Sapphire Radeon HD 4850 X2
As a general rule, our belief is that pairing two slow-performing cards
using SLI or CrossFire is a bad idea?you?re usually better off running
a single faster card. However, the Radeon 4850 X2 delivers astounding
performance compared to the single-GPU boards in its price range,
spanking the Radeon 4870 and the GeForce GTX 280, with none of the
pitfalls that have plagued dual-GPU boards in the past. At the heart of
the board is a pair of ATI?s RV770 GPUs running at 625MHz, just like
the single-GPU in the 4850 boards. Each GPU features a full complement
of 800 stream processors, which are connected to identical 1GB GDDR3
frame buffers running at 993MHz on a 256-bit bus. Although X2 boards
are labeled as featuring 2GB of memory, because the contents of each
GPU?s frame buffer must be mirrored, applications can utilize only 1GB
of video memory. Like its 4870-powered predecessor, the 4850 X2 sports
ATI?s advanced video decode acceleration, allowing you to view fully
accelerated picture-in-picture Blu-ray discs. It?s fully compatible
with multiple-monitor displays, and we love that this board features
four DVI ports for multi-mon madness. In our performance testing, the
4850 X2 unseated the fastest single-GPU videocards, the GeForce GTX 280
and Radeon HD 4870, in almost every benchmark. The exception to the
rule was Crysis at high visual-quality settings and high resolution.
When running at 1920x1200 with 4x antialiasing and the visual-quality
settings cranked to Very High, we hit the 4850 X2?s memory bandwidth
wall. Despite running at a higher clock speed than the 4870-family
boards, the GDDR3 on the 4850 transfers half as much data per clock
cycle. With a street price that?s less than $300, the 4850 X2 is a
great deal for owners of lower-resolution 22-inch monitors. However, if
you use a 24-inch or 30-inch panel, it?s probably worth ponying up for
a card with a peppier memory pipeline.
EVGA GeForce GTX 260 Core 216 Superclocked
When Nvidia unveiled its G200 GPU, we were immediately drawn to the
shiny, speedy GeForce GTX 280. Why wouldn?t we be? With high core and
memory clocks and 240 stream processors to churn through the toughest
shaders, it was sexy and fast. We were less excited about the 260,
which sported 192 stream processors and slower clocks speeds but cost
about $100 less than the 280 (at the time). Since then, ATI has
released its R700-based Radeon 4870, which outperforms the original 260
but costs the same amount. And that?s where the Core 216 edition of the
260 GTX comes in. With the same stock clock speeds but 24 more shader
processors than the original, the new version of the 260 GTX delivers
comparable performance to the 4870 at a similar price. The speeds and
feeds are about the same as the original 260?s, although EVGA clocked
this card?s core at 626MHz (up from 576MHz stock) and includes 896MB of
GDDR3 running on a 448-bit bus at 1053MHz (stock is 999MHz). Aside from
the additional shader processors, the Core 216 version of the GTX 260
is identical to the original. The card features all the video decoding
and playback power of the GTX 280, including hardware-accelerated H.264
decoding to accelerate Blu-ray playback. Performance was about what we
expected; the card delivered scores that were slower than a GeForce GTX
280?s but slightly faster than the 4870?s in shader-intensive games
such as Crysis. We?re seeing significantly better performance with both
Nvidia and ATI cards after recent driver releases, so we?ve
rebenchmarked both the 4870 and GeForce GTX 280 to maintain an even
playing field. We?ve seen this card for less than $300 online already,
which puts it firmly in the midrange category. You can find 512MB
Radeon 4870 HDs, which are slightly faster, for less money online, but
the GTX 260?s extra memory will likely help the card do better than the
512MB Radeons with games released in coming years. Of course, the 1GB
4870 boards are available for about $300, but unfortunately, we don?t
have one in the Lab for comparison.
