Wednesday, October 25th 2006
PowerColor to Release AGP version of Radeon X1950 Pro
According to DailyTech, PowerColor will be releasing a Radeon X1950 Pro based product for AGP systems. Images of the proposed Radeon X1950 Pro AGP graphics card are also revealed. The upcoming Radeon X1950 Pro AGP is expected to be clocked higher than its PCI Express counterparts at 600 MHz core and 700 MHz instead of the 575 MHz core and 690 MHz memory on PCI Express version. In addition to the higher core and memory clocks, PowerColor will equip Radeon X1950 Pro AGP graphics cards with a quiet Arctic Cooling Accelero X2 cooling solution. If this thing really hit the market, this will probably be the fastest available AGP video card.
Source:
DailyTech
50 Comments on PowerColor to Release AGP version of Radeon X1950 Pro
Bc if they didnt make it big and powerful enough then it will easily kill the card or drastically shorten the life of the card.
Its better to be safe then sorry.
I got my home-made SFF Pentium M 1.5Gz totally silent this way. You can do it too!
Reason that its being sold as a 2 slot? Retail value-added, and quieter. Nearly all enthusiasts put an aftermarket cooler on their GPU, so they are doing this as part of the original package. For 90% of buyers this is a good thing and a lower cost overall. Downside to single slot fans, but you can fix it.
If this card is to be had on the market, then ATi will still be King of AGP. :D
Is it worth paying for this when you have maybe the second best agp card with an ATI X850Xt? Definately will need to see some benchmarks.
One other thing:
I suppose this will run DirectX 10 and have all the latest eye candy thingies..?
But from what I have heard they say that the 8 lanes on the AGP bus is not enough and will bottleneck this GPU, unlike the PCI-Express version.
If this is true than this card will be the first to do so and will show that the PCI-Express version is better than the AGP.
@AnneCore. Yes. Just compare PCIe X850 vs X1950pro. Big difference. Second. Encoding DivX. That ATI codec that operates 8x faster than a standalone CPU means you can encode a DVD in tens of minutes not hours. Third. Dual-link DVI to drive 2 TFT's or a super hi-res TFT like the Apple Cinema Display 30". You can't do that on an X850.
@bruins004. Most tests to date show that the AGP isn't even saturated yet. So in theory although the PCIe lane has 2x the bandwidth (one way... since we write, dont read to the GPU), it makes no performance difference.
AGP was originally designed to deal with ripping textures from main memory onto the GPU which (5 years ago) had only 8 or 16MB of RAM. Now that GPU's have 256MB or 512MB, there really isn't that much data going from main memory to the GPU. If there is any transfer, its coming off the HDD. And that's the bottleneck. The HDD interconnect, not the GPU interconnect.
So I think the benchies will show identical performance.
If not bottlenecked at PCIe4, then with AGP the speed is 2x > than PCIe4, there wont be a problem.
1./ There are many versions of AGP. What I am refering to, and only refering to, is AGP version 3.0 also known as AGP 8x with a bandwidth of 2112MB/s. I will refer to AGP 3.0 (8x) as AGP for short. Any mainboard with a lower AGP specification is an old dinosaur and is not part of this discussion.
2./ PCIe16 has twice the "one way" bandwidth compared to AGP. PCIe16 can sustain 4000MB/s one-way.
3./ Because PCIe16 can read and write in both directions, sometimes PCIe16 is claimed to have 8000MB/s bandwidth that is nearly four times the bandwidth compared to AGP.
4./ However, since we "write" and dont "read" to GPU then this point is moot.
5./ Hence, from a GPU perspective, we can think of the PCIe16 as being twice as fast as AGP, 4000MB/s compared to 2112MB/s. Also, PCIe8 is about the same speed as AGP in graphics bandwidth terms, 2000MB/s compared to 2112MB/s.
