.php?name=Forums&file=faq&notfound=1&code=1" target="_new">The THGC Photo Album, send your pics and see others'!</A>
 

Copenhagen

Distinguished
Oct 21, 2001
552
0
18,980
I can't believe THG were beleiving them and praising the results which are so horrid.
I absolutely agree. It's like they're focusing on the achieved clockspeed, rather than the resulting performance. Generally I like these kind of reviews where systems are pushed to the edge, and this review was supposed to have shown impressive results, but somehow the expected results didn't show up.

I think the Granite Bay chipset is to blame. It really looks like a bad performer to me. It overclocks extremely well, granted, but it doesn't really gain from it, so what's the point. Given that the memory speed and FSB are synchronous (bus run at 780MHz, close to Springdale specifications) it's surprising to see the how bad it scales. I fully understand that Intel is working hard on Springdale because Granite Bay is certainly a disappointment. It also shows that the P4 Northwood really needs achitectural improvements in order to provide a substantial increase in overall execution speed. The increased L2 cache of Prescott is more than welcome.


<i>/Copenhagen - Clockspeed will make the difference... in the end</i> :cool:
 

eden

Champion
Yes but it's more in the physical limits of the paths or traces, that I think is the reason for horrid scaling.

I also like these kinds of articles, enthusiast oriented, but god, the pride of doing such a system is great and really there, but they threw that away by praising the results more, especially when these were absolutly disappointing!

Oh and I really thought that was one fanboy comment when they yelled Rambus is dead.
I thought this was the key phrase to fire the authors away and destroy THG's rep!
To yell such comments is unprofessional, against a company and really outputs bias of the writers and image of THG. This is bad, VERY bad in fact. It's like if the P4 lost against the new CH, and THG yells: Intel is finished!!!

--
<A HREF="http://forumz.tomshardware.com/modules.php?name=Forums&file=faq&notfound=1&code=1" target="_new">The THGC Photo Album, send your pics and see others'!</A>
 

slickstaa

Distinguished
Apr 7, 2002
406
0
18,780
Oh come on guys! I thought the article was great! It wasn't meant to predict future performance exactly, just to do something out of the ordinary and show off their OC'ing abilities.

<font color=blue> If it ain't broke, don't fix it...tweak it.</font color=blue>
 

tnadrev

Distinguished
Jan 6, 2002
269
0
18,780
AMD has no answer to this: breaching PR8600 on a daily basis. Granite bay scores to come
not sure if this is a joke or what, but you realize the Sandra PR ratings have no correlation to the AMD PR ratings?
i don't have a copy of 2002, which i assume that was in, but if i recall, my AMD 1600+ rated as about PR4900... at stock, so yeah, stfu

(bb || !bb) - Shakespeare
 

ritesh_laud

Distinguished
Nov 16, 2001
456
1
18,780
NOT TO MENTION, they had made sure the Granite Bay pushed the memory to 200MHZ DC
Actually it wasn't 200 Mhz, it was 175 albeit at CL2. Look at the test setup again. The future P4 will be on real DDR400 technology at 200 MHz but perhaps dropping to CL3. I'm not sure why THG made a big deal of DC DDR400 in this test, because their benchmarks weren't run at 200 MHz.

From the 3.06GHZ NON HT to the 2.5GHZ for example, in Quake 3 640*, thanks to zero's idiocy and hypocrisy by shooting himself in the foot, there is over 70FPS gained. From the 3.06GHZ NON HT to the 4.1GHZ, there is a mere 50FPS, for a full 1GHZ jump.
You're comparing apples to oranges. Quake 3 is largely memory performance limited. Though it has greater bandwidth, dual-channel DDR-SDRAM at 175 MHz does not always perform better than PC1066 RDRAM as you can clearly see from the WinACE compression benchmark. The same concept is probably working against the Granite Bay setup here in Quake 3.

