Slashdot Mirror
K7 vs. Pentium III benchmarks
The Register has published some benchmarks which they got
from an engineer who works at ACER. Personally I would not count on
those numbers because: 1. the motherboard is just a sample and not a final
product, and 2. The OS is Win2K Beta 3 WinHEC APR 99 release. What do you think?
> WinBench99 Version 1.1 (5 times measure)
> CPUMark99
> - K7@600MHz 68.4 Mark
> - Pentium III@600MHz 70.1 Mark
> FPUWinMark (CPU Floating Point Performance does
> - K7@600MHz 2,819 Mark
> - Pentium III@600MHz 3,104 Mark
(1 - 70.1/68.4) * 100 ~= 2.49%
(1 - 3104/2819) * 100 ~= 10.1%
2.49% It's a benchmark. Benchmarks don't measure
real world performance. 2.49% difference is
nothing.
10.1% That is a little more, but once again, it
is just a laboratory benchmark. Plus, the bus
speed on the K6 is signicantly lower.
All sites have problems, but TheRegister and BBC seem to not filter their news very well. Just a week or so ago TheRegister reported that Apple was switching to Intel chips. This has thoroughly been denied by all parties involved and confirmed by none. If I recall, the BBC also had a really fake story a while back, but I don't remember what it was.
Lesson: Don't trust the Register unless you write for them.
Having said that, AMD has a real uphill struggle on it's hands. They are ramping up the K7 on their 0.25um process while Intel is just moving to its 0.18um process for the P-III and P-III Xeon. And it's not just any old 0.18um process, according to a Register article I can't find right now and another recent article on comp.arch it has a lot of stuff that makes it more like a 0.13um process in some ways.
AMD really need to break into the high end on x86, otherwise Intel can keep killing them by pressing them into the sub-$100 space where they can't make any money. I wouldn't mind a K7-500 for under $100 though :-).
I will never understand why Motorola doesn't buy up the x86 part of AMD. AMD is cheap right now, they have a very good x86 design in the K7 and Motorola has the fabs. Are Motorola just going to leave the x86 market to Intel? Are they going to keep betting on the PowerPC to take off? Again and again, AMD has had a good design only to be outspent and outmanaged on the fab front. Motorola has the fab experience, they just need something with a high profit margin to build in them. Take a look at the Intel bottom line, there's money to be made in that business.
By the way: Interesting to see how crippled the x87 architecture really is. This is said to be a similar implementation technology to the Alpha 21264 (done by some of the same people if I recall correctly), yet the floating point performance of the 21264 in its current 600MHz 0.35um process totally smokes current high end x86s including, it seems, the K7 at 0.25um.
Erik Corry, who can't remember his password
This is a fake benchmark perpetrated by a guy who called himself AcermanPS. He lurks around at a Thai webboard called www.pantip.com at tech-exchange section.
He claimed to be working for Acer in Singapore but his IP address comes from Kasetsart University in Bangkok, Thailand.
He is the same guy who, a few months ago, claimed that he can overclock Celeron 300A to 600 MHz using only a fan. When pressed, for details, he defaulted with no proof whatsoever.
Funny, my copy of that Win2k beta has a licence that says that it isn't allowed to be used for benchmarking or performance testing. I wonder if these guys read their licence...
\\'
Yep ... I don't trust theese numbers for some obvious reasons ...
...
... Delay loops ??
... For people who don't know, ALI stands for Acer Labs Inc. so why the hell an Acer engineer is not using an ALI based test system ??! Moreover, the AMD Irongate chipset is not supposed (but don't quote me on this) to support DRDRAM.
... I read some K7 FPU tech docs (well, the few that are available), and there's absolutely no way that the K7 performs worse than a P3. K7 has 3 fully pipelined FP units, P3 has 2 semi-pipelined units, and P3 has more latency on FMUL, FDIV, etc ...
:)
Ok, so a K7 is slower at WinStone than a P3 at the same frequency ?? This would be VERY SURPRISING given that even the K6-III WS scores are significantly higher than a P3 at same clock speed. This would mean that the K7 is outperformed by a K6-III and I just can't belive this.
