Engstrom, Reviews Editor
Review Update: Our initial review of XLR8's MACh Carrier
G4/350 card was a frustrating experience to say the least.
We spent several weeks trying to track down the cause of repeated
crashes and boot problems. With the help of XLR8 we discovered
that the problems stemmed from running the card with "write-through"
mode disabled. Check out the side
bar below for an explanation of write-though. Unfortunately,
in version 1.4.2 of XLR8's control panel write-through is
disabled by default and the option to enable it is hidden.
Enabling write-through fixed all of the stability problems
but caused a performance hit. XLR8 has since released version
1.4.3 of their software so we popped their card back in our
9500 and put it through its paces. The updated
scores below show that the revised software has brought
XLR8 back up to speed and right on par with the competition.
We have left our original review in place but have posted
amendments to the text in red.
General Impressions: We have always enjoyed receiving XLR8
cards for review. We have always found their G3 cards to be
stable and easy to install. I also appreciate the ability
to fiddle with a card's settings to push its performance to
the limit. When XLR8 sent us their G4/350 Carrier Card for
review we expected a similarly pleasant experience. We were
unfortunately disappointed, at least at initially.
Software Installation: XLR8's installer places the usual
extension and control panel in their respective folders. XLR8
also provides their own AltiVec (aka Velocity) plugin for
use with Photoshop 3.x and 4.x. boosting the speed of certain
operations by up to 300%. See the "AltiVec Enabled!"
sidebar below for more information. The control panel provides
information on current settings and allows you to adjust the
backside cache speed and toggle the following on or off:
- power conservation,
- speculative addressing
- motherboard cache.
Another setting which is unfortunately hidden by default
is the "write-through" state of the backside cache.
We will go into this in detail in our stability
section. [Write-through is now visible and on by default.
Attempts to disable write-through result in a warning about
stability issues along with instructions on how to re-enable
it should you run into problems]
Speculative addressing, also called speculative processing
has been the subject of heated online debate after Newer Technologies
released a "white
paper" warning of data corruption, boot problems
etc. if speculative addressing is not turned off properly.
This was followed by a rebuttal
from XLR8 andPowerLogix. Newer takes a hardware approach to
dealing with the speculative addressing issue, turning off
speculation in input/output space but leaving it on in RAM/ROM
space. The other manufacturers take a software/firmware approach.
The issue with the software fix is that the patch can be erased
in the unlikely event you "zap" your NVRAM. You
should note that NVRAM and PRAM zaps are two
different beasts and zapping your PRAM (a useful troubleshooting
procedure) will not remove the SA patch. XLR8 now provides
an emergency startup floppy which will boot your machine and
restore the NVRAM patch if needed.
Hardware Installation: XLR8 wins the prize when it comes
to helping you through the installation process. Their documentation
is well written and comprehensive covering all of the appropriate
machines. In the past, their chart outlining all possible
speed settings may have been a little overwhelming for some.
Their latest documentation now thankfully includes a trimmed
down chart with just the most common settings. The full chart
is still included as well for the adventurous. XLR8 continues
with its fine tradition of including a high quality, reusable
grounding strap to help avoid static mishaps.
Performance/Stability: The XLR8 card was the first G4 to
arrive on our doorstep so we didn't have an idea what to expect
in terms of performance or stability. The first reboot after
install was successful so we proceeded to run through our
benchmark run. Unfortunately we ran into problems when we
attempted to run Unreal frame rate, MacBench 5.0 and the SoundJam
encode tests. On each application the 9500 repeatedly crashed
requiring a reboot. We also had sporadic problems getting
the system to boot. Fortunately in this case we tracked down
outdated FWB disk drivers as the culprit. We reformatted with
Apple's latest drivers and the boot problems disappeared entirely.
