some ATI 4800 details

Across architectures ? It's the SAME architecture, which is why 800 SPs doesn't seem reasonable. There is no confirmation yet, but I would like to see those "official slides". You mean the suspicious looking slides, with extreme compression artifacts around the 800 number ? :rolleyes:

Good God, Silus. You are wrong. It's 800SP. Please put down the keyboard and step away before you hurt yourself.

-------------------
pcper said:
The GPU-Z utility has been updated with support for RV770 cards:
You can see how fast our HD 4850 is running: 625 MHz core clock and 993 MHz memory clocks. Notice the 800 unified shaders, GDDR3 memory, 256-bit memory bus width and a die size about half that of NVIDIA GT200 chip.

AMD Radeon HD 4850 512MB Preview - RV770 Discovered
---------------------
Legion Hardware said:
Despite being the third fastest graphics card that has been named in the Radeon HD 4000 series, the Radeon HD 4850 specifications are still extremely impressive. The Radeon HD 4850 features a staggering 800 (160x5) SPUs, while there are 40 TAUs and 16 ROPs. Compare this to the 320 (64x5) SPUs, 16 TAUs and 16 ROPs of the Radeon HD 3870, and you quickly begin to understand just how impressive this new graphics card really is. The Radeon HD 4850 has a memory bandwidth of 64GB/s, while the Radeon HD 3870 was limited to 57.6GB/s of bandwidth.

http://legionhardware.com/document.php?id=755
---------------------

ATI Radeon HD 4870 ATI Radeon HD 4850 ATI Radeon HD 3870
Manufacturing Process 55nm 55nm 55nm
Stream Processors 800 800 320
Texture Units 40 40 16
Render Back-Ends 16 16 16
Core Clock Speed 750Mhz 625Mhz 775MHz
Memory Data Rate 3.6 Gbps GDDR5 2.0 Gbps GDDR3 2.25 Gbps GDDR4
Math Processing Rate (multiply-add) 1.2 TeraFLOPS 1.0 TeraFLOPS .497 TeraFLOPS
DirectX Support 10.1 10.1 10.1

AMD Zone: ATI Radeon HD 4850
----------------------

You can just say "Ok then.", or "Fine. I was wrong", or "Wow! ATi really did some great things with this chip!".
 
Good God, Silus. You are wrong. It's 800SP. Please put down the keyboard and step away before you hurt yourself.
Silus said:
yeah...but....they're inefficient!!!one

ffs people. specs don't matter in today's world. Anybody with 1/2 a brain realized that during the Athlon vs Pentium days.
 
Hey look what's around:

ExtremeTech RV770 Architecture Analysis

Wow, they did a LOT of changes to the architecture to fit all that within 256mm^2. 956M transistors in 256mm^2 makes me wonder why Nvidia's 1.4B seems to inefficient. Anyways, some juicy parts (and pics)



This may be our first hint at what is new with the dual-chip "R700" product coming in a few months—these hubs may communicate between two RV770 chips in a fashion that is more efficient than in past multi-GPU boards.



Last but not least, we have the render back-ends, commonly called ROPs. There are still 16 ROPs in the RV770 chip, but ATI has focused on improving their performance degradation when using anti-aliasing—another real sore spot of the RV670 chip. Depth/stencil ops are doubled per ROP, which effectively doubles the fill rate with 2x or 4x AA to a full 16 pixels per clock (up from 8 in the RV670). 8x AA is 8 pixels per clock, again double the RV670. All 16-bit per component color modes have doubled AA resolve, and depth/stencil only rendering passes are of course doubled as well. The RV770 ROP is now closer in capabilities to that of Nvidia's G92 and GTX 200 GPUs.

Of course, with all these additional texture units and stream processors, ATI had to roughly double the intra-chip communication bandwidth to make sure it doesn't all get bottlenecked. It's really rather impressive that the company was able to accomplish all these architectural changes and fit so many more texture units, stream processors, and improved ROPs in a chip that is still on a 55nm process but manages to be only around 33% bigger, and still manages to be over 20% smaller than the G92 GPU.

Maybe all that talk about ATI really employing wizards is true.

800SP's, 40 TMUs, 16 ROPS (at twice the clocks of before), and a new hub

Now I wonder what the new hub does for R700... interesting
 
Good God, Silus. You are wrong. It's 800SP. Please put down the keyboard and step away before you hurt yourself.

:rolleyes:

I questioned the 800 SPs number, because there was no real proof of it, until the NDA lifted and previews showed up. With the specs overview in some of those sites that posted their HD 4850 preview, did I see something more palpable than flacky slides, with compression artifacts all over the 800 number. Now, it is confirmed that it is indeed 800 SPs and so I was wrong in doubting the rumor.

