View Full Version : RV740 Announced: Mobility Radeon HD 4830/HD 4860
AnarchX
03-Mar-2009, 06:52
AMD announces 40nm RV740 based Mobile HD4830 and 4860:
http://techreport.com/discussions.x/16506
The 4860 should be a very good competitor against NVs GTX 260M, while holding the possibility to be offered a bit cheaper.
Die-shot!
http://img15.imageshack.us/img15/1707/3324176030a4bb3d11e4b32.jpg (http://img15.imageshack.us/my.php?image=3324176030a4bb3d11e4b32.jpg)
~136mm˛ is really a bit over the rumored 100mm˛...:wink:
Press release:
http://www.amd.com/us-en/Corporate/VirtualPressRoom/0,,51_104_543~130418,00.html
AMD announces 40nm RV740 based Mobile HD4830 and 4860:
http://techreport.com/discussions.x/16506
The 4860 should be a very good competitor against NVs GTX 260M, while holding the possibility to be offered a bit cheaper.
Die-shot!
http://img15.imageshack.us/img15/1707/3324176030a4bb3d11e4b32.jpg (http://img15.imageshack.us/my.php?image=3324176030a4bb3d11e4b32.jpg)
~136mm˛ is really a bit over the rumored 100mm˛...:wink:
Press release:
http://www.amd.com/us-en/Corporate/VirtualPressRoom/0,,51_104_543~130418,00.html
I just read about this on cnn. I was really surprised, since you rarely ever see a mobility part announced first. Hopefully this means I can buy a laptop with one soonish, AMD is saying Q2.
AnarchX
03-Mar-2009, 08:07
Another interesting bit:
ATI Mobility Radeon™ HD 4830 / HD 4860 Product Features
* 826 million transistors on 40nm fabrication process
http://ati.amd.com/products/mobilityradeonhd4800/4860_specs.html
6 million transistors per mm^2. Around 70% percent more than at RV770/730.
Pressure
03-Mar-2009, 08:15
Interesting, I guess we won't have to wait too long to see some initial benchmarks (besides Guru3D's).
rpg.314
03-Mar-2009, 08:40
its definitely rare that mobile parts are coming out first. But with notebooks outrunning desktops, it may become the norm in the future. 70% density increase. Well then a 1600 alu rv870 doesn't seem impossible.
I think it's pretty clear why we see the notebook announcement first. The press doesn't like soft launches for desktop cards, but it's pretty comon for notebook parts.
And it's coming today as a major spoiler for nvidias GTX260/280M launch. If we look at Guru3D's test the 4860M will be a bit faster than the tested part and is going up against a (full) g92 at @585, ie slightly faster than the 4830/9800gt.
Could be nice if they could add a 585mhz downclocked 9800gtx to the test :)
Awesome, all we need now is the RBE and MC configurations.
The chip on the left appears to be RV770. If we take the indicated dimensions into account, it's 265mm2.
Recently we discovered from GT200 that packaging the die adds 0.5mm to each dimension. Assuming the same for ATI GPUs that would make this apparent RV770 ~ 249mm2.
So, erm, take your pick.
For RV740 this would mean 125mm2.
Jawed
That's surprising news, didn't know they were going to announce it today. Guess that's what "fresh and new" brings against "old and stale."
Now.. why did they call this 4830Mobile (which it honestly is) and not use the name of the RV740 products , or is 4860 a hint at that?
My team here at AMD also plays a huge role in pushing the visual experience on notebooks to new levels. Today we announced a milestone by launching the world's first 40nm graphics processors, the ATI Mobility RadeonTM HD 4860 and ATI Mobility RadeonTM HD 4830. This is the fifth consecutive time that AMD is first to launch graphics processors based on a new process node, but the first time we've chosen to make such a debut in the mobile market. (Not to worry desktop fans: 40nm desktop parts are coming soon.)
http://blogs.amd.com/play/archive/2009/03/03/compromise-free-mobile-computing.aspx
Awesome, all we need now is the RBE and MC configurations.
The chip on the left appears to be RV770. If we take the indicated dimensions into account, it's 265mm2.
Recently we discovered from GT200 that packaging the die adds 0.5mm to each dimension. Assuming the same for ATI GPUs that would make this apparent RV770 ~ 249mm2.
So, erm, take your pick.
For RV740 this would mean 125mm2.
Jawed
According to the flickr page that's the 4850 "mobile"
http://flickr.com/photos/amd_unprocessed/3323338543/
compares:
ATI Mobility Radeon HD 4850 vs.
