The new Intel platform isn't a secret for anyone who surfed the Web at least once during last half-year. Actually all the important information was officially introduced yesterday. Since there's nothing else to add, we'll just list the facts about "Socket 775" or "LGA775" again:
Socket 775 becomes a leading platform for Intel processors, but Socket 478 will have been supported for a long time. However, all the novelties will in the first place be announced for Socket 775.
LGA775 will become the world's first x86 desktop platform with DDR2 support.
LGA775 processors will be marked using processor numbers instead of real clock rate. For example, today we'll test two such processors: Pentium 4 550 (3.4GHz) and Pentium 4 560 (3.6GHz).
Along with the new processors, the company announces two chipsets: Intel 915P Express and Intel 925X Express. The former is positioned for low- and middle-end systems and supports both DDR2 and DDR. The Intel 925X Express is designed for high-performance desktops and workstations and supports DDR2 only.
Both chipsets do not support AGP. Instead there's higher-speed PCI Express x16. For other devices the usual PCI backward compatibility is provided along with up to 4 x PCI Express 1x slots.
We have dedicated a whole article to Socket 775, so in this one we'll talk about the new processors. Let's look at their appearance first!
So that's what new Pentium 4 and Pentium 4 eXtreme Edition look like. It's easy to see that they differ by layout and number of passive elements in the internal pin-free area. Eh, my bad. No pins anymore!
Unusual, isn't it? The processor back is totally naked. Socket 775 CPUs do not have pins. Instead they feature flat contact pads, while pins are situated in the socket. However, let's get back to the new socket and it's peculiarities and take a look at diagnostic readouts. This time, along with CPU-Z results we provide CPU Info tab of RightMark Memory Analyzer. The latter is not finished yet, but even in the state it is, it did a good job.
So, CPU-Z thinks Pentium 4 550 (Socket 775, Prescott, 3.4GHz) is a Nocona-based Xeon. What a funny mistake :). The clock rate and other parameters are recognized correctly though. RMMA isn't familiar with Processor Numbers yet, so it says it's a mere 3.4GHz Pentium 4 with Prescott core. Well, at least the core is determined correctly.
RMMA gave up a little this time and couldn't recognize the "extreme character" of the new Pentium 4 eXtreme Edition 3.4GHz for Socket 775. But the core is seen well by both CPU-Z and RMMA: server Gallatin (also used in the latest Xeons).
And again CPU-Z considers Prescott Socket 775 a server CPU! Actually all these screenshots just indicate that you shouldn't believe such software right after its new announcements. While it won't mistake in clock rate, instruction sets and cache, CPU official names and core codenames might be a problem. Anyway, the latter are usually fixed within days after release.
A little about the socket
This is how Socket 775 looks without a processor. Resembles an upside-down Socket 478 CPU, doesn't it? On the one hand, we understand unspoken grumble of motherboard vendors, because the socket should become more expensive now. On the other hand, in the end we just have "responsibilities redistributed". The headache just passes to motherboard makers. Users haven't benefited at all, as they still can bend socket pins, instead of processor ones. But from the angle of money Pentium 4 processors cost more in average than boards for them.
On this photo we tried to demonstrate socket pins as clear as possible. You can see they are somewhat complex being designed to slightly spring-load processor pads. According to Intel, conical peaks (not visible on the photo) are not accidental, because if contact is bad, the heat will partially soften pin peak and strengthen the contact.
The processor on the socket background. Just a nice photo. No technical meaning at all :)
Finally, the operating state. The processor is installed and the socket is locked. Everything together looks like a metallic monolith protected from any hazards. Considering the solidity of a locked Socket 775, it seems even a dropped screwdriver or pliers can hardly damage it. Most likely, the board will get damaged instead.
And this is what bundled reference cooler looks like. Peculiarities are clear:
Copper hub not completely covers processor
Large impeller... is naked open without any protective covers!
Redesigned latches fix cooler directly on motherboard
We hope that the impeller is such only because it's a Qualification Sample, because a few times during our tests the cooler cable got under fan blades.
