Future of Athlon
In our next article, we will cover the different graphics adapters available for
Slot A, Slot 1, and Super 7 systems. For those of you who are technically inclined and
dying to get the scoop on what's under the hood of the Athlon, don't despair. We are
currently working hard on a deeper analysis of the Athlon and the various operational units
providing the results you see here.
What's So Special About the Athlon?
The Ace's crew just loves advanced CPU's, and you will find at least
five articles about the Athlon's architecture alone, here at Ace's.
Don't miss the links at the end of this article to know more.
The most remarkable features of the Athlon:
- Three generalized CISC to RISC decoders
- Nine-issue superscalar RISC core
- Deeply pipelined and therefore build for high clock speeds (more
info here)
- The first out of order, tripple issue pipelined x87 FPU
- 128KB of L1 cache (K6: 64 KB, PII/PIII: 32 KB)
- Programmable backside L2 cache (Size:512 KB-8MB) interface
- Enhanced 3DNow!TM
- 200-MHz interface between CPU and chipset, based on the Alpha
EV6 bus protocol, with multiprocessing support
- 184mm2 at 0.25-micron, six-layer metal process, and counting
22 million transistors
As the K6 name was associated with low cost, budget PCs, AMD renamed
the K7 to champion of competition or Athlon. Ok, now that we know what it is,
let's get to know it!
First Encounters of the 7th Kind
Just about two months ago, I got an exciting phone call: a conference
call with AMD! Lance Smith, Drew Prairie, and Jacub Rise honored your's truly
with the first Athlon conference call in the world!
I had a tough time hiding my excitement. Do not worry, although these
crafty AMD fellows could convince Andy Grove to buy an Athlon, I got back to
my senses. Two weeks later, when I got my very own Athlon system, I began
searching for the weaknesses of this Athletic CPU.
The Athlon has been my roommate for more than a month now, and it has behaved
well. This must be one of the most stable systems I ever worked with. Only CPU ID
manages to crash this system, most likely because it does not recognize the CPU.
In Windows 98, you only have to load the right chipset drivers before you
install your videocard to get the Athlon running smoothly. In Windows NT, however,
you need to apply a patch to enable the write combining features of the Athlon.
Look and Feel
Athlon Startup (POST) Screen
The 600 MHz Athlon derives the core clockspeed by multiplying a 100 MHz memory
bus clock by six (the multiplier). The multiplier is configured on the
processor, so unless you open the Athlon module and have access to a special
device, you can not change the multiplier.
As I caught a glimpse of the internals of the Athlon system, I quickly
found out that this motherboard was not meant for enthusiasts and hobbyists.
The AMD Fester motherboard does not include jumpers to modify voltage settings
FSB (memory) speeds. It looked more like an OEM board that you'd find in a Compaq or
Packard Bell PC.
A View Inside the Case
Making the Athlete Run Faster
As you can no doubt tell from the size of the heatsink pictured above, the Athlon
dissipates a great deal of heat. The following table (compiled from information found on
Kim Noer's Site) displays the maximum heat dissipation
for many high-end microprocessors, including the Athlon.
|
CPU
|
Max. Power Dissipation
|
|
Sun UltraSPARC II 400
|
21.5 W
|
|
PIII 500
|
25.3 W
|
|
Celeron 500
|
30 W
|
|
PIII 550
|
31.2W
|
|
PIII 600
|
36 W
|
|
Athlon 500
|
37 W
|
|
Athlon 550
|
40 W
|
|
Athlon 600
|
43.6 W
|
|
Athlon 650
|
47 W
|
|
Alpha 21264 600
|
109 W
|
Of course, thanks to the large heatsink attached to the CPU, the CPU/heatsink
package never became too hot to touch, even after hours of intense benchmarking.
A Close-up View of the K7 and Heatsink
The already impressive clockspeeds of the Athlon are ramping up even better than
expected, as the fastest Athlon currently available is at 650 MHz. This is 50 MHz
higher than AMD announced and reliable sources tell us that the Athlon-550/600 are good
overclockers, capable of reaching 650 MHz and more.
However, The Athlon will not be easily overclockable in comparison to current products.
It is necessary to have a hardware device that connects to the Golden Fingers of the CPU
to change the multiplier settings.
Golden Fingers
This device is not widely available, but industry sources tell us that
this will change in the future. The first Slot A motherboards can't use memory bus speeds
greater than 100 MHz, but do not despair. Other, well informed sources informed us that
future (Q4 ’99?) Slot A motherboards from ASUS and FIC will provide selectable memory speeds.
So, despite the fact that changing the multiplier will be difficult, if not impossible, you
should be able to overclock.
A New Enhanced 3Dnow! For the Athlon
3Dnow! gave the K6 family the ability to perform up to four floating-point operations
per clockcycle. This has made the K6-2 a viable gaming CPU and, resultantly, 3DNow! has
appeared primarily in games.
Intel's ISSE went a bit further. In addition to prefetching and SIMD FPU capabilities,
Intel also implemented cache controlling instructions. While Prefetch could
only help control the L1-cache, cache controlling streamlines the whole
memory and cache architecture. This has made ISSE applicable to a wider software
palette.