Sapphire Radeon 4870 X2
For a long time, we?ve considered videocards that sport two GPUs
second-class citizens. They have all the problems of multi-card
solutions?namely application incompatibilities and no multi-monitor
support?but fail to perform as well as dual-card solutions, since
multi-GPU cards usually use slower midrange GPUs. That?s finally
changed with the new RV770-powered Radeon 4870 X2, which mounts two of
ATI?s fastest GPUs on a single card, without sacrificing power-user
features like multi-mon support. The 4870 X2 is essentially two Radeon
4870 HD cards running in CrossFire mode packed onto a single board: The
X2 has the same GPUs, the same 800 shader cores running at the same
750MHz core clock, and the same GDDR5 memory running at 900MHz. But,
there is one difference. The single-GPU 4870 includes 512MB of memory,
while the X2 has a whopping 2GB. However, the memory is duplicated
between the two GPUs, so the effective frame buffer for the card is
just 1GB. The X2 also features a high-speed PCI Express interconnect
between the two GPUs, which should, theoretically, boost the efficiency
of the shared GPUs. However, in our tests, we didn?t see an appreciable
performance difference between a traditional CrossFire solution and the
X2. The 4870 X2 outperformed the previous single-card performance champ
in most of our benchmarks, delivering playable frame rates at 1920x1200
and 2560x1600 in nearly every game we tested. Naturally, the exception
remains Crysis, which, at its highest quality settings, punishes nearly
every system we?ve tested. We?re slightly concerned about the accuracy
of our Crysis benchmarks; the ATI card seemed to render far-off
textures at a higher resolution than the Nvidia card. We?ll test
further and report back next month. As always with high-end cards, if
you?re running a low-resolution display?pretty much anything below
1920x1200?you won?t be able to harness the full power of this card. At
lower resolutions, the 4870 X2 performs exactly the same as the
single-GPU 4870. For anyone running a high-res panel, the X2 truly
kicks ass.This card is a significant upgrade if your GPU doesn?t
support DirectX 10?and is much better than some last-gen cards that do.
If you?ve been waiting to make the jump to DX10, now?s the time to
shell out the bucks?you won?t see a better performer for quite a while.
Nvidia GeForce GTX 295
We?ve made no secret of the fact that we love the pulse-pounding speed
that ATI?s Radeon 4870 X2 boards deliver, but there?s a new speed king
in town?the GeForce GTX 295. On paper, the two GPUs on the 295 fall
somewhere between the GTX 260 and GTX 280, but this board delivers a
crushing performance blow to ATI?s fastest part. The GTX 295?s GPUs
feature 896MB of GDDR3 memory and the full complement of 240 shader
cores previously seen only on GTX 280 boards (current GTX 260 boards
have just 216 shader units). However, the core and memory clocks are a
touch below those of the single-GPU GTX 280 boards?576MHz and 999MHz
respectively. Additionally, the new GPU is Nvidia?s first to step down
from a 65nm to a more efficient 55nm process. The benefit? Mega-speed
in one double-wide card. Even with the process-size shrink, the card
requires a new mid-mounted cooler?that?s right, the heatsink/fan is
sandwiched between two boards, each with its own GPU and memory.
Naturally, the card is outfitted with all the accoutrements we?ve come
to expect, including a pair of dual-link DVI ports and a single HDMI
port. Like the rest of the cards in the GeForce G200 series, the GTX
295 also supports general-purpose GPU computing, using both the open
OpenCL platform as well as Nvidia?s proprietary CUDA platform. While
we?re optimistic about the promise of general-purpose GPU computing, we
don?t see any proprietary API gaining enough traction with consumers
and developers to make a long-term impact. The same is true for
Nvidia?s PhysX accelerated physics API. With just a handful of games
supporting PhysX acceleration, and then only for superficial eye candy,
we?ll continue to base our purchasing recommendations on performance in
popular games rather than proprietary APIs that may or may not gain
mainstream popularity. Finally, the boys in green strike back at the
4870 X2 So where does that leave the GeForce GTX 295? With this board,
Nvidia has shown that it can build kick-ass technology and doesn?t need
to hide behind proprietary APIs to protect and expand its market. The
GTX 295 is demonstrably faster than the Radeon 4870 X2 in every
benchmark we use. And that makes our recommendation easy, without even
considering PhysX or CUDA. Expect a street price of around $500.