Chipsets: forget AGP
Here we provide two screenshots of Windows XP Professional Device Manager. If you are familiar with system devices list, you will surely notice the absence of "Processor to AGP Controller". Well, it's time you forget it. It's 2004 and AGP is not favoured anymore. There's just one little problem: Windows doesn't know yet the difference between PCI Express and PCI, so it can't supprot any texturing from RAM, no GART, etc. The OS considers PCI Express graphics cards mere PCI devices.
We guess you are tired of looking at pictures, so look at testbeds configuration instead, followed by test results charts :)
Pity, but we have no gaming section in this article. Socket 775 is a young platform, so it was hard to find a top-end graphics card with the same clock rates and chips in AGP and PCI Express variants. We obtained it late, so we'll have to get back to this in the nearest future.
CPU RightMark 2003
Math solver (physycal model)
As before, AMD64 is unbeatable in this test. If you remember, CPU
RightMark now supports all progressive instruction sets and activates
SSE2 in math solver on all platforms, and even SSE3 in rendering (for
Prescott). Thus AMD fairly wins a "visitor match" over Intel in the
SSE2 field. However, Intel's clock rate increase is also noticeable
in case of 3.6GHz Pentium 4 Socket 775. The performance boost is 5%
proportional to the clock rate growth. So, given no architectural
changes are made to Pentium 4, only a 4.1GHz Prescott will achieve
results of Athlon 64 3800+ and Athlon 64 FX-53 in CPU RightMark. It's
also clear that neither chipset, nor L2 cache affect performance in
this test, as all 3.4GHz Pentium 4 processors show similar results.
Rendering module (drawer)
This is just opposite in the rendering module: Intel is a clear leader,
with Prescott showing especially good results. It's interesting that
this is not the benefit of doubled cache (comparing to Northwood),
as the new Pentium 4 eXtreme Edition (actually a Northwood with 2MB
L3) shows not a satisfactory result. There's no considerable difference
between the same-clock Prescott S478 and S775. Remember that not only
chipsets, but memory type is different here. Prescott 3.6 GHz is faster
than 3.4 GHz due to good scalability, at least in this application.
RightMark Memory Analyzer 3.2
Minimal and peak latency
We wrote some time ago that due to Hardware Prefetch peculatiries,
that seems to be considerably improved in Prescott, a round about
a chain in 64-byte strides shows a really interesting picture - tested
latency turns out to be lower than minimal possible! But let's demonstrate
it again. See for yourself, due to hardware prefetch with 64-byte
strides Prescott latency is even smaller than that of AMD64 systems
with their memory controller! But if you double the stride, it becomes
There is a way to corner Prescott's Hardware Prefetch. But its improvement
is still respectable. Note the serious lag of both older Pentium 4
processors. At that, Prescott in its best form (Socket 478 + usual
DDR400) almost catches up with both AMD processors despite not integrated
memory controller. On the other hand, Socket 775 doesn't close the
gap, even vice versa. Well, it uses DDR2-533 with timings worse than
those of DDR400 and a way too narrow bus coupled with asynchronous
operation (133MHz actual memory clock rate and 200MHz FSB). Pentium
4 eXtreme Edition has rather high minimal latency. We guess it's the
bad performance of L3 cache on a relatively slow 64-bit bus.
Memory read performance
Results of all processors except Northwood-based (which conditionally
include Pentium 4 eXtreme Edition though it has Gallatin core) are
about the same and lie near theoretical peak bandwidth. We have reasons
to "forget" about the bus bandwidth, as DDR2-533 has theoretical peak
of 8400 MB/s, but 800MHz FSB of today Pentium 4 just can't support
such speed. Northwoods performed rather bad. It seems Prescott's hardware
prefetch also touch serial memory access (that is actually logical,
because Prefetch is needed here most).
Memory write performance
Test subjects are clearly divided into two groups that witnesses in
favour of AMD's integrated controller. At least there's no difference
between Intel processors independently on chipset. L2 cache doesn't
play a role here as well, because our memory write mode just ginores.
3ds max 5.1 + Brazil r/s
We don't even want to distribute this parity into elements. Such a
difference is not in any way impressive to choose a platform on its
A standard situation for this software. AMD loses to Pentium 4 Northwood
and is on the level of new Prescotts. 3.6GHz Prescott demonstrates
a slight boost comparing to 3.4 GHz. Scalability is visible and even
the highest-clock-rate Pentium 4 is not limited by bandwidth.