ATI Mobility Radeon HD 4860
Cookie Monster
03-Mar-2009, 09:38
Now.. why did they call this 4830Mobile (which it honestly is) and not use the name of the RV740 products , or is 4860 a hint at that?
Well, look at what nVIDIA is doing at the mobile GPUs. They are using G92Bs and branding them as GTX280/260M. So I think its best not to think too much about branding practices from both companies when it comes to mobile GPUs.
According to the flickr page that's the 4850 "mobile"
http://flickr.com/photos/amd_unprocessed/3323338543/
That's RV770 isn't it?
Jawed
If the Mobility 4860 is 650/1000 Mhz, I guess the desktop 4770 must be considerable faster.. 800?
Except this (which is kinda source-less) I dont see any mention of the RBE configuration: http://forums.vr-zone.com/news-around-the-web/399485-amd-ati-pushes-ahead-launches-40-nm-products.html
Any more official places talking about it, or just mentioning pixel fill rates or similar?
Very interesting - better news that ASAP. A couple of points:
- Jawed, I'm not sure it makes sense to remove 0.5 in this case because this isn't a real die shot; it's a marketing photo. They are more likely to have written the real die size there than what you'd measure.
- Remember 40nm wafer prices are substantially above 65nm ones right now. Let's assume they're at $7K vs $5K; that means it costs more like a 190mm˛ chip on 65nm (ala RV670/G94b). Still, it does have a clear cost advantage assuming yields are decent versus G92b/RV770.
- Jawed, I'm not sure it makes sense to remove 0.5 in this case because this isn't a real die shot; it's a marketing photo. They are more likely to have written the real die size there than what you'd measure.
Having dug around I think you're right:
http://www.extrahardware.cz/je-vetsi-g92b-nebo-rv770
These measurements (even if a bit woolly) indicate that the measurements provided by AMD are die size, not packaged die size.
G92b is clearly smaller in this comparison, about 96% the size of RV770.
Jawed
The measurement isn't minutely precise, because the clamps are touching the glue, which surrounds the core and not the core exactly, so the difference can be up-to 0.7mm.
About that AMD official size comparison... I can't quite read what it says on RV770, but if I read it correctly (18 point something × 16 point something) that would come out to ~300 mm2, that ain't right :nope:
About that AMD official size comparison... I can't quite read what it says on RV770, but if I read it correctly (18 point something × 16 point something) that would come out to ~300 mm2, that ain't right :nope:
16.434x16.110. Your browser prolly supports Ctrl-mouse-wheel to zoom the page.
Meanwhile I've just noticed the Mobility HD4860 has 64GB/s bandwidth:
http://ati.amd.com/products/mobilityradeonhd4800/4860_index.html
which is quite a bit healthier than the Guru3D engineering sample :grin: That should make it match HD4850.
With a bit of luck the desktop part will be about the same.
Jawed
About that AMD official size comparison... I can't quite read what it says on RV770, but if I read it correctly (18 point something × 16 point something) that would come out to ~300 mm2, that ain't right :nope:
That's the die size of the RV770/HD4850 (16.434 x 16.110), the RV740 is 11.7 by 11.7
full size at flickr!
http://flickr.com/photos/amd_unprocessed/3324176030/sizes/o/ (2504x1524)
That saves you a lot of CTRL-clicking
DegustatoR
03-Mar-2009, 21:00
Well, look at what nVIDIA is doing at the mobile GPUs. They are using G92Bs and branding them as GTX280/260M.
Actually ATI was doing this trick way earlier than NVIDIA even had a mobile GPU.
Meanwhile I've just noticed the Mobility HD4860 has 64GB/s bandwidth:
http://ati.amd.com/products/mobility...860_index.html
which is quite a bit healthier than the Guru3D engineering sample That should make it match HD4850.
Well it's called 4860. So it's better match 4850. Or even be a little bit faster. No?
What puzzles me are the lenient HD 4830 specifications. Clock it between 450 and 600 MHz and you can call it a 4830, eh?
What puzzles me are the lenient HD 4830 specifications. Clock it between 450 and 600 MHz and you can call it a 4830, eh?
You'd say they'd make it a 4810 then. ;)
Well it's called 4860. So it's better match 4850. Or even be a little bit faster. No?
The mobility HD4850 (against which this mobility HD4860 is being compared and named) is lower performance than desktop HD4850, so we won't know till the desktop RV740 cards are actually announced :sad:
Fingers-crossed.
Jawed
So they are RV740 rather then RV770 then.. since they only have 640 Shader Core.....
If that is the case then it is noting special is it? Since we expect RV740 in April anyway
You'd say they'd make it a 4810 then. ;)
Perhaps. Or just specify one frequency.
