Forums | MacLife

Theres also a new Mac card coming out. Mac/PC version of the 3870. Doesn't look like X2 as its only $219, but for a Mac board that also works with PCs, thats the first reasonable price ATI has done in all the years they have shipped Mac boards. AMD buying them out was such a good thing , at least for that reason. Old ATI would have probably charged $250-$350 for the same board.

Wow, that was depressing.  NVIDIA was kicking ass and taking names, only a few short years ago.  Now they're loosing the high-end for sure.  I guess we'll have to wait and see how things turn out for the mainstream segment.

MAC GAME SHOOTOUT:
Radeon HD 3870 Mac & PC Edition versus GeForce 8800 GT and others

At Barefeats.


changed to game comparison article

Last edited by Bat (2008-06-13 6:59 pm)

...and NC's STD will be a debutante at Siggraph.

Larrabee to be unveiled at Siggraph
Rattner's jewel
Thursday, 12 June 2008, 10:37 AM

INTEL HELD ITS annual Research Day at its facility in Mountain View today.

And one of the most important things that was said was an off hand remark by Justin Rattner.

He mentioned that Larrabee will be unveiled at Siggraph 2008.

We thought we'd pass it on. µ

This is interesting.

Moving on, we have the other piece of Apple's new performance technology, the Open Computing Language (OpenCL). OpenCL is a C language for GPGPU computing, similar to CUDA and Brook+ for NVIDIA and AMD respectively. Currently CUDA and Brook+ are incompatible, both are C but the languages are different enough that programs are not portable between the two, and neither can compile for the other company's GPUs. Full GPGPU support has been notably absent from the Mac so far while CUDA and Brook+ have been supported on Windows and Linux for some time now. We have heard that Apple has wanted to add full GPGPU support to the Mac for some time now (having been one of the first companies to embrace early GPGPU usage for their video editing applications) but we have also heard that they are unhappy about the incompatibility between the GPGPU languages. They don't want to have to write two of everything, nor do they want their developers doing so. There are 3rd parties that offer GPGPU programming environments that are cross-GPU compatible, but these are expensive and not at all open in any sense of the word.

So we have OpenCL, Apple's proposed universal GPGPU programming language and API. From what we know it looks like Apple is trying to make OpenCL the GPGPU sibling of OpenGL, but like Grand Central Apple has said very little so far. We're not sure who (if anyone) else is backing OpenCL, although we expect AMD and NVIDIA to be on-board otherwise the whole effort will fall flat on its face without support from the GPU manufacturers. Apple has said it's beyond CUDA and Brook+, although we're not entirely sure in what way. From what we've seen of CUDA and Brook+, both are very powerful and very functional languages, so it's unlikely OpenCL is adding any new features that will expand what developers can do; nearly anything can already be done. Rather it seems Apple will be going the simplification route, as CUDA and Brook+ are anything but simple to program for; they may use C but they currently have many nuances that have to be dealt with. An even further generalized framework that's more open to less technical programmers may be what Apple is shooting for. But like Grand Central, we're going to be waiting some time until Apple fully explains where they're going and why.

http://www.anandtech.com/weblog/showpost.aspx?i=461

What... no one planning on dropping $650 on one of these in a few hours? NC? Levi? Books?


...Sci?

Bat wrote:

What... no one planning on dropping $650 on one of these in a few hours? NC? Levi? Books?


...Sci?

I'd rather drop $200 and go HD3870 Crossfire

Hot on the heels of NVIDIA's GTX 200 family launch, AMD will introduce its 55nm RV770-based Radeon 4850 next week. 

The Radeon 4850 features a 625 MHz core clock and GDDR3 clock in excess of 2000MHz. Corporate documentation explains that the 480 stream processors on the RV770 processor offer considerable enhancements over the 320 stream processors found in the RV670 core, though AMD memos reveal little about how this is accomplished.

The RV770 includes all the bells and whistles of the RV670 launched in November 2007: Shader Model 4.0, OpenGL 2.0, and DirectX 10.1.  The only major extension addition appears to be the addition of "Game Physics processing" -- indicating a potential platform for AMD's recent partnership with Havok.

http://www.dailytech.com/AMD+Preps+Rade … e12118.htm

280 GTX Tri-SLI a disappointment

WHEN WE READ THIS ARTICLE the image of toddlers smashing plastic shapes in the wrong slots crossed our mind, only in a Tri-SLI version. Let us explain: TweakTown was brave enough to try the Tri-SLI config, but… First they had to jury-rig a power supply just because their 1000W PSU couldn’t keep up; Second, Tri-SLI offered anything but performance gains over two card SLI (which was fine, btw). Tri-SLI is just the wrong fit... Read on.

Lessee, that's only... ~$2,000 for the cards, plus 1200-1500w PSU and... ouch. Anyone who buys three of these needs a head exam after the walletectomy. Someone at the AT forums early this AM was already asking about buyers for his.

http://www.theinquirer.net/gb/inquirer/ … -17jun2008

Many people, especially in huge articles like the GT200 launch article, skip over the very text heavy pages I tend to write. Especially as I have a flair for low-level technical detail that not everyone enjoys.

In this recent foray into GPU architecture guess work, we spent some time speculating about G80 and GT200 SP pipeline depth. Our guess was 8 stages based on the depth of other architectures at that speed and the potential of wasted power with very deep pipelines. It turns out that we may have guessed way too low on this one (Anand: ahem, actually someone came up with 15).