You can just say "Ok then.", or "Fine. I was wrong", or "Wow! ATi really did some great things with this chip!".

How is it a "great thing" to only be 30-40% faster than a HD 3870 (i.e. at the levels of a 8800 GT 512 or 8800 GTS 512), after an increase of 2.5 times in SP count and 2.25 times in TMUs ?
 
ffs people. specs don't matter in today's world. Anybody with 1/2 a brain realized that during the Athlon vs Pentium days.

That would actually make sense, if we were talking about different architectures, but we are not. It's the exact same architecture as R600/RV670, but with tweaks and additions, yet performance numbers are not adding up.
 
That would actually make sense, if we were talking about different architectures, but we are not. It's the exact same architecture as R600/RV670, but with tweaks and additions, yet performance numbers are not adding up.

Ever considered drivers? They obviously don't add up, but it's confirmed to have 800SP so you can only go up.

The current drivers don't even support Powerplay (2D Clocks) so I assume they're pretty primitive for the cards.
 
That would actually make sense, if we were talking about different architectures, but we are not. It's the exact same architecture as R600/RV670, but with tweaks and additions, yet performance numbers are not adding up.

New cards are always based on the previous hardware. There are no completely new architectural revamps, you know that. Even the GT2xx series uses some basic architecture from the G80. It's like being surprised the sun rises in the morning.

In either case, the numbers may be due to drivers. It's officially been out for, what, 24 hours? the GT2xx cards have received 3 driver releases so far. Even the latest official drivers from ATi (8.6) don't have the 4800 series labels in them. The big NDA on the 25th will go into much more detail. So take the PREVIEWS as what they are, have some patience, and we'll see what happens in a few days.

EDIT: Looking at Extremetech's Preview, the performance numbers certainly are adding up, expecially for a card that is "just a tweaked 3870":

Much like Enemy Territory: Quake Wars, Nvidia leads by a few percent with AA and AF disabled with a newly price-reduced GeForce 9800 GTX. Turn AA and AF on, and ATI pulls ahead, especially at 1920x1200 where the new card leads by a healthy 30%. Performance is around 20–25% better than the 3870 without AA or AF, but turn those features on and the performance increase jumps to 70%. ATI has done wonders to improve their AA performance. etc etc etc

The Little Wonder - AMD Radeon HD 4850
 
I'm looking forward to the 4870 X2. I just put together a new system for it, which is patiently waiting for it. :D

I hope it truly performs. We need to keep the competition going between ATI and Nvidia. Without ATI, no more competition, thus nothing will keep Nvidia from jacking the prices up. That, as well as we'll see a slow down in new technology. The same goes for AMD and Intel.

.
 
Ever considered drivers? They obviously don't add up, but it's confirmed to have 800SP so you can only go up.

The current drivers don't even support Powerplay (2D Clocks) so I assume they're pretty primitive for the cards.

Sure, drivers may not be fine tuned for the cards yet, but there are no magic drivers that boost performance that much.
 
:rolleyes:

I questioned the 800 SPs number, because there was no real proof of it, until the NDA lifted and previews showed up. With the specs overview in some of those sites that posted their HD 4850 preview, did I see something more palpable than flacky slides, with compression artifacts all over the 800 number. Now, it is confirmed that it is indeed 800 SPs and so I was wrong in doubting the rumor.



How is it a "great thing" to only be 30-40% faster than a HD 3870 (i.e. at the levels of a 8800 GT 512 or 8800 GTS 512), after an increase of 2.5 times in SP count and 2.25 times in TMUs ?

You are always confident with nVidia rumours that said "a card is at least x% faster than the other card" but it is hard for you to believe a posive rumour on ATi cards :p

The HD4850 is a "great thing" because it is cheap and it takes less hit with AA and higher resolution than the G92. The Crossfire scaling is also better than SLi most of the time so when the G92 and HD4850 have the same price, it is hard not to choose the HD4850 especially when someone already has a mobo with an Intel chipset.
 
New cards are always based on the previous hardware. There are no completely new architectural revamps, you know that. Even the GT2xx series uses some basic architecture from the G80. It's like being surprised the sun rises in the morning.

Of course there are. GeForce FX to GeForce 6, GeForce 7 to GeForce 8, etc. And I had already mentioned multiple times that neither RV770 or GT200 were new architectures.

Karma said:
In either case, the numbers may be due to drivers. It's officially been out for, what, 24 hours? the GT2xx cards have received 3 driver releases so far. Even the latest official drivers from ATi (8.6) don't have the 4800 series labels in them. The big NDA on the 25th will go into much more detail. So take the PREVIEWS as what they are, have some patience, and we'll see what happens in a few days.

