[Prev][Next][Index][Thread]

Re: Cyric 6x86



On Tue, 30 Jul 1996, Ofer Maor wrote:

> I am standing in front of a big hardware upgrade, and while browsing the net 
> I saw a lot of refferences to the Cyrix 6x86 CPU, which according to 
> benchmarks and reviews is supposed to be quite good. however, as we all know 
> - all reviews and becnhmarks work generally for MS stuff.. and I heard some 
> rumours that the Cyrix 6x86 on Linux makes some probs, (some VICIOUS rumours 
> say it makes linux resemble MS stuff ;-)

half on topic on linux-il, so I am crossposting this in il-board, where I
know a few Hardware mavens lurk.

www.cyrix.com is the first place to look and the last place to trust, I
hope you payed your visit...

it tells you it's a super-scalar CPU (like P54 and P55) and that it is
compatable with Linux. However I heard from at least one guy it doesn't
have a 1-1 Opcode compatibility with more than the 486, so _maybe_ Pentium
optimization will not work.

> C. Are the performances of the Cyrix (either pentium optimized or 486
> optimized) PXXX+ compared to a standard intel PXXX better or worse? 

humm... I think you'll have to ask that in a Linux hardware Usenet group.

> 
> D. Can I run a Cyrix in good preformances on a Triton II 430HX
> motherboard, with no problems? (Tyan TomCat I is my preffered option
> at the moment)

why not get a good Iwill with 430VX? the HX is not even as good as the FX,
as far as I saw.

and now back to the Cyrix: Architectural Comparison

--------------------------------------------------------------
                           6x86        Pentium Pro   Pentium
                           Processor   Processor     Processor
--------------------------------------------------------------
Full x86 Instruction Set
   Optimization               X
Superscalar                   X           X             X
Superpipelined                X           X
Register Renaming             X           X
Data Dependency Removal       X           X
Multi-Branch Prediction       X           X
Speculative Execution         X           X
Out-of-Order Completion       X           X
80-Bit Floating Point Unit    X           X             X
16K Primary Cache             X           X             X
--------------------------------------------------------------
humm... Cyrix keep on claiming they are fully compatable...

speed differences? depends...

=========================================================================
              Ziff-Davis       Norton SI       Power Meter     Landmark
              Winstone® 96     Ver. 8.0*       MIPS 1.7*       2.0 (CPU)*
=========================================================================
6x86-P200+       91.6           1020.4           139.2           1813.9
200 MHz Pentium  89.0            632.0           139.2           1151.2
-------------------------------------------------------------------------
6x86-P166+       86.7            901.4           123.5           1604.8
166 MHz Pentium  82.7            525.8           114.6            957.7
-------------------------------------------------------------------------
6x86-P150+       81.9            814.5           111.4           1447.9
150 MHz Pentium  77.6            475.4           102.6            865.9
-------------------------------------------------------------------------
6x86-P133+       76.6*           746.6           102.1           1327.2
133 MHz Pentium  76.0            420.6            92.8            766.2
-------------------------------------------------------------------------
6x86-P120+       71.7            678.7            92.8           1206.5
120 MHz Pentium  70.9            380.3            82.9            692.7
=========================================================================
+‡ Source: MicroDesign Resources (MDR) Labs.
* Source: Cyrix Corp.

from the FAQ:

Q. How does it differ from the Pentium® processor? 

A. Most significant are the architectural differences which result in the
6x86 processor's superior performance gains. While both the 6x86 and
Pentium® processors are superscalar and contain an 80-bit floating point
unit and a 16-KByte primary cache, other architectural features are found
only on the 6x86. The 6x86 processor's integer and floating point units
are optimized for maximum instruction throughput by using advanced
architectural techniques including register renaming, out-of-order
completion, data dependency removal, branch prediction and speculative
execution. These design innovations eliminate many data dependencies and
resource conflicts to achieve high performance.

Q. What is the difference between the Cyrix 6x86 and 5x86 processor? 

A. There are significant differences between these two Cyrix processors.
The 5x86 processor, formerly the M1sc, is pin-compatible with a 486. 
However, the 5x86 uses many fifth-generation features (branch prediction,
data forwarding, superpipelining, etc.) to achieve performance equal to
Pentium processors. The goal for the 5x86 was to achieve an efficient
design, obtaining the highest performance possible using the fewest
transistors to obtain entry-level Pentium processor performance. A 120 MHz
5x86 performs on par with a 90 MHz Pentium processor. 
The Cyrix 6x86 processor is a high-performance, sixth-generation CPU that
is P54C socket-compatible. It incorporates a 64-bit external data bus
(5x86 use a 32-bit external data bus) and a variety of innovative
architectural features that enable it to outperform higher-megahertz
Pentium processors to achieve best-in-class performance. Additional design
information can be found in the 6x86 Processor Brief.

(Cyrix also claim that the M2 will have MMX, and you will be able to
upgrade it for a tiny fee... yeah, and my real name is Mr. HotPotato...)

(ok, now look at THIS wierd part:)

Q. Why is there a plus mark in each 6x86 processor name? 

A. For the 6x86 processor, a P+ designation is used to indicate better
performance.

                Winstone 96 6x86-Pxxx+ > Winstone 96 Pentium-xxx
                          xxx = P-rating or Pentium megahertz

     For example, a 6x86 processor with a P-rating of "P166+" (regardless
of processor name and clock speed) indicates performance faster than a 166
MHz Pentium processor. The plus mark indicates performance that
consistently exceeds the same Pentium processor megahertz level.

