T43p graphics obsolete?

[bert]

I am not sure what you guys are agonizing about. Why would you want to pay the same for old technology as for new? Based on a small difference running some particular benchmark? In real life you probably could not tell any difference in performance between a 9600 and an x300. They are both based on the same level of technology. One is the last member of the AGP era, and the other is the first member of the PCI Express generation.

When I started this thread the issue was dissatisfaction with the T43p coming out with the x600 and not the x700. I have a T40p with a 9000. To me it is not worth upgrading to an x600, but I would upgrade to an x700. At the pace that technology is moving ahead, I think that it takes at least twice the performance to justify an upgrade.

[danny_isr]

exactly


there another reason why i didn’t want to buy T42 with similar spec.
on the express models they don’t have a 14" sxga Radeon 9600 with blue tooth and finger print.
they do offer models like this now with the T43

[aamsel]

Well, first off I NEVER agonize. I may go on and on, but no agony is ever involved!

1.) I would want to "pay the same for old technology as for new" if the new technology was inferior. That is what is being discussed here.
2.) You said "One is the last member of the AGP era, and the other is the first member of the PCI Express generation" but the AGP is the 9600, and the X300 is said to replace the 9200. I have now read that in many, many places. Can you point me to anything other than the information at www.rojakpot.com that says X300 replaces 9600?
3.) As for "Based on a small difference running some particular benchmark" I have stated that I normally don't run benchmarks, and don't put a lot of weight on them. However, it is well known that IBM does not turn up clock rates on hardware, or use optimized drivers. They are very conservative. So, if the X300 was superior, I would have simply expected it to get better scores across the board. It didn't. Do I put much weight on that...no. Simply stating what I found when I ran the tests.
4.) I value yours and every opinion here. That is why I posted it. I haven't made a final decision and wanted input from users here.

My hope that someone with more knowledge than me would post something valid that would sway me one way or the other.

Thanks,
Andrew
Austin, TX

I am not sure what you guys are agonizing about. Why would you want to pay the same for old technology as for new? Based on a small difference running some particular benchmark? In real life you probably could not tell any difference in performance between a 9600 and an x300. They are both based on the same level of technology. One is the last member of the AGP era, and the other is the first member of the PCI Express generation.

When I started this thread the issue was dissatisfaction with the T43p coming out with the x600 and not the x700. I have a T40p with a 9000. To me it is not worth upgrading to an x600, but I would upgrade to an x700. At the pace that technology is moving ahead, I think that it takes at least twice the performance to justify an upgrade.

[/img]

[aamsel]

Good point (not your saying "exactly")

Features ARE more important than specs.
Really ticked me off that the sales guy I bought the T43 from told me I could not add a/b/g at time of purchase. I am calling today to verify that it can be done.

Lastly, please once again understand that I am NOT discussing whether an upgrade from T42 to T43 is warranted. Not at all.
I can purchase either, and the discussion is which offers a better value.

Andrew
Austin, TX

exactly


there another reason why i didn’t want to buy T42 with similar spec.
on the express models they don’t have a 14" sxga Radeon 9600 with blue tooth and finger print.
they do offer models like this now with the T43

[Conmee]

Folks,

I love this thread... because it indicates just how much we all love our ThinkPads, and how frustrating it is to get consistent, comparable data for all the g*@@am ATI parts out there.

Not everyone can get the employee/shareholder or educational discount, I understand this. But I can... and by the way, I apologize for adding in FireGL bench data... besides drivers and more VRAM, it's the same MR9600 core, and thought, for comparison sake, people would be interested in seeing that v. X300 as well. Anyhow...

I can get the 2378FVU for $1146.75 + Free Shipping + 6% tax in Michigan as an Employee/Shareholder discount. The 266846U isn't currently available on the IBM site, so I don't know what it's going for. But the lower model 266844U with XGA+fingerprint is going for $1913.13 retail and $1429.35 + Free Shipping + 6% tax in Michigan. All things equal (and they aren't since the 2378FVU has an SXGA+ screen) there is a $282.60 pre-tax premium for a T43 over a T42 with higher resolution screen (the 266846U will be somewhat pricier than the 266844U just because of the screen difference). So what, in this scenario, does the additional $282.60 get you? A 1.86GHz CPU v. 1.7GHz, DDR2, and a PCIe GPU that wouldn't even stress an AGP bus.