I doubt it is proportional and I will always do, even under idea conditions, removing any bottlenecks.
Of course it is proportional. The maximum theoretical difference in pure CPU performance should be 4025/3067 = +31.2%. Look at the SiSoft CPU benchmark comparing the 4025 MHz to the 3067: 38% higher for Dhrystone, 23% Whetstone, 30% Integer, 31% FP. That's pure transistor speed there, virtually no other bottlenecks. Nearly perfect scaling. The minor differences can be accounted for by hyperthreading doing funny things to the execution.

I think you underestimate the constriction of memory and video bottlenecks. The P4 scales quite well at 4.1 GHz and if the memory performance were improved to 200 MHz it would do even better. Also, there are some apparent differences architecturally between RDRAM and DC DDR-SDRAM that allow RDRAM to still come out on top in certain apps, despite lower bandwidth capability.

Ritesh
 

imgod2u

Distinguished
Jul 1, 2002
890
0
18,980
From the 3.06GHZ NON HT to the 2.5GHZ for example, in Quake 3 640*, thanks to zero's idiocy and hypocrisy by shooting himself in the foot, there is over 70FPS gained. From the 3.06GHZ NON HT to the 4.1GHZ, there is a mere 50FPS, for a full 1GHZ jump.

The 4.1 had HT enabled, which reduces performance somewhat in Q3A (according to Toms only btw). So the actual increase for both HT enabled processors from 3.06 to 4.1 was 58.2 fps in the case of demo001 at 640x480x16. It should also be noted that the 3.06 was put on a rambus platform while the 4.1 was on a Granite Bay platform, which doesn't seem to perform as well.

NOT TO MENTION, they had made sure the Granite Bay pushed the memory to 200MHZ DC, which is a full 6.4GB/sec, AT CAS2 I remind you, and even other memory timings that are aggressive. They then proceeded to use a very high performing video card, and under Quake III, if I am not mistaken, the programmability is made to make sure CPU and video card are used, and here the MAJOR addition of clock speed and bandwidth yeilded a mere difference. Both the bus and memory were insanely high and using aggressive timings. How on earth you are this blind and fail to see that it did miserably is beyond me.

The FSB and memory were synchronized at 175 MHz QDR and DDR respectively. That is a 31.6% increase in theoretical memory bandwidth. In order for a microprocessor to maintain a linear scaling (i.e. a 33.6% increase in clockrate, as is the case for 3.06 to 4.1, you get a 33.6% increase in performance), you'd need an equal scaling of all other critical components. And with a game like Q3A, the memory bus is the critical component.

This is just like the AthlonXP 3400+, it was losing performance in the end. I suspect the P4 and Athlon are reaching their 0.13m physical limits, OR that the paths inside the chip need some updating to handle the new clocks and transfers.

That's rediculous. Processor trace paths limit clockrate. Everything inside an MPU <b>must</b> follow the clock timing or you'll have an unstable MPU. What you're suggesting is that certain components fail to meet their timing and perform slower than 4.1 GHz. If this were the case, the system would crash and you'd get a BSOD in Windows. The fact that the chip runs stable at 4.1 GHz means it is reaching that timing and all components (and paths) are working up to spec.

And I am still not convinced that clock speed scales with performance as well. I am not convinced yet that a 2GHZ CPU will be twice the performance of a 1GHZ CPU under same architecture, just as a 6GHZ CPU is needed to make twice the performance of a 3GHZ. I doubt it is proportional and I will always do, even under idea conditions, removing any bottlenecks.

Provided that the processor is the only critical component, this would be true. However, there is no realistic case of such. The processor cannot gain instantaneous access to memory (like it should) nor can it fetch instructions and data instantaneously and send it to the graphics card instantaneously. Without the ability to achieve this, it will always be limited as far as scaling. You can pump up the bandwidth all you want, but if access commands take 3 clocks to reach memory, and then 2 clocks at best for the data to <b>begin</b> its way to the processor, you'll still have quite a bit of lag time.