One of our readers, Denis Riedijk, pointed us to NVIDIA's own forums and CUDA programming guide. These sources reveal that properly hiding instruction latency requires 6 active warps per SM. The math on this comes out to an effective latency of 24 cycles before a warp can be scheduled to process the next instruction in its instruction stream. [..]

Just one page, with links to moar if you likes you the tech stuff.

NVIDIA GeForce GTX 280 Pipeline Update

NVIDIA releases beta PhysX driver

Eager to try out GPU-based PhysX acceleration? If so, you'll be glad to hear that NVIDIA has posted beta PhysX drivers for the GeForce GTX and 9800 GTX GPUs. This beta ForceWare 177.39 release is intended to expose the public to the technology, eventually PhysX supported will be rolled out for NVIDIA's entire GeForce 8/9 family of GPUs.

links 'n' moar

You also can’t miss ATI’s nod to GPU-based physics in the specs listed above. ATI has partnered with archrival Intel (who owns Havok) to make this possible. While both sides haven’t announced anything specific yet when it comes to implementation, we’ve been told that they’re actively looking into areas where it makes sense for the GPU to handle physics rather than the CPU. In those cases RV770 would presumably be used to handle those specific effects rather than the CPU.

Basically they’re not looking to replace the CPU for handling in-game physics anytime soon, but complement it. We’ll have to wait for more details on which games may potentially take advantage of the technology though.

FS Radeon 4850/70 Review

Mr. T wrote:

Bat wrote:

Mr. T might start kicking hisself a li'l over this...

...

Nvidia 9800GTX+ tips up

Indeed.  Review up.

Since I posted that, the cards've come out, excepting that one. ATi seems to've pretty much kicked NV where it hurts. NV wins little this round but the now-dubious title of fastest single-chip solution. Ati wins on price/perf, and when Crossfire is taken into account, absolute perf. In fact, in CF-enabled games, 2 4870s should blow a GTX 280 out of the water, for less money. One already is nearly competitive at $300 vs. $650. 4870s are already in the channel, a few have hit shelves, and the 4870X2 is expected shortly.

I've been keeping most of the latest in WFA lately, as a member is designing a BIY system. Anyway, AT and Ars T. have reviews up (23 pages at AT). GDDR5 is a win that enables better perf at lower PCB cost, and NV is just now hitting 55nm.

Priced at $299 the Radeon HD 4870 is clocked 20% higher and has 81% more memory bandwidth than the Radeon HD 4850. The GPU clock speed improvement is simply due to better cooling as the 4870 ships with a two-slot cooler. The memory bandwidth improvement is due to the Radeon HD 4870 using GDDR5 memory instead of GDDR3 used on the 4850 (and GDDR4 for 3870); the result is a data rate equal to 4x the memory clock speed or 3.6Gbps. The Radeon HD 4870 and 4850 both use a 256-bit memory bus like the 3870 before it (as well as NVIDIA's competing GeForce 9800 GTX), but total memory bandwidth on the 4870 ends up being 115.2GB/s thanks to the use of GDDR5. Note that this is more memory bandwidth than the GeForce GTX 260 which has a much wider 448-bit memory bus, but uses GDDR3 devices.

23 pages later...

That being said, AMD's strategy has validity as we've seen here today. A pair of Radeon HD 4850s can come close to the performance of a GeForce GTX 280, and a pair of Radeon HD 4870s are faster across the board - not to mention that they should be $50 less than the GTX 280 and will work on motherboards with Intel chipsets. Quite possibly more important than the fact that AMD's multi-GPU strategy has potential is the fact that it may not even be necessary for the majority of gamers - a single Radeon HD 4850 or Radeon HD 4870 is easily enough to run anything out today. We'll still need the large monolithic GPUs (or multi-GPU solutions) to help drive the industry forward, but AMD raised the bar for single-card, single-GPU performance through good design, execution and timing with its RV770. Just as NVIDIA picked the perfect time to release its 8800 GT last year, AMD picked the perfect time to release the 4800 series this year.

Like it's RV670 based predecessors, the Radeon 4850 and 4870 both implement DX10.1 support and enable GPU computing through their CAL SDK and various high level language constructs that can compile down SPMD code to run on AMD hardware. While these features are great and we encourage developers to embrace them, we aren't going to recommend cards based on features that aren't yet widely used. Did we mention there's a tessellator in there?

I hope MGS isn't holding back 360 ports because it's unfair to NV, they can't tesselate yet. Gimmee my H3! :shakefist:

The new 9800 is nice, but its biggest edge is in Quake Wars and for those who just bleed green. Other than the GTX+ it looks like close to a Red Team sweep. (And tho physics is being enabled for GT200, there's no test to try a combined physics/ rendering workload, and NV won't be submitting that driver for Futuremark approval, ever. It's for demo purposes and bullet points). 9800 GTX+ should hit the streets in a couple weeks, 4870s are trickling out now; and in 2-3 weeks there'll be 4870X2, tho you'll need a stout PSU for that one.

Edit: And F@H has run on ATi for quite some time now; NV just got their client. 4870 should be a Folding beast, and X2 a monster.

Last edited by Bat (2008-06-26 12:51 pm)