Posted on 12/06/2002 2:54:57 AM PST by JameRetief
Truly remarkable results gobsmack us
A demo on its site – it requires Flash to view – compares a spreadsheet calculation. That takes five minutes 39.9 seconds on a 3.06GHz Pentium 4 without hyperthreading while with hyperthreading, Intel claims it takes 35.58 seconds. [ Actually, if you see the demo, it is supposed to take the non-hyperthreaded P4 an even longer 6 minutes 15 seconds. (5:40 + :35) ]
This is a remarkable feat and leads us to believe that hyperthreading is better than anyone could possibly have anticipated.
Indeed, with the performance of your processor boosted by over 900%, truly no-one should be without one.
You can run the demo from this Intel page, while the Ace's thread starts here. µ
IMO, the later ones were weak re-hashes of the characters. If you've read the first half, the second half can be given a pass.
WTF are they calculating ?? All statistics on EVERYONE on the planet ?
I've never seen a spreadsheet take more than a second or two to load.
"My AMD XP2000 kicks the stuffing out of most Intels up to 2.3GHz. (it runs a modest 1.81 GHz)
The old Intel/AMD numbers games.
Tell you the truth ~ I had a really nice 1GHz AMD I built last year with 512mb SDram. So, I got a bit of cash back from the IRS, and built this new one. 1.53GHz / 512mb DDRram running the fastest mem settings in the BIOS.
REAL WORLD DIFFERENCE? It took me 23 minutes to install WinXP Pro on my new one. It took 24 minutes to install it on the 'old' one.
Personally, if I were to build right now, I'd get an Epox motherboard, at least 512 megs of fast solid ram, and an AMD XP2000 [co-incidentally, that's what I DO have now) or (if I were going to be bleeding edge) an XP2700 on a new 333MHz board. Again, the speed difference is neglible, but at least I'd be cutting edge.
Once we reached the 1GHz mark, it seems the actual speed increase became more 'on paper', as in, SiSoft, 3DMarks, PCMarks benchmark results. I have pages of them from my own system. According to the numbers, I'm flying. According to realworld ~ It's ok. FreeCell disappears after the last card is played, that's all I care about. hehe."
My cheepo computer that I would build today would be a board with 512mb of PC133 and an Athlon XP. With the Athlon 1600+ and 2100+ being the best buys. Not too many boards run pc133 and Athlons but they're out there. One of my computers has 512mb/2100 DDR. But 256mb/pc133 modules can be gotten cheap and even cheaper on sale.
Once we reached the 1GHz mark, it seems the actual speed increase became more 'on paper', as in, SiSoft, 3DMarks, PCMarks benchmark results.
Did you see much difference between 512 and one gig of memory? With windows 2000 or XP?
Unfortunately, no. Keith Laumer, the author of the Retief series died in 1993. There have been a few posthumous releases though. Hopefully, there will be more unpublished works released.
I'm pretty sure my next build will have on-board USB2 and firewire. Fewer cards the better.
If you tried to run, say, Doom 3 and Halflife/Counterstrike at the same time on a regular 3 GHz machine, the computer wouldn't be able to handle the two games at all.....or, if they did run side-by-side, they would run increadibly slow....too slow to be playable. But if the two games were coded to hyperthread, (which is very similar to coding for dual processor motherboards) and were running on a hyperthreaded platform, it's entirely possible that you'd actually be able to play them both at the same time.
Currently, hyperthreading is too expensive for the desktop market but, over the next few years, expect to see two-way and four-way servers replaced with hyperthread processors which, is actually cheaper than the currently situation.
That is a totally inaccurate statement....especially the "3DMarks" part. If you're running a high-end card like the GF4600 or Radeon 9700 a 1 GHz processor kills your 3D performance. If, for no reason other than the front bus speeds.
.
ASUS A7N8X nForce2 Technology Review
... it's probably better to view this as a performance gauge rather than a full-on review
of the ASUS A7N8X which will ... Hypertransport for maximum system throughput ...
www.3dvelocity.com/reviews/nforce2/nforce.htm - 42k - Cached - Similar pages
I don't see how. As I understand it, hyperthreading puts two sets of registers on the processor, allowing it to execute two threads simultaneously. The best-case performance increase I'd expect to see is 100%; 900% is ludicrous.
That would depend entirely on what you're doing with the computer: if you are running Enterprise sized databases, or rendering abnormally large graphics files, you would see an increase in performance however, it would be slight because at that point you're usually maxing out the CPU and you probably wouldn't be multi-tasking anyway.
For day to day use, such as web browsing and working on Office documents, you'd find no perceivable increase. Although, you'd notice less swapping when returning to the OS from games - if you were running huge-environment games like Battlefield 1942. But for running games like Tribes II or even Americas Army, you'd see very little improvement in swapping back to the OS.
Where you would really see a speed increase would be if you were to upgrade the Motherboard to a DDR400 or DDR333 memory system. Going from a PC133 system to a DDR400 system is amazing but, would probably be a waste of money for the average user...currently, I'd only suggest that kind of upgrade for a gamer or graphics artist.
One of the major principals behind hyperthreading is that the 90% which is being claimed by the application, but not actually being used, can be siphoned back to other applications.
That is why a 900% increase is plausible, but only if Intel were running multiple, CPU-intensive applications during the test. If Intel is claiming a 900% increase of speed while running ONE application, I don't believe a word of it -- that news would be plastered all over sites like Anandtech.com and DeviantPC.com in 48 Point, Bold font.
For Office applications, 512 is probably overkill. I'm running a really big Access database right now -- hundred tables, several hundred defined queries, several hundred reports and forms -- and only using 181 megs.
On the non-hyperthreaded processor they probably ran the SAS software on high priority and then concurrently ran the Excel calcs on low priority in the background. With only one logical processor it would take the Excel software an exagerratedly long time to compute.
By using the hyperthreading processor, the second logical processor is able to make the difference that would "account" for the 1000% increase.
If this were a valid way to benchmark, Intel would have made the case for dual processor systems the same way. You would see an even greater and more outrageous increase if you ran the same test on a dual (non-hyperthreaded) processor system. There is no way to reduce the processing time by over 90% by simply adding another processor, either physical or logical, unless the test has been rigged.
Let's say you were working on 800 Mb of data, perhaps a big 3D model, or a very high res image, or a monster spread sheet, or indexing your Magnum Opus in Microsoft Word. With 512 Mb of memory, you're running at disk speeds, which is perhaps 10 milleseconds per block (seek time). With 1 Gb memory, you're running at memory speeds, which is perhaps 100 microseconds (roughly 1000 loads from 100 nanosecond memory).
That's the difference of a 100 to 1.
For accessing a single word, main memory is perhaps 100,000 times faster than disk.
So don't ask how much faster a system with 1 Gb is than a system with 512 Mb. Rather ask how much memory your application and data need, and size your system to have that much.
Having more memory than you need makes almost no difference at all. Having less memory than you need, even just a little bit less, can cause huge performance losses.
But if it's plausible then smart coders will take advantage of the circumstances under which the 900% increase was achieved.
>:)
Although, you're misunderstanding the technology...Hyperthreading is not simply a case of "Single Hyperthreaded CPU = Duel Processor". Even two-way and four-way systems aren't as fast as they should "theoretically" be. You don't automatically get doubled-performance by adding another processor, even with the best SMP apps.
An excellent article explaining Hyperthreading is at Anandtech.
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.