There are no magic drivers that increase performance that much. You should also know that.

Karma said:
EDIT: Looking at Extremetech's Preview, the performance numbers certainly are adding up, expecially for a card that is "just a tweaked 3870":

But of course that the numbers are good for the HD 4850. I never even questioned that, considering its price point. The point is that for a 2.5 times increase in SP count and 2.25 times in TMUs, on the SAME architecture, I would expect much better.
 
Of course there are. GeForce FX to GeForce 6, GeForce 7 to GeForce 8, etc. And I had already mentioned multiple times that neither RV770 or GT200 were new architectures.



There are no magic drivers that increase performance that much. You should also know that.



But of course that the numbers are good for the HD 4850. I never even questioned that, considering its price point. The point is that for a 2.5 times increase in SP count and 2.25 times in TMUs, on the SAME architecture, I would expect much better.

Afterall it is only 30% increase of the die size.
 
You are always confident with nVidia rumours that said "a card is at least x% faster than the other card" but it is hard for you to believe a posive rumour on ATi cards :p

Oh please, not another fanboy attack. :rolleyes:

I always included "based on rumors and/or rumored performance numbers" in any of my comments related to unreleased products. It's not fact, it's rumor, but obviously your fanboy eyes, only see it that way. So be it. I don't expect a fanboy to be rational about it.

As for the 800 SPs rumor, it was conflicting with the early rumors of both transistor count - 800+ million - and SP count - 480, so I had my doubts about it. Now that the previews give a glimpse of specifications, my doubt is cleared and it is indeed 800 SPs.

alg7_munif said:
The HD4850 is a "great thing" because it is cheap and it takes less hit with AA and higher resolution than the G92. The Crossfire scaling is also better than SLi most of the time so when the G92 and HD4850 have the same price, it is hard not to choose the HD4850 especially when someone already has a mobo with an Intel chipset.

Again, I always agreed with the great value these new cards present, though of course you disregard those posts of mine, which is a big surprise :rolleyes:

Now, I only wait for real-world gameplay numbers, which should be released next week. It's definitely a great time to upgrade for any consumer, since the prices are fluctuating so much on the low side.
 
Anyway, we shall see when the first/second tuned driver for the 4850 comes.
 
Anyway, we shall see when the first/second tuned driver for the 4850 comes.

How come I keep hearing that the 4850 will be insane after the drivers issues are sorted out, but the 280 will gain nothing from drivers?
 
How come I keep hearing that the 4850 will be insane after the drivers issues are sorted out, but the 280 will gain nothing from drivers?

I'll give you a hint.

It starts with "fan" and ends with "bois".

Can you guess what it is?
 
How come I keep hearing that the 4850 will be insane after the drivers issues are sorted out, but the 280 will gain nothing from drivers?

Insecure much?


I speculated in another forum that the GT200, more notably 280 will gain 20-30% if there are no major bottlenecks, the closest I know of is triangle setup rate.


The RV770 does NOT have a ringbus, I assume a lot of stuff is still done very funnily compared to the other chips. Bite it. :)
 
I questioned the 800 SPs number, because there was no real proof of it, until the NDA lifted and previews showed up. With the specs overview in some of those sites that posted their HD 4850 preview, did I see something more palpable than flacky slides, with compression artifacts all over the 800 number. Now, it is confirmed that it is indeed 800 SPs and so I was wrong in doubting the rumor.



How is it a "great thing" to only be 30-40% faster than a HD 3870 (i.e. at the levels of a 8800 GT 512 or 8800 GTS 512), after an increase of 2.5 times in SP count and 2.25 times in TMUs ?


Silus,
actually there were a lot of good reasons to believe that 800 SP rumor before the NDA lifted. We had leaked tests from those with the card running part of 3Dmark that backed up this claim, we had leaked slides, we had reliable info from CJ over at B3D that backed it up (and his record in the past was amazing) and then we had the developer of the folding program (Dr Vijiay) that slipped up in one of his presentations when he said that there also was 800 SP on the new ATI hardware. Now had this been some random posting on a forum or started by INQ we could have had doubts. But there was just too much credible information from different sources saying the same thing.

I also don't understand that 30% comment. We have benchmarks from Anadtech, TechReport and ExtremeTech showing it to be faster than the 9800 and even faster than the 260 in some cases. How is that not amazing for a card that is selling under 199 bucks right now?
 
Insecure much?

lol.

That comment wasn't directed at anyone person in general. (Mostly because I didn't want to waste my time looking up quotes) But I'm just sick of reading in both threads that XXX's card sucks, but YYY's card will be awesome when the new drivers come out.
 