Let's compare the CPUs die sizes also :
K6-III : 91 sqmm
Pee!!! : about 120 sqmm
K7 : about 185 sqmm
K6-III 91 sqmm includes 64Kb L1 and 256Kb L2, K7 just have 128Kb L1 on die
So, if you got my point, the K7 is more than double the silicon size of the K6, so what do you think they put in
Another strange thing is that the K7 system uses and AMD chipset and DRDRAM. ALI is one of the supposed K7 chipset manufacturers
The FPU numbers are quite funny too
I won't trust any benchmark numbers unless they come from a reliable source, and the register, obviously, is not one (but we already knew it
Doesn't exactly look like responsible reporting to me. Upon a close inspection the system stats look hokey (Especially regarding RAM and HD). And if I understand correctly, ALi and VIA were to be creating the chipset. Additionally, according to AMD the backside L2 cache bus supposedly "scales past 1GHz," so I'd naturally assume that the L2 cache would be core speed, or at least 1/2 of it. And even the size (not to mention the Ultra3 speed) of the IBM hard disk looks fishy. What happened to 4.55, 9.1, 18.2, 36.4... From our favourite news source, even.
This has already been discredited.
Look at it: it talks about USB2.0 (not ratified).
600MHz rdram doesn't exist.
one minute it says 133mhz fsb, next it says 200mhz.
it's a fake.
I use Friend/Foe + mod-point modifiers as a karma/reputation system.
Just because USB 2.0 isn't ratified doesn't mean that can't possibly have test implementations of it.
And there doesn't seem to be any sort of inconsistancy with the FSB rates. They're exactly what every media story has mentioned for the associated chips, 200mhz for the K7 and 133 for the PIII.
Go back and reread the article.
K7: 7th gen cpu, 200mhz, bus, 3 fpu units, huge pipeline, large L1
PIII: ppro w/mmx w/sse and slower cache (6th gen)
Even if the K7 sucked (probably won't) it would outperfrom the Pxxx at the same clock speed by at least 20%. The Register isn't exactly a reliable news source.
-- d'arcy poirot
If I recall the presentatino sheets correctly, the K7's L2 cache is a _backside_ cache and can be clocked at whatever speed they decide to ship it as. The front-side cache is an L3 cache.
From all I have read about the k7, architecturally there seems to be no good reason that its fpu would be slower than that of the Pentium III, ie, more parallel execution, more pipelining, and all the rest. Or the problem could be cropping up in the operating system, god knows microsoft plays enough tricks in its final releases, god only knows what they are up to in thier beta's.
:)
Either way, if these are true I am really dissapointed. I was planning on replacing this machine with a dual K7 system just about as soon as the chips and boards came out. Also if these are true it could spell the end to that small thorn in Intel's side known as AMD, and that would be a real shame. It might just be an american thing about rooting for the underdog, but damn if i didn't want the K7 to come out and just wipe the floor with intel and its crappy PIII adds. (how in god's name does your processor make the internet faster? someone want to explain that one to me?)
If AMD goes down the only real competition I can see Intel meeting is the DEC Alpha.(sorry Compaq, in my heart it will always be a dec) Once everyone realizes that they are going to need to move everything to epic, both on the development side, porting to epic, and on the consumer side, buying all new apps, people might really consider the alpha because atleast in Linux as of now its distributions are stable, its compilers and libraries are getting more and more mature, and no matter what it will be more mature and more stable than merced, just because it will have been around a lot longer. But I digress.
Someone please prove these benchmarks are crap...besides the fact that they came from acer
I would hate to see AMD go down now, getting so close but not quite knocking intel down.
In any good experiment its always nice to see where you might have screwed the pooch.
:)
1. Win 2k beta 3? why in gods name would you try and pass any type of reputable benchmark off on an operating system that isn't done yet.
2. Acer computers...need anyone say any more about that?
3. Intel FUD, acer, being a crappy company that makes crappy computers is just fine with being used as a toy for Intel's PR department. The last time I checked I didn't know acer made anything with an AMD chip in it, but I could be very wrong, I've been building my own machines for a while and don't stop to look at the specs on acer's most recent POS at compusa or wherever they sell them.