The system continued to crash when running the above applications
though and we went through a variety of troubleshooting efforts
to resolve the problem including the following:
Reformatting the drive and installing a fresh copy of OS
Installing four AltiVec specific extensions off the OS 9
CD using the Tome
Lowering the system bus setting
Running off a drive with OS 8.6 installed
Installing the software that PowerLogix provides with their
Only the last two setups yielded favorable results. The card
seemed stable both under OS 8.6 and when running the PowerLogix
software. Conducting a review under an older system software
or using a competitor's software is hardly ideal though. [running
under version 1.4.3 has been very stable so far. We have not
experienced any of the problems mentioned above]
| Write-Through Mode Explained
"write-through is a cache operation mode. Every
time data is stored to the L2 cache, it is also *immediately*
written back to main memory. This means that cache lines
are never "dirty", i.e. main memory always matches what
is in the cache.
Obviously having to write every modification back
to main memory every time will impact performance, usually
by about 10% and this is the reason we do not turn it
on by default."
-Chris Cooksey, Director of Software Engineering, XLR8
After numerous messages back and forth with XLR8's support
and engineering team we stumbled on the solution to the stability
issues. Hidden in the control panel is an option to set the
backside cache to "write-through" mode. This checkbox
option can be accessed by holding down the option key when
selecting the "Advanced" tab in the XLR8 control
panel. [See note above. Write-through is now visible and on
by default] I have been running the card with this setting
for several days now and have not experienced one crash. I
have rebooted several times and run a variety of applications
all of which seem stable. The sidebar to the right explains
what write-through is in detail. It would be interesting to
run comparative benchmarks with write-through enabled and
disabled. Unfortunately, disabling write-through is not an
option, at least not on our 9500 under OS 9. Comparing the
XLR8 card with write-through enabled to the PowerLogix G4/350
reviewed (also with write-through enabled), we noted the
XLR8 card ran between 2% and 20% slower than the PowerLogix
card. We also ran the XLR8 card using the software
package from PowerLogix. While we didn't conduct any real
world tests with this setup, MacBench showed a 19% difference
in processor score and 7% in FPU. Check out the MacBench scores
below for performance
information at a variety of settings and configurations. [Version
1.4.3 brings XLR8's card right on par with PowerLogix and
others. Check out the scores
below for a comparison of 1.4.2 vs 1.4.3. We will also
post updated results to our G3/G4
upgrade card overview in the next few days. It is also
worth noting that I was unable to overclock this card to 400MHz
without repeated crashes.] Interestingly enough, we were able
to use the XLR8 software with write-through disabled just
fine under OS 8.6.
| AltiVec Enabled!
While the rumors have been flying about Adobe posting
the AltiVec enabling software on their site, the ETA
for this posting has come and gone. We have heard (indirectly)
from Adobe that the software has been submitted for
posting and should appear in the near future. At this
point it is unclear whether Adobe's software will be
limited to Photoshop version 5.5 or if it will cover
all versions of Photoshop 5. XLR8 includes their own
AltiVec plugin that will work with versions 3.x and
4.x of Photoshop. While the plugin doesn't boost performance
across the board, it provides amazing results for the
following key operations:
Blend Color, Blend Gradient Tool
Blur, Blur More, Blur Motion, Blur Radial, Blur Smart,
Gaussian Blur, High Pass, Motion Blur
Despeckle, Feather, Find Edges
Rotate, Rotate Canvas 180, Rotate Canvas 90, Rotate
Canvas Arbitrary, Rotate Canvas Flip Horizontal, Rotate
Canvas Flip Vertical
Sharpen, Sharpen Edges, Sharpen More, Unsharp Mask
Image Size, Resize, Canvas Size
It is not clear whether all machines will need to enable
write-through for stable performance. However, I believe it
is a big enough issue that XLR8 should set write-through as
the default and let users try running with it off. Hopefully
this will be the case with their next software release. At
very least, the write-through check box should be visible
and enabling write-through should be added to the troubleshooting
section of their documentation as a potential fix.
As we mentioned before, XLR8 provides an AltiVec plugin for
Photoshop, versions 3.x and 4.x. We ran our Photoshop tests
with and without the plugin and the performance gains were
very impressive! Some filtering operations completed roughly
4x as fast when using XLR8's plugin! As with other G4 upgrade
cards, you will need to run applications that have been optimized
for AltiVec to see this kind of striking performance gain.
If you are running applications that have not been optimized
for AltiVec the G4's are roughly equal in performance (MHz
for MHz) to the G3's. After we have completed all of our G4
reviews we will put up a page with scores comparing all of
Conclusions: All of the stability issues we experienced early
on seemed to arise from running the card Under OS 9 with write-through
mode diabled. It is unclear at this point whether enabling
write-through caused the performance hit we noticed. Running
the card using PowerLogix software (which enables write-through
by default) provided stable performance without an apparent
performance impact. Based on this information I think it is
safe to assume that the XLR8 software is somehow hampering
the card's performance. Hopefully future revisions of XLR8's
software will resolve the stability and performance issues
we noted in this review. If we are able to conduct tests with
new software we will update this page with our findings.
Revised Conclusions: All of the stability problems we noted
in our original review were traced to the disabled write-through
mode of operation. The performance hit we noticed with write-through
on was related to some early (and now unnecessary) software
code which has since been removed. It looks like we are finally
able to give XLR8's card a clean bill of health!
Test Machine Configuration
Our test machine was a 9500 with 96MB RAM and OS 9 installed.
We tested with an extension set comprised of all OS extensions
plus those installed by XLR8 and, for some MacBench tests,
PowerLogix. For the MacBench tests virtual memory was turned
off and disk cache was set to 512k. These settings are consistent
with those used on the MacBench base reference machine, a
beige G3/300. For the real world tests we turned virtual memory
on and set it to 97MB.
|| Includes Photoshop 3.x and 4.x plug-in
which provides excellent performance gains, carrier design
reduces future upgrade costs by allowing the use of ZIF
processor cards, well written documentation.
|| Unable to overclock to 400MHz
MacBench 5.0 Scores
MacBench 5.0 is a subsystem-level benchmark that measures
the performance of a Mac's processor, disk, and graphics subsystems
to name a few. MacBench normalizes all scores relative to
the base machine, a Power Macintosh G3/300. The base machine
receives a score of 1000. For all MacBench tests, higher numbers
mean better performance. For more detailed information on
MacBench click here. Remember, MacBench 5.0
came out well before the G4 processor and was consequently
not written to take advantage of or test the AltiVec (AKA
Velocity) instruction set. Since we ran MacBench under a variety
of different card settings and OS versions we have spelled
out these configurations below.
| Test Configurations
| Card Settings & Software Version
|Backside Cache Setting
|XLR8 G4/350/175/1M v1.4.2
||175MHz (2:1 ratio)
|| This was the setup we used for all of our
"real world" tests below. Ran very stable with
|XLR8 G4/350/233/1M v1.4.2
||233MHz (3:2 ratio)
|| Ran fine, only used for MacBench testing.
|XLR8 G4/350//175 PL Software
||175MHz (2:1 ratio)
||Using software from PowerLogix. System appeared
stable. No significant performance hit.
|XLR8 G4/350/233/1M v1.4.2 OS 8.6
|| 233MHz (3:2 ratio)
||Ran fine even with write-through disabled.
|XLR8 G4/350/175/1M v1.4.3
||175MHz (2:1 ratio)
||Using new XLR8 software, version 1.4.3.
Card was stable and performance on par with other G4 cards
in its class.
"Real World" Tests
(Shorter bars are better)
Time to Scroll a 574 page AppleWorks document
from top to bottom.
Using the same document as above we did a search/replace
command to replace the word "the" with the word
"macbench," over 12,900 occurrences total!
Photoshop 4 "Real World" Test Results
All scores are relative to the stock 9500 which was assigned
a score of 100. Lower numbers and shorter bars are better.
Obviously, the first three tests are filter operations that
can make use of the plugin and the rest are not. We ran these
tests both with and without XLR8's Velocity plugin for comparison.
Render Boy 2.2.0
Time to render "Pool Table" Example
SoundJam MP3 Encode
Time to encode a CD track 4 minutes 26 seconds in length.
for this test we underclocked (blasphemy!) an XLR8 G3/400
to 350MHz. Scores are absolute (time in seconds) and, of course,
shorter times are better...
Variable/Fixed Clock Rate
With ZIF Daughtercard
Toll Free Phone