Visiontek Radeon HD 4850
Last month, we spent a ton of time talking about the efficiency and
overall pixel-pushing prowess of ATI?s new GPU, so we won?t waste much
ink on the subject here. Suffice it to say, the 4850 delivers enough
power to drive your sweet, new 22-inch monitor at its native
resolution. Visiontek?s Radeon HD 4850 delivers entry-level DirectX 10
performance at a compelling price. The card?s silicon is equivalent to
that of previous-gen high-end cards. It?s equipped with 512MB of GDDR3
memory running at 993MHz. Unlike the Radeon HD 4870 boards (which cost
$100 more), the 4850 doesn?t sport GDDR5 (GDDR5 transfers twice as much
data per clock cycle as GDDR3). The upshot? The HD 4850 has the slowest
memory interface of any card in the current generation, and benchmarks
show that?especially at high AA/anisotropic filtering levels. The HD
4850 does sport the same GPU as the 4870, but it?s clocked down to a
modest 625MHz. Unlike the lesser Nvidia parts, which feature fewer
stream processors, the 4850 includes a full complement of 800 stream
processors paired with 40 texture units, just like the 4870. This means
the HD 4850 is at its best in shader-heavy benchmarks such as Crysis.
At the $200 price point, this card?s main competition is the old
GeForce 8800 GT/9600 GT line of parts, against which it compares
favorably. In benchmarks that are limited by shader performance, the
4850 absolutely slaughters the older GPUs. In memory-bandwidth-limited
benchmarks, the older GPUs close the gap. While the benchmarks we list
are primarily geared toward high-resolution screens, we also run some
lower-resolution tests?Crysis on Very High chalked up a respectable
15.3 fps, on High it averaged 28 fps. Our image-quality tests didn?t
show any anomalies, and high-def video playback was flawless. For
anyone riding an old DirectX 9-era GPU, the HD 4850 is your ticket to
full DirectX 10 capability?and a more than capable upgrade from your
old card. For folks who already own a DirectX 10 card, there?s really
nothing to see here.
BFG GeForce GTX 280 OC 1GB
Sporting almost the same configuration as the reference design we
previewed last month, BFG?s GeForce GTX 280 delivers amazing
performance with the second-generation DirectX 10 chipset from Nvidia.
It soundly spanks ATI?s new 4870, as well as all but the dual-GPU
graphics solutions from the previous generation?and even against those,
the GTX 280 wins all but a few benchmarks. The real question we?re
asking is, Do we need this much power? Luckily for Nvidia, the answer
is yes. The company?s GT200 GPU, which forms the heart of the GeForce
GTX 280 and 260 boards, is a great performer, despite its massive
footprint and huge energy requirements. BFG overclocked the GPU core
ever so slightly?it runs at 615MHz?while the GDDR3 memory ticks along
at a stock 1107MHz. The GTX 280 features 240 stream processors running
at 1350MHz?a touch more than double the GPU?s core speed. It?s not
surprising that the GTX 280 fares so well against older Nvidia cards;
far more interesting is its performance compared with the new Radeon HD
4870, which streets for about half the GTX 280?s $660 list price. The
good news for Nvidia is that the 280 is faster across all our
benchmarks than a single 4870 board. However, the ATI card delivers as
much as 80 percent of the performance of the single GTX board at a much
lower price. Still, despite ATI?s accomplishment, the GeForce GTX 280
is the fastest single videocard you can buy today. The GTX 280 features
all the great video-processing capabilities of the 9000-series Nvidia
boards?it accelerates the decode of H.264, MPEG-4, and MPEG-2 videos
and supports HDCP and dual-link DVI. Before you rush out and purchase a
GTX 280 board, make sure your system supports its power requirements.
This is the first single-GPU board we?ve tested that requires both a
6-pin and 8-pin PCI Express power connector, and you need to make sure
your power supply is up to snuff. We recommend a 650W PSU to run a
single card?potential SLIers will need at least a 1000W supply, but you
should check Nvidia's PSU compatibility list before you buy.