4. Timing, last I heard the K7 wasn't in final production yet so its also a beta chip.
Beta OS, beta chip, crappy company, looks like a load of BS to me.
> I'm a hardware research engineer from Acer Inc. (Singapore)
Do we have any proof of this? No.
> K7 test machines CPU K7 600MHz (FSB 200MHz x 3.0) 600MHz 128KB L1 Cache
> 2-way associative 200MHz 512KB L2 Cache 4-way associative EV6 BUS
> Controller L2 Cache Speed 1/3 FSB
Everyone on this message board has mentioned this part spec dozens of times, this could easily be expected.
>RAM TI Direct RDRAM 256MB at 600MHz
I believe Intel RDRAM spec is twin concurrent 16bit busses at 800mhz, not 600mhz. This would support the claims of 3.2GB/sec and multiples thereof with interleaving. Also, Sony has speced their PSX2 as having an 800mhz RDRAM bus. Nintendo RDRAM runs at 433mhz. I don't know of anyone making 600mhz rdram because nobody apparently intends to use it. I believe this is fake.
VGA
> Matrox Millennium G400MAX 32MB SGRAM 360MHz RAMDAC Resolution 1600x1200 32
> Bpp 85Hz
The card's been announced, everyone who read the press release saw it and full specs.
> Harddisk IBM Ultra3 SCSI 6ms 10.0 GB
Nobody makes a SCSI HD in 10.0GB. They're all multiples of 2.255GB. 4.55, 9.1, 18.2, 36.4 are all currently shipping capacities. I don't believe IBM has shown a U3W prototype yet, only Quantum has that I'm aware of. Also, there is no controller listed in this spec, and I haven't yet seen any manufacturer talking about and upcoming U3W controller for anytime this year. Not believable.
> Mainboard AMD Sample Chipset
> Northbridge AMD Irongate AGP4X SouthBridge AMD Cobra ATA66 USB 2.0
Chip names but no numbers. I believe the names were on AMD presentations or discussed at comp shows or shown on roadmaps. We obviously expect AGP 4x. Intel has not even finalized the USB 2.0 spec yet, so support for this in an already existing chipset is nonsense.
> OS-Windows 2000 Beta 3 WinHEC April 99 Release
Nobody in their right mind would run benchmark comparisons on a beta OS.
> P-III test machines CPU Pentium III Coppermine 600MHz (FSB 133MHzx4.5)
> 600MHz 32KB L1 Cache 4-way associative 600MHz 256KB L2 Cache on-die 4-way
> associative
We all expect this part eventually. Nothing new in the specs for it.
> Mainboard Intel Sample Chipset Northbridge
> FW82820 AGP4X Southbridge FW82801AA ATA66 USB 2.0
Chip numbers for this can probably be easily extrapolated from current intel partnumbers. We've been hearing for weeks about delays in the "intel 820" chipset. We know it will support agp4x and ata66 (not that anyone cares about ata66). I'm very confident it will NOT support USB 2.0 as intel has announced that will be coming up for early 2000.
> WinBench99 Version 1.1 ( 5 times measure ) CPUMark99 (CPU Integer
> Performance Not include MMX register) - K7@600MHz 68.4 Mark - Pentium
> III@600MHz 70.1 Mark
This is quite likely made up. The scores on the PIII which could be considered a "known" are way off from projections from standard clocks and results from overclocked systems. Sorry, this is most likely numbers drawn out of a hat and adjusted to make the K7 look bad.
> FPUWinMark( CPU Floating Point Performance Not Include
> MMX,SSE,3DNow!,FSTORE Extendtion register ) - K7@600MHz 2,819 Mark -
> Pentium III@600MHz 3,104 Mark
Again, probably random numbers.
Now I have to ask, how many people on here would not know where to find the press releases, product roadmaps, and rumored info to construct this kind of sham? The accuracy of the description is the only thing that makes it believable to any of us. The more you actually know about this upcoming hardware, the easier it is to fake, because you can get the audience's confidence with known facts and then slip the lies in at the end.
--
--
My sometimes helpful blog