6./ For interested parties, the bandwidth obtained through simultaneous read and write IS VERY INTERESTING for network cards. Basically a PCIe1 network card can operate at 2-4x the throughput of a PCI network card. And on a server, you can install network cards with multiple connections that run on PCIe4, which has 8-16x the throughput of a PCI card. PCIe also doesn't "steal" bandwidth from other slots. Whereas on PCI, this bandwidth is shared across all PCI slots.
For reference:<div align="center"><table width="60%" border="4">
<tr>
<td><div align="center">Graphics Standards</div></td>
<td><div align="center">Possible Bandwidth</td>
</tr>
<tr>
<td>
<div align="center">16-bit ISA</div>
</td>
<td>
<div align="center">16MB/s</div>
</td>
</tr>
<tr>
<td>
<div align="center">EISA</div>
</td>
<td>
<div align="center">32MB/s</div>
</td>
</tr>
<tr>
<td>
<div align="center">VLB</div>
</td>
<td>
<div align="center">132MB/s</div>
</td>
</tr>
<tr>
<td>
<div align="center">PCI</div>
</td>
<td>
<div align="center">132MB/s</div>
</td>
</tr>
<tr>
<td>
<div align="center">AGP 1x</div>
</td>
<td>
<div align="center">264MB/s</div>
</td>
</tr>
<tr>
<td>
<div align="center">AGP 2x</div>
</td>
<td>
<div align="center">528MB/s</div>
</td>
</tr>
<tr>
<td>
<div align="center">AGP 4x</div>
</td>
<td>
<div align="center">1056MB/s</div>
</td>
</tr>
<tr>
<td>
<div align="center">AGP 8x</div>
</td>
<td>
<div align="center">2112MB/s</div>
</td>
</tr>
<tr>
<td>
<div align="center">PCIe x1</div>
</td>
<td>
<div align="center">250 / 500MB/s
(Per direction / Both Directions)</div>
</td>
</tr>
<tr>
<td>
<div align="center">PCIe x2</div>
</td>
<td>
<div align="center">500 / 1000MB/s
(Per direction / Both Directions)</div>
</td>
</tr>
<tr>
<td>
<div align="center">PCIe x4</div>
</td>
<td>
<div align="center">1000 / 2000MB/s
(Per direction / Both Directions)</div>
</td>
</tr>
<tr>
<td>
<div align="center">PCIe x8</div>
</td>
<td>
<div align="center">2000 / 4000MB/s
(Per direction / Both Directions)</div>
</td>
</tr>
<tr>
<td>
<div align="center">PCIe x16</div>
</td>
<td>
<div align="center">4000 / 8000MB/s
(Per direction / Both Directions)</div>
</td>
</tr>
</table>
Now the main point is this. If AGP is as good as PCIe8 and is 2x the speed of PCIe4. But we know that PCIe4 and PCIe8 are good enough for SLI and crossfire, then we know AGP is also good enough and we will not hit a bandwidth problem on AGP.
Another issue for this card if it ever sees the light of day will be cost. AGP versions always seem to cost more and in NVIDIA’s case have fewer pipes too.
So will it even be worth the cost to upgrade now or just save up and upgrade your whole system later? I know some people here argue about waiting for DX10 however if you (we) have waited this long to upgrade why not wait till you can do your whole system and support the next generation of software/hardware.
Actually from the looks of it its clocked higher than the pci-e version and has the same pipes and specs and all that jazz. Only difference is AGP. And if you have a decent CPU to go with this card then its a good upgrade for you.
Currently the pci-e version without the stock cooler cost $164 on newegg. So I dont think this card is gonna get over $210 newegg wise.
tweet this wouldnt be a good upgrade for you because of your cpu. So the move to pci-e is actually your only real choice gaming wise.
Okay, so this card doesn't really NEED the two slot cooler after all. The GeCube version will be clocked the same as the PowerColor and the GeCube is using the one slot cooler.
I will buy the GeCube version, problem solved.
It's funny how people think that AGP is so out of date, but yet its interface bandwidth has not yet been exceeded for single card solutions.