8.6 catalyst beta drivers FTW. No seriously it is rather amazing how the 4850 is changing around the mid level gpu market with it's performance numbers. Anyone here remember the days when gt were for $200 at least gts was at least $50 higher and 9800 gtx was around $300? Anybody notice the crazy price drops recently? That can't be all atributed to the gtx280 launch, you really have to thank the 4850 for that.
 
8.6 catalyst beta drivers FTW. No seriously it is rather amazing how the 4850 is changing around the mid level gpu market with it's performance numbers. Anyone here remember the days when gt were for $200 at least gts was at least $50 higher and 9800 gtx was around $300? Anybody notice the crazy price drops recently? That can't be all atributed to the gtx280 launch, you really have to thank the 4850 for that.

The 8.6 Betas nor Finals change things substantially.
The ringbus assumptions/corrections are still there I guess. This increase might take longer than I suspected, but getting a 4850 now is so viable already that extra performance, be it 10% or 40%, is just icing on the cake.
 
956 million transistors according to vr-zone, with the 4870 clocked at 750/1800.

I would love to see a O/C'ed version at 800/2000!
 
and 9800 gtx was around $300? Anybody notice the crazy price drops recently? That can't be all atributed to the gtx280 launch, you really have to thank the 4850 for that.

not really, there was a deal in the hot deals forums about a 225 9800GTX 2 days before the launch of the 4850.
 
not really, there was a deal in the hot deals forums about a 225 9800GTX 2 days before the launch of the 4850.
....you don't get it do you? Might as well drop the price and sell more before the competition moves in.
 
There's definitely more performance to be had from RV770 now it has been confirmed as having 40 TMU's - the guy at xtremesystems (I'm sure you know the thread!) posted his fillrate results and they indicated 32 texture units based on clockspeed calculations (came out at 31.7 or so). RV670 came out at 16 (which is bang-on), so in theory there's 25% more texturing power sitting there idly (while 3D'06 sucks, if a pure fillrate test can't make use of all available units... )
EDIT: Perhaps 20 ROP's were in order? I know they've doubled the efficiency (at least with AA enabled), but as the shading/texturing subsystem has had a 150% increase perhaps rasterization has become the bottleneck?
 
Silus,
actually there were a lot of good reasons to believe that 800 SP rumor before the NDA lifted. We had leaked tests from those with the card running part of 3Dmark that backed up this claim, we had leaked slides, we had reliable info from CJ over at B3D that backed it up (and his record in the past was amazing) and then we had the developer of the folding program (Dr Vijiay) that slipped up in one of his presentations when he said that there also was 800 SP on the new ATI hardware. Now had this been some random posting on a forum or started by INQ we could have had doubts. But there was just too much credible information from different sources saying the same thing.

I also don't understand that 30% comment. We have benchmarks from Anadtech, TechReport and ExtremeTech showing it to be faster than the 9800 and even faster than the 260 in some cases. How is that not amazing for a card that is selling under 199 bucks right now?

You have to forgive silus... he has been a wet blanket on anything ATI since the 4800 rumors first started.
 
Yeah. I haven't seen so much denial since I asked my 7 year old niece if she ate some cookies, despite the chocolate she had on her face.

btw, just like I predicted yesterday, AMD has finally officially released drivers/hotfix that includes the 48xx series. It'll be interesting to see the differences from before:

AMD said:
The information in this article applies to the following configuration(s):

* Radeon™ HD 4870 series
* Radeon™ HD 4850 series

This Hotfix improves overall performance
and stability. The Hotfix includes the Display Driver and Catalyst Control Center.

http://support.ati.com/ics/support/default.asp?deptID=894


EDIT: grammar.
 
You have to forgive silus... he has been a wet blanket on anything ATI since the 4800 rumors first started.

Yea but you have to admit the last few ATI/AMD releases (looking at you R600) have been really well..."flop"-ish at best. So one would be wise to be a bit more hesitant. I mean when I saw how much bandwidth the R600 cards had with its 512 mb interface, it was game over...well it was...just not how we all thought :)
 
Yeah. I haven't seen so my denial since I asked my 7 year old niece if she ate some cookies, despite the chocolate on her face.

btw, just like I predicted yesterday, AMD has finally officially released drivers/hotfix that includes the 48xx series. It'll be interesting to see the differences from before:



http://support.ati.com/ics/support/default.asp?deptID=894

no doubt scores will be higher with the hotfix.... need to see new scores!!!
 
What is the story with anti-aliasing did they ever come up with 8x or 16x modes that don't blur the image? I remember this was an issue with the 2900XT and one reason why I switched to Nvidia.
 
Back
Top