Resultantly, many industry analysts have been tempted to suggest that one day
AMD will have to adopt ISSE because ISSE would become so dominant and so
ubiquitous that any CPU without ISSE would become unmarketable.
But this industry is not static, and while Intel is still the dominant player, AMD
has managed to capture enough marketshare and mindshare to be able to offer
alternative standards successfully.
Nevertheless, AMD understood that the Athlon needed more than 3DNow!
if the Athlon was to be superior over the PIII in every way.
The new enhanced 3DNow! for the Athlon contains 19 additional streaming
and cache controlling MMX instructions, which are very similar (almost identical) to
the Intel ISSE Streaming instructions. In fact, if a programmer has optimized for Intel's streaming, it is a piece of cake
to optimize for the Athlon's streaming as only the instruction names need to be changed.
In other words, although the AMD's Streaming instructions are not binary compatible
with the Intel's ones, a significant amount of work isn't required to optimize for both the
Athlon and the PIII. It gets even better. Five new instructions have been included to accelerate
DSP, softmodem, and MP3 software.
To summarize, 3DNow! Enhanced (or Athlon 3DNow!) includes:
- SIMD FPU and prefetch instructions (21)
- Cache controlling instructions (19)
- DSP instructions (5), do not exist in ISSE
So right now, 3Dnow! enhanced is more complete than Intel's SSE. It
is, of course, backwards compatible with the original 3Dnow! and will be
supported in DirectX 6.2.
What this also means, unfortunately, is that developers will need to optimize for
Athlon 3DNow!, original 3DNow! (the K6-2 and K6-III do not recognize the additional instructions
and will crash when used with them), and ISSE.
As you'll see later on in the review, 3DNow! boosts the performance of the Athlon
significantly. The latest drivers for the Matrox G400, Nvidia TNT2, and
3DFX Voodoo 3 already contain optimizations for the Athlon and its
new 3DNow!. S3 and ATI are currently working on optimized drivers for their products.
Beasts at Battle
AMD Athlon System
- AMD Athlon 600 MHz
- 512 L2 SRAM Cache on Processor module
- AMD production level reference board Fester Rev. C
- AMD 750 chipset
- Northbridge: AMD-751 Irongate
- Southbridge: AMD-756 Viper
- 200 MHz EV6 interface speed,
- 100 MHz SDRAM,
- 2X AGP,
- PCI 2.2
- ECC support
- UDMA-66
- 4 port OHCI USB
- IOAPIC support
- ACPI 1.0 Compliant
- APM 1.2 (Power Management)
- 4 PCI/2 ISA/1 AGP and 3 DIMM (up to 768 MB PC100 SDRAM)
- 2 IDE, 1 floppy, 2 serial, 1 parellel and 2 USB ports with 1 IR connector
- CPU speed auto-detection
- AMD bus mastering EIDE Driver V1.03
- AMD AGP miniPort Driver V4.41
- Thosiba SCM 212 6x DVD ROM drive
- Linksys HPN 1000 home Ethernet card
Intel Slot 1 System
- Pentium III 550, PIII 450, Celeron 400
- 512 L2 SRAM Cache on Processor module (PIII), 128 KB on die (Celeron)
- ABIT BH6 2.0
- ABIT Busmastering EIDE Driver and AGP driver (06/1999)
- ASUS 40x CD ROM
AMD Super 7 System
- K6-2 400, K6-III 400
- AOPEN AX59 Pro with 512 KB cache
- ASUS 40x CD ROM
Common Hardware
- Diamond VIPER ULTRA running at 150/183 and 175/200 (overclocked).
- Creative Labs Sounblaster Live Valuie
- Western Digital WD41800 18GB.
- 128 MB SDRAM CAS 2
Common Software
- Windows '98 First edition with DirectX 6.1A
- Windows NT 4.0 Service pack 4
- Nvidia 1.88 Detonator Drivers
Notes
- The PIII-560 is running at 124 MHz FSB * 4.5 multiplier (overclocked
PIII-450)
- The PIII-616 is running at 112 MHz FSB * 5.5 multiplier (overclocked
PIII-550)
- The Celeron 450 is running at 75 MHz FSB * 6 multiplier (overclocked
Celeron 400)
- The K6-2 448 is running at 112 MHz FSB * 4 multiplier (Overclocked
K6-2 400)
- The K6-2 450 is running at 100 MHz FSB * 4.5 multiplier (Overclocked
K6-2 400)
Before we continue, I'd like to give a big thanks to
PC Magazine-Belgium for
the time I've spend at their labs with the PIII-600 and other hot hardware.
I'd also like to thank Huis Peter, A Belgian shop (Very
good hardware prices) that was so kind to
lend us the Intel P-III 450, the K6-III 400 and the ABIT BH6.
The Athlon System was courtesy of AMD
The AMD Athlon FAQ is available here.
The half-life Blowout and Quake 2 crusher benchmarks are courtesy
of the master of benchmarking, Brett
3 Fingers Jacobs.
All Content is Copyright (C) 1998-2004 Ace's Hardware. All Rights Reserved.