PNY XLR8 GeForce 9800 GTX
If you?re already gaming with a G92-based 8800 GTS, there?s very little
reason to move up to a G92-based 9800 GTX such as PNY?s XLR8. The
architecture in both GPUs is nearly the same, with 128 stream
processors, a 256-bit interface, and 512MB of GDDR3. Slightly faster
clock speeds yield only a modest bump in performance. The most
important difference between these two architectures is the fact that
you can build a rig with three 9800 GTX cards, thanks to the presence
of three SLI edge connectors on the top of the board (the 8800 GTS has
only one, so it?s limited to two-way SLI). The new card also consumes
more power and requires two six-pin connections to your PC?s power
supply (the 8800 GTS requires only one). The 9800 GTX also supports
Nvidia?s HybridPower technology, which will be of interest only to
consumers who own a motherboard that also supports HybridPower
(currently, that means a motherboard outfitted with an nForce 730a
chipset). When running less graphics-intensive applications (surfing
the web, using productivity software, or watching a movie, for
example), HybridPower will shut down the videocard in the PCI Express
slot and rely instead on the integrated graphics built into the
motherboard. PNY bumped the 9800 GTX?s clock speeds just a wee bit
beyond Nvidia?s reference design: The core runs at 725MHz (from a stock
675MHz), the shaders at 1.813GHz (from a stock 1.688GHz), and the
memory at 1.160GHz (from a stock 1.1GHz). These tweaks mark the extent
of PNY?s customization, as the card features a reference-design cooler.
All 9800 GTX cards are equipped with two six-pin power connectors and
two SLI edge connectors; the 8800 GTS has one of each. In our benchmark
tests, PNY?s implementation of the 9800 GTX proved to be roughly 10 to
15 percent faster than a stock 8800 GTS?a performance delta that we
don?t think justifies a price premium that ranged from $50 to $75 at
press time. Adding a second 9800 GTX to run in SLI mode resulted in a
34-percent boost in Crysis performance (at 1920x1200 resolution with 2x
AA and all other values set to high), but even with SLI, the game
remains just barely tolerable at those settings. And that?s
unfortunate, because we imagine most people won?t see how exquisite
this game can look because there just isn?t any hardware available
today that?s capable of delivering it. So what if you?re moving up from
an older GPU architecture? AMD still doesn?t have anything worthwhile
for the hardcore gamer, and frankly, we?d still stick with the cheaper
G92-based 8800 GTS. Aside from making three-way SLI a possibility and
supporting HybridPower, the 9800 GTX has no significant new features
worth the extra dough
Gigabyte GeForce 9800 GX2
Compared to AMD?s gracefully engineered Radeon 3870 X2, Nvidia?s
GeForce 9800 GX2 (represented here by Gigabyte?s implementation) is
something of a kludge. But when we consider the performance that
Nvidia?s design delivers, it?s hard to complain about elegance. Both
AMD?s and Nvidia?s solutions deliver dual-GPU performance from a single
PCI Express slot without requiring chipset support (and in both cases,
adding chipset support allows you to run four GPUs in one system).
AMD?s solution, however, plants two Radeon 3870 GPUs and two 512MB
frame buffers on a single printed circuit board; Nvidia?s design
entails taking two PCBs (each with a G92 processor and 512MB of GDDR3
memory), bolting them together, and sticking a ribbon cable in between.
It?s a quick-and-dirty brute-force solution, but it delivers frame
rates like nobody?s business. Gigabyte hewed closely to Nvidia?s
reference design, clocking the GPU cores at 600MHz, the shaders at
1.5GHz, and the memory at 1GHz. Pay no attention to what Nvidia?s
website says, the chip?s memory interface is 256 bits wide, not 512. It
seems both Nvidia and AMD have realized that wider interfaces weren?t
delivering as much performance as they had anticipated; these days,
256-bit memory interfaces are de rigueur even at the high end. We?ve
given up benchmarking Crysis at 1920x1200, and no card has yet
delivered what we?d call acceptable performance even with the 1280x720,
2x AA settings we?ve been using?until now: The 9800 GX2 pumped out 41.4
frames per second. Maximum PC readers don?t live by games alone, and
you obviously don?t need two G92s to watch a Blu-ray movie, but we?re
happy to report that the 9800 GX2 includes Nvidia?s excellent
PureVideoHD for offloading high-definition video decoding from the host
CPU. The HDMI socket on the mounting bracket saves you from having to
use an adapter, but getting digital audio to that connector involves
another kludge: a cable from your motherboard header to a socket on the
card (AMD?s solution routes digital audio over the bus). But if we
can?t have both, we?ll take beast over beauty any day of the week.