As someone else noted, the setup did scale around 20% on average. Not bad considering a 33.6% boost in clockrate. Although the test results were pretty much useless in judging such. It used different setups (including different motherboards and memory types) for differently clocked processors. It didn't bother to disable HT on the 4.1 and in doing so, skews the results in ways we will never know in comparison to older models. It was more of a "look what I have done" article. In reality, it proves nothing about the scalability of either the P4 or Athlon.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.
 

lhgpoobaa

Illustrious
Dec 31, 2007
14,462
1
40,780
Did anyone else notice that the 4.1Ghz still doesnt beat AMD's 2.1Ghz in ALL benchmarks.
With close to 100% clock speed difference i owuld have thought it would at long last.

<b><font color=purple>[Rik_]</font color=purple> I wonder how many people have made their own phasechange system?
<font color=blue>[LHGPooBaa]</font color=blue> I get phasechange whenever i eat a hot chillie :lol: </b>
 

Spitfire_x86

Splendid
Jun 26, 2002
7,248
0
25,780
This review opens our eyes that how much it is justified to buy P4 to overclock. Many people prefer P4's overclocked price/performance ratio than AXP's stock price/performance ratio. I never beleived an overclocked P4 2.4 GHz @ 2.8 Ghz/3 Ghz with i845PE and DDR333 would really perform like PC1066 RDRAM + 2.8/3 GHz P4. These overclocked systems covers the performance gap between Rambus and DDR, then give some extra performance.

Also this review reveals how much bad granite bay is. Dual DDR400 and 700 MHz FSB gives memory performance theoritically equal to i850E + PC1066 RDRAM; practically it falls behind of i850E.

This article is another bad piece coming from THG. If this kind of things conitinue to come in every month......, I don't like to think. Tom should seriously think about his site. He should write most of the important article as he used to write before. Start reading THG articles from '95, then you will see a sudden THG quality drop since last year. The reason is simple, Tom isn't writing regularly.

Let us know <A HREF="http://forumz.tomshardware.com/community/modules.php?name=Forums&file=viewtopic&p=25703#25703" target="_new"> What File compression format you use? </A>
 

Quetzacoatl

Distinguished
Jan 30, 2002
1,790
0
19,780
Call me biased, but at $200, i'd rather get the 2400+ than a 2.4Ghz or 2.53Ghz and overclock them. AMD has a much better chipset now, the Nforce2 which is the perfect pair with DC DDR and a 2400+. Personally, I like the Granite Bay chipset. Provides a noticeable gain in performance over the i845pe/ge and is VERY comparable to the i850e. This overclocking potential from the new chipsets are even greater with the C1 stepping of the Pentium4, a good advancement of the processor.

Instead of Rdram, why not just merge 4 Sdram channels...
 

eden

Champion
The 4.1 had HT enabled, which reduces performance somewhat in Q3A (according to Toms only btw). So the actual increase for both HT enabled processors from 3.06 to 4.1 was 58.2 fps in the case of demo001 at 640x480x16. It should also be noted that the 3.06 was put on a rambus platform while the 4.1 was on a Granite Bay platform, which doesn't seem to perform as well.
Yes I can see that, except that under the tests we've seen, GB was still able to tie Rambus occasionally. Yet here we see that even CAS2 timings, and I still mention againa aggressive ones, are not helping.




The FSB and memory were synchronized at 175 MHz QDR and DDR respectively. That is a 31.6% increase in theoretical memory bandwidth. In order for a microprocessor to maintain a linear scaling (i.e. a 33.6% increase in clockrate, as is the case for 3.06 to 4.1, you get a 33.6% increase in performance), you'd need an equal scaling of all other critical components. And with a game like Q3A, the memory bus is the critical component.
You are the one saying that the bus is a critical thing for Quake III. The R9700 PRO has yet to flex its muscles in that benchmark btw and it would seem to me more CPU speed would still yeild high results and bottlenecks are not there yet. So please explain why here we do not see any advantage? You are 1GHZ higher than the current market processor, you have a higher bus speed, YET you yeild a small improvement?
I am really starting to wonder at why some are even arguing against this, the scaling was horrible, it is clear and there. And I know it is an OCed P4. But I am ranting this far because of how THG's writers of the article added such comments, they seemed to think it was so spectacular. This is pure bias, the P4 4.1GHZ OC system shows that it is not performing up to speccs, despite having better memory timings and higher bandwidth.

That's rediculous. Processor trace paths limit clockrate. Everything inside an MPU must follow the clock timing or you'll have an unstable MPU. What you're suggesting is that certain components fail to meet their timing and perform slower than 4.1 GHz. If this were the case, the system would crash and you'd get a BSOD in Windows. The fact that the chip runs stable at 4.1 GHz means it is reaching that timing and all components (and paths) are working up to spec.
Did you see the benchmarks?
Everyone saw them, everyone knew there was something inside the AthlonXP that was not working anymore right at the high clock speed of 2.66GHZ. Unless you'd tell me it's normal. The XP3400+ had a higher bus and memory, so that is not the bottleneck. It seems clear to me that it has something inside that needs fixing. It's not like a silicon can perform anytime at top speed no matter what frequency you set it at.
To me it sounded there was a possibility the P4 is also having such silicon problems and that perhaps the architecture needs some tweaking for higher speeds. Is it so much impossible for that not to happen?

As someone else noted, the setup did scale around 20% on average. Not bad considering a 33.6% boost in clockrate. Although the test results were pretty much useless in judging such. It used different setups (including different motherboards and memory types) for differently clocked processors.
I still see it as horrid performance. If I OCed my P4 to 4.1GHZ with such expensive cooling, and aggressive memory timings yet found out my scores merely jumped by 20%, I'd be so disappointed, I wouldn't bother for a refund, I'd just throw the whole setup out the window and depress myself. Would you pay over 1000$ for the CPU setup alone and get a mere improvement and then still try to justify yourself as that being normal?

Provided that the processor is the only critical component, this would be true.
Were you agreeing with me that it isn't necessarily proportional? Unless someone would prove me that you really need 6GHZ to double the performance when upgrading a 3GHZ CPU. And I am fully aware of bottlenecks and all, but to me it'd seem as generations advance (nowadays a P4 2.4GHZ can be twice more performing than a 1.2GHZ Tualatin, so it seems to me everything is coming back to normal)the IPC does as well, so a 6GHZ CPU could as well be over 3 times more performing than the 3GHZ one. So again, it's not proportionally, because otherwise, the physical limits would really create a problem, because to double a 10GHZ's performance, we'd need 20GHZ and I doubt any pipeline can manage that in a few rounds of advancements and scaling.

It didn't bother to disable HT on the 4.1 and in doing so, skews the results in ways we will never know in comparison to older models.
I agree on that one, perhaps, just perhaps, HT enabled played a huge role here. After all the HT enabling under Q3 did take off 9FPS at some point. But I somehow have doubts about this one.

It was more of a "look what I have done" article.
Which I have said many times, but I wish it were! Unfortunatly it was shadowed by the benchmarking part that just removed all dignity in calling it that, an enthusiast article. Do you at least see what I am saying?

In reality, it proves nothing about the scalability of either the P4 or Athlon
Which I also have been saying and keeping in mind, however I am looking at it objectively here based on the comments (on THG and THGC) that seemed to make those results look amazing and normal!

(not directed to you)
Please people, I said this too many times, if this article was purely about the achievement and the enthusiast content, it'd been praised by me a lot, and it'd have been a classical THG one. BUT the fact they added such unprofessional comments after each benchmark literally trashes the initial goal and really destroys the article. It is sad that it was this way, really. No pride in the end.

--
<A HREF="http://forumz.tomshardware.com/modules.php?name=Forums&file=faq&notfound=1&code=1" target="_new">The THGC Photo Album, send your pics and see others'!</A>
 

LED

Distinguished
Apr 28, 2002
511
0
18,980
And I am still not convinced that clock speed scales with performance as well. I am not convinced yet that a 2GHZ CPU will be twice the performance of a 1GHZ CPU under same architecture, just as a 6GHZ CPU is needed to make twice the performance of a 3GHZ. I doubt it is proportional and I will always do, even under idea conditions, removing any bottlenecks.
That's like saying a car going 60mph wont reach a destination twice faster than the same car, on the same road going 30mph.........

Cache size may be the reason the P4 doesnt scale progressively by mhz.

[-peep-].