(humm... reading on:)

Q.What board MHz setting do I use for each P-rating? 
  A.P120+ = 50MHz CPU bus with a 2X clock (50/100MHz)
    P133+ = 55MHz CPU bus with a 2X clock (55/110MHz)
    P150+ = 60MHz CPU bus with a 2X clock (60/120MHz), same as Pentium-120 
    P166+ = 66MHz CPU bus with a 2X clock (66/133MHz), same as Pentium-133 
    P200+ = 75MHz CPU bus with a 2X clock (75/150MHz) 

(and here comes the biggest wierdy:)

Q. When I run 3D Studio and Clipper on my 6x86-based PC, the programs are
stalling. What's the solution? 

     A. Both 3D Studio and Clipper Applications use software timing loops
in the code. The 6x86 processor executes these loop instructions faster
than previous x86 CPUs which interferes with timing dependent code in both
software programs. Solution: Download the pipeloop.exe file and put it in
your autoexec.bat file. This utility inserts a delay in the loop
instruction. The utility is not memory resident. NOTE: This file may not
correct all timing issues.  Other solutions will be listed when available. 

(humm! sounds like a probable confuser for Bogomips...)

and now for a quick spin through DejaNews:

> 
> I know on some compilations for programs the make file queries the cpu
> to decide which source code to use.  Does the cyrix chip answer as if
> it is a intel chip?
> In essence has anyone had any problems running linux on a Cyrix 6x686?

I had no problems running Linux on my 6x86/120 MHz at all. I've compiled
the kernel for Pentium and it runs just fine. All that bothers me (not
too much of course) is that the processor is recognised as i486.

Which is immidiately explained by Cyrix' FAQ again:

Q. When I run Windows 95 on my 6x86 computer, the Windows 95 control panel
system information identifies the CPU as a 486. Why?
 
     A. The algorithm used in Windows 95 to detect the CPU was completed
before the Cyrix 686 was released and therefore the Cyrix 686 responds to
the algorithm just as a 486 does. To verify that your system is using the
6x86 microprocessor, please reference your computers BIOS startup screen. 
In addition, the following software utilities identify the 6x86 CPU
correctly: Diagsoft QA Factory 6.02 and QA Plus/FE 5.42; Quarterdeck
Manifest 4.01; and PC Doctor Rev 1.5.162 from Watergate Software. Cyrix
will update this site with new software utilities that can detect a 6x86
CPU in the future.  Please note that the 6x86 processor was certified by
Microsoft for Windows 95, and Windows 95 performance results are
available.

so I guess it confuses Linux too (for now).

and back to DejaNews, here are a few more un/satisfied people:

--------------- 
I am using a 686 P150+ pushed up to 133MHz clock (P166+)
and it flies along.  There are supposed to be shortcomings with the FPU
unit (i.e. supposedly slower than the equivalent intel Pentium)  but it
seems remarkably fast to me.  I am going to have a crack at writing some
FPU test progs myself to compare. 

Total Linux kernel re-build (i.e. make clean etc... to make zImage)  in
under 4 minutes (yes it astonished me too!)  -used to take 20+ on my old
486/66 (admitedly, I have a new Triton VX motherboard amd 16Meg EDO RAM) 

        I don't know why there is such a price jump from the P150 to P166,
after all there is not that much of an improvement (11%)
---------------
686 doesn't run quake nearly as well as the equivilent pentium,
due to the poor performance of the fpu... or so most people say

that would go for any floating point math intensive tasks also
---------------  (and the last one from a few pros, note the newsgroup)
Newsgroups:   comp.os.linux.development.system

>>>> What's the best optimization setting for the Cyrix M1/6x86
>>>>processor in the Linux kernel? Pentium or Pentium Pro?

>>>i486/i386.  x86 clones are just that clones of x86 processors.
>>>Pentiums are not cloneable because they have a Trade Marked Name and
>>>command set that Cyrix and AMD can't use.

>> That's not right.  It's not a matter of intruding on some trademark.
>>The Cyrix 6x86 line benefit from having the kernel compiled with the
>>processor type set to PPro.  It's got the same kind of superscalar architecture
>>and benefits from the optimizations made for that chip.

> Not as easy as either answer, I would suspect.  From what I understand,
> the Cyrix 686 is:
>  a) completely Pentium compatible (all instructions included), but
>  b) optimized to run 486 code.

> Intel can't copyright or patent the instructions themselves, only the 
> way they are implemented on the chip.  The only thing restricted by
> Intel is the name itself (they were told they couldn't trademark a
> number).

> So, while you could tell your compiler to use Pentium instructions, the
> Cyrix design will actually improve 486 code the most.
> This is the indication given by most of the benchmark tests I've seen.
> BTW, it's FP processor is also not quite as good as the Pentium, so...

Isn't the PPro also optimised to improve 486 code the most?

I have found on my new PPro that the redhat binary of gzip is about
10% faster than the binary I compiled with -O3 optimisation using PGCC
2.7.2p-pl9.
----------------------------------------------------------------

To sum all this up, I think I would buy one for testing on a station, try
Ppro, Pentium and 486 optimizations, run benchmarks and crashme, and see
what I end up with.

if anyone is willing to donate me the hardware, or loan it for a week,
I'll be glad too :-) I only need the CPU since my nice new Iwill P55/TV
can handle all of the Cyrix line.


   -------------------------------------------------------------
   Ira Abramov          <ira@scso.com>        Scalable Solutions
   SITE Web Presence ("webspace for rent") http://www.site.co.il
   Beeper 48484 at 03-610-6666, 02-294-666    FAX (972)2-430-471
   POBox 3600, Jerusalem 91035, Israel        Tel (972)2-6426822
   http://www.scso.com/~ira   Check out: http://www.linux.org.il


References: