CES is over and the hope, promise, and promise of Intel’s Sandy Bridge CPU with embedded processor graphics (EPG) has been revealed. There is rejoicing through the press and the webosphere.
We’ve been testing and using a 4-core version Sandy Bridge Core i5-2500k running at 3.3 GHz. We ran a variety of tests and compared the Sandy Bridge against the previous generation, Clarkdale.
As you might expect, Sandy Bridge, code named “SNB,” is a “Tock,” and considerably faster in all the operations we tested, including CPU tests, and graphics.
Intel’s new Sandy Bridge desktop processor architecture has a lot going on, from a shared 8 MB Smart Cache that feeds eight-way independent processing threads with Turbo Boost 2.0, to the improved Advanced Encryption Standard (AES) Instruction Set that significantly improves encryption-decryption. However, as great as they are, they don’t bring much to the gaming enthusiasts.
Nonetheless, Sandy Bridge is the most efficient processor Intel has ever built, and can be overclocked beyond 5 GHz before melting.
One of the big new features for Sandy Bridge is the Advanced Vector Extensions (AVX). These new extensions widen the vector registers out to 256 bits. That means Sandy Bridge’s floating-point hardware can run 16 single-precision and 8 double-precision floating point operations per cycle—that brings a lot of FLOPs to the party and challenges GPUs for that bragging point.
Sandy Bridge also adds a GPU-like, video transcoding engine that shares the L3 cache and with it Intel introduced a ring bus.
The ring bus is the same as used in Nehalem EX and Westmere EX. Each core, each slice of L3 (LLC) cache, the on-die GPU, media engine and the system agent (AKA North Bridge) all have an address on the ring bus.
The bus is made up of four independent rings: a data ring, request ring, acknowledges ring and snoop ring. Each stop for each ring can accept 32-bytes of data per clock.
It’s the Graphics
As good as AES is, it’s the GPU that really makes the difference in Sandy Bridge. The GPU treated like just another processor in Sandy Bridge and as such has equal access to the L3 cache. The graphics driver controls what goes into the L3 cache and can be used to limit how much of cache the GPU is allowed to use so it doesn’t become a resource hog and slow down other operations – you know how those GPUs can be. Storing graphics data in the cache is always important because it saves cycles to/from main memory.
The largest performance improvement on Sandy Bridge vs. the Westmere architectures is the graphics. The CPU cores show a 10 – 20% improvement in performance over Clarksdale. However, the Sandy Bridge graphics performance is an average of 75%+ what Clarkdale can do. Also, some games Clarksdale couldn’t even run and Sandy Bridge can.
The GPU in Sandy Bridge has some programmable shader hardware units that Intel calls Execution Units (they’d choke or be beaten if they used the term GPU). The EUs can duel issue picking instructions from multiple threads and Intel says the ISA maps one-to-one with most DirectX 10 API instructions resulting in a very CISC-like architecture. Moving to one-to-one API to instruction mapping increases inter-processor communications by effectively increasing the width of the EUs.
Surprisingly transcendental math is handled by hardware in the EU (and not handed off to the CPU) so its performance has been sped up considerably. Intel said at IDF that sine and cosine operations are several orders of magnitude faster now than they were in current HD Graphics.
There will be two versions of Sandy Bridge graphics: one with 6 EUs and one with 12 EUs. The first version of mobile parts will use 12 EUs, while desktop processors will be available with either use 6 or 12.
Not for everyone – maybe
At IDF Intel engineers discussed the markets they can, and cannot, address with Sandy Bridge’s graphics. And said they are not trying to target the most high-end discrete (standalone) segment. Intel acknowledges they don’t have the memory bandwidth, or the power (watts) budget.
Sandy Bridge is a UMA design (some have described it as just an IGP in the CPU, but that’s neither accurate nor fair). However, it does have to share the PC’s main memory, and the best PCs will still only have DDR3 DRAM, so that becomes the asymptote on performance. However, for a few dollars, less than $100, you can add a graphics board and dramatically increase performance. And, if LucidLogix gets any traction in the ODM/OEM world you can have the best of both worlds (see story this issue.)
The tests
Testing the Sandy Bridge wasn’t as easy as we thought it was going to be. Given all the things the processor is good at we could probably have spent the next month or two running various tests.
General tests
The first battery of tests we put the new processor through were general CPU like tests, using Super PI, and 3D Mark Vantage CPU, and GPU tests.
Games tests
We used three popular DirectX 10 games to test the Sandy Bridge: Resident Evil, Starcraft 2, and Stalker COP. To be fair, Sandy Bridge is not, nor was it intended to be, a gamer’s machine, and these game are pretty heavy-duty.
However, in the case of Stalker, at least Sandy Bridge could run (whereas Clarksdale couldn’t.) And we ran the tests at max screen resolution – 1680 x 1050. However, lowering the screen resolution to 1280 x 1024 will raise the FPS to an acceptable level.
And, by adding a low cost, $65, graphics board, all the games we used ran quite well. $65 doesn’t seem like much of cost is someone wants to play serious RPG or FPS games.
What do we think?
Sandy Bridge is cool running (literally). The processor has advanced power management capabilities, and can turn things as needed. We expect to see it become the industry’s mainstream processor.
Pros
Sandy Bridge has a lot of good to great qualities: great floating point capability, very good at transcoding (QuickSync), excellent decryption-encryption, low power operation, pretty good graphics (up from good enough), and good video performance. See TheCyberLink story for the transcoding performance.
Cons
As stupid as it sounds, at this stage of testing and working with it, we don’t have any complaints. The graphics are as good as they can be for the thermal envelop.
Read more at Jon Peddie tech Watch