I happen to have another lame inquiry.
http://www.extrahardware.cz/files/images/novinky/2009/03brezen/lf/AMD_ATI_Mobility_Radeon_HD_4860_03.jpg
http://www.extrahardware.cz/files/images/novinky/2009/01leden/lf/mobility_radeon_hd_4570.jpg
What kind of a form factor & slot size is that? The Mobility HD 3800 was a standard MXM module. Has AMD started to use their proprietary form factor lately?
MXM-II
http://arstechnica.com/hardware/news/2009/01/hd-4000-series-goes-mobile-brings-performance-with-it.ars
Not proprietary though
As partly offtopic, was there any other real reasons behind adopting MXM everwhere instead of ATIs AXIOM, except for nVidias marketing power / dragging more companies behind their solution?
What puzzles me are the lenient HD 4830 specifications. Clock it between 450 and 600 MHz and you can call it a 4830, eh?
I suspect core clock won't really make much of a performance difference since it will be quite limited by memory bandwidth anyway (if I'd were to buy a NB with such a mobility 4830 I'd be way more concerned about memory clock...).
I suspect core clock won't really make much of a performance difference since it will be quite limited by memory bandwidth anyway (if I'd were to buy a NB with such a mobility 4830 I'd be way more concerned about memory clock...).
I think you're right
Maybe it's 430 for the GPU's using the shared memory and 650 for the cards on an MXM module.
I think you're right
Maybe it's 430 for the GPU's using the shared memory and 650 for the cards on an MXM module.
mobility 4830 never use shared memory. But still the 128bit ddr3 (likely) will limit it (similar core clock to 4860 but half the memory bandwidth).
mobility 4830 never use shared memory. But still the 128bit ddr3 (likely) will limit it (similar core clock to 4860 but half the memory bandwidth).
Yes they do!
http://ati.amd.com/products/mobilityradeonhd4800/4860_index.html
check the 4860 says gddr5, the 4830 says gddr3/ddr3 (900 and 800mhz.) I'm pretty sure the inq said:
Memory-wise, the cheaper part, the 4830, will support a shared-memory architecture or dedicated VRAM, depending on the notebook model. The 4860 will have GDDR5 from Qimonda (yes, still churning out the memory).
The only problem is that DDR3 is not mentioned in the official press release since they only talk about GDDR3/5. that's why it doesn't get mentioned on most sites.
mobility 4830 never use shared memory. But still the 128bit ddr3 (likely) will limit it (similar core clock to 4860 but half the memory bandwidth).
Of course it IS the inq, but they claim 4830 does:
http://www.theinquirer.net/inquirer/news/229/1051229/ati-40nm-gpus-arrive
Yes they do!
Not really bashing the inq, but seriously that doesn't make any sense whatsover. Being non-igp chip it would need to access all memory through pcie. I don't think that's even theoretically possible (need some latency guarantees for display scanout), and even if it is that still makes no sense - even those poor 40 alu igps are quite memory bandwidth limited!
What's the big deal? About every GPU on the market today supports some kind of "TurboCache" / "HyperMemory" thingy, the mobile HD 4830 not being an exception. Of course if you put enough dedicated memory on the card (that would be the GDDR3/DDR3), you don't need the shared one (although the option is always there should you run out of graphic memory).
No the Mob 4830 does not use hypermemory.
It uses either specialized GDDR3 (higher clockspeed, excess inventory), or commercial RAM-like DDR3 (cheaper) modules, just like the desktop 4670 does.
So is it a CCC bug that the Dark Knight 4870's all have hypermemory enabled?
So is it a CCC bug that the Dark Knight 4870's all have hypermemory enabled?
More likely BIOS, though.
What if it was intentional?
AFAIK anything over _6__ in ATI's DX10+ series does not use HM.
More likely BIOS, though.
What if it was intentional?
AFAIK anything over _6__ in ATI's DX10+ series does not use HM.
My DXDiag wants to disagree with you, showing Approx total mem. for my HD3850 512MB at 2298MB, due the fact that it CAN use my main RAM too if needed.
My DXDiag wants to disagree with you, showing Approx total mem. for my HD3850 512MB at 2298MB, due the fact that it CAN use my main RAM too if needed.
WDDM management. Happens on my 2900XT, and obviously nVidia cards too. :wink:
This is different from the "hypermemory" indication found in ATI's CCC, which shows the local (or non-OS-managed-shared?) memory of the GPU.
vBulletin® v3.8.6, Copyright ©2000-2013, Jelsoft Enterprises Ltd.