It's almost a straight line with a slight outstanding of the new Pentium
4. Just another example of good scalability of Intel processors.
Windows Media Video 9
Prescott feels assured on all its platforms as well as AMD processors.
Intel Northwood is not favoured, even with Pentium 4 XE cache. If
you remember, we have repeatedly mentioned the effect of Hyper-Threading
in Prescott. And this application utilizes this technology very well
and provides a good boost.
Mainconcept MPEG Encoder 1.4
Prescott is favoured again. But this time Athlon 64 lags behind -
due to good multi-threading implementation again we guess.
The dispersion is not great. Prescott shows the best results among
Intel products, but both Athlon 64 are generally victorious.
LAME and Oggenc encoding
AMD64 mostly competes with Intel Northwood. Prescott lags even at
3.6 GHz, but the scalability is noticeable again.
Again it's about total parity, so let's look a particular details.
First, Prescott performed good again. Second, we for the first time
see almost zero boost of 3.6 GHz processor compared to 3.4 GHz one.
It's early for conclusions, but it seems we managed to find an applicationo
to demonstrate unsatisfactory bandwidth (or why clock rate growth
doesn't boost performance then?) There's also no benefits of DDR2.
But considering recent
tests, it's good there no drawbacks at least.
And as if purposely another archiver turns results upside-down: Northwood
performs good, Prescotts perform bad. It's symptomatic that AMD is
often a winner in such situations. This time was no exception. Socket
775 generally loses even to Socket 478 seemingly due to high DDR2
The most mysterious result of all provided in this article.
Neither low-level test above indicated considerable Socket 775 superiority
over older platform with the same processors. But nevertheless, this
superiority is clearly visible in a real application! Note the difference
between Prescott 3.4 GHz on i875P (Socket 478, DDR400) and theoretically
the same Prescott 3.4 GHz on i915/i925X (Socket 775, DDR2-533). For
now we have to just state it and assume some rather deep changes in
Intel's new memory controller that still stay hidden from our sight.
In general, despite new processors and chipsets, we have nothing extraordinary. Yes, there's a new socket. Yes, there are new chipsets along with new memory. But we still think that the situation would hardly change, if Pentium 4 550 and Pentium 4 560 were named just Pentium 4 3.4E GHz and Pentium 4 3.6 GHz and were designed for the good old Socket 478. Some results would be better, some worse maybe (but hardly considerably bad to stress them). The main thing that Intel's new platform demonstrates is vendor's priorities and future plans: noone is going to decline Prescott despite it loses to Northwood sometimes. Especially, considering it wins over Northwood in other situations, and is particularly good with Hyper-Threading. Prescott's drawbacks will seemingly be fixed in a good old way - by clock rate. And the release of 3.6 GHz Pentium 4 should assure everyone who doubts that clock rates will grow. Moreover, we assume that Prescott 3.8 GHz won't be the one to wait for long. At least, it will be strange, if it's not announced this year. Besides, Intel will surely work with software vendors to move as many applications as possible to multi-threading base. It seems in some complex cases, it's the only chance for Prescott to show what it's good for.
However, the main meaning of this announcement is not related to performance or clock rate growth. The meaning is about the arrival of the new platform bound to live at least as much as Socket 478. It's also clear that new socket favourites will be not present, but future processors: Pentium 4 will have to obtain at least 1066MHz FSB to utilize all DDR2-533 potential. The bet on DDR2 is also clear, because Intel declined DDR in the new top-end i925X. Therefore, it's obvious that i915P is a replacement of i865PE, i925X is the new i875P. It's also good that LGA775 platform, according to information that we have, will be compatible with future multi-core processors, so that another considerable bandwidth change won't require a new socket.
To draw a bottom line, we can say that Socket 775 started as optimally as it was possible: on the one hand the company introduced a new higher-clock processor to prove clock rate growth potential. On the other - there's a reserve for the future, so the performance growth is not solely related to CPU clock (remember that 1066MHz FSB). In other words, there's something to be pleased with already today, with more benefits coming up. It's actually a clever step - to introduce a leading platform (among other prorietary products) and then allude to more bonuses in the future. Well, we just have to wait for this future, so let's hope it won't take long.