So the question remains, as it did at the start of this thread: Is the performance of the X300 v. the MR9600 (and the benefits of slight faster CPU/mem) worth the price differential in buying a 266846U v. a 2378FVU? Based strictly on video benchmarks, I'd say no. I think a post above misquoted me when I mentioned "don't know why MR9600 scores are so low." The scores I received on MY T42 (not p) with an MR9600 were in line with the X300. Others have had different experiences. Who knows.

I love IBM and ThinkPads, and I, like most folks here, don't bash one ThinkPad v. another. This is an economic price v. performance discussion, not a love-hate argument. lol And in that light, I'll stick with the follow OPINION of my own:

If you own a T42/T42p, I'd wait until the T43/p has the X600/700/800 or V5200 FireGL before upgrading, IF.... graphics performance is the key driver. If you don't own a ThinkPad, and you are looking at the mid-range or low-end T43's, and you want to save a little money, there is tremendous value in getting a T42 of comparable specs right now, especially if you can get an Employee/Shareholder/Education discount. There's not so much new whiz-bang technology in the T43 to feel you're buying an 'obsolete' system if you choose a T42/T42p. Of course, I'll change my tune on this point later in the year, early next year when dual core Pentium M's show up and 800MHz FSBs... lol

It's all good, and I love my T42p... just like I loved my T41p, T40p, T22, etc...

Daniel.

[jo5h]

WEB price for 266846U is $2129 (35% off with EPP). If 3 year warranty and 512MB RAM upgrades are added on 2378FVU, price difference is about $20.

T43 would be a better deal, don't you think ?

I can get the 2378FVU for $1146.75 + Free Shipping + 6% tax in Michigan as an Employee/Shareholder discount. The 266846U isn't currently available on the IBM site, so I don't know what it's going for. But the lower model 266844U with XGA+fingerprint is going for $1913.13 retail and $1429.35 + Free Shipping + 6% tax in Michigan. All things equal (and they aren't since the 2378FVU has an SXGA+ screen) there is a $282.60 pre-tax premium for a T43 over a T42 with higher resolution screen (the 266846U will be somewhat pricier than the 266844U just because of the screen difference). So what, in this scenario, does the additional $282.60 get you? A 1.86GHz CPU v. 1.7GHz, DDR2, and a PCIe GPU that wouldn't even stress an AGP bus.

[/quote]

[aamsel]

I have ended up ordering an "Open Bay" T43, and customizing it the way I wanted it. I ended up with an IBM a/b/g card, 512MB, and 60GB 7200 for about $25 more than a 266846U, and close to the 2378FVU. (after adding memory and 3 year warranty to the FVU.)
All in all, the best "bang-for-the-buck" that I have seen, and I am quite pleased with the decision.

Andrew
Austin, TX

[zhaozhong]

Can anyone having T43 with X300 do some graphics benchmark with two dimms both installed with same capacity memory? (e.g., 512MB+512MB or 256MB+256MB)

Sonoma is suppoed to support dual channel DDR II memory. With only one SODIMM installed, T43 will definitely use single-channel mode, which has 4.2GB/s memory bandwidth. Suppose the BIOS activate dual channel mode when two identical SODIMMs get installed, the memory bandwidth will be 8.4GB/s. The dothan cpu itself only requires 4.2GB/s bandwidth, so the other 4.2GB/s in dual channel mode is mainly reserved for graphics.

According to Radeon mobility X300 specification, it is basically same as 9600 except that the built-in 64MB memory is 64-bit instead of 128-bit. And that explains the difference in benchmarks. However, X300 CAN actually use 128-bit memory interface. The other 64-bit comes from PCI express, which is a direct port to system memory, and can befenit from the extra 4.2GB/s bandwidth in dual channel mode.

The theory is, currently all the benchmarks we saw are with single channel mode. If T43 can enable dual channel mode, then the benchmark should increase a lot. In fact, in some desktop X300 Hypermemory (which is essentially same as mobility X300) test, benchmark scores like 8695 in 3dMark01 can be found for a 32MB version of X300 with i925 dual channel memory. Since our T43 has 64MB X300, as long as we enable dual channel mode, we should be expecting at least 9xxx 3DMark01 score, which is on par with the T42 in 9600.

[aamsel]

My understanding was that the Thinkpad only supports Sonoma in single-channel mode.
Correct me if I am wrong.

Andrew
Austin, TX

[zhaozhong]

This may very well be the case. But we do not have direct proof. There were previous posts that says single and dual channel configuration show similiar system memory bandwidth scores in testing, which is not really surprising as the Dothan CPU can only take 4.2GB/s bandwidth even if dual channel setting can provide 8.4GB/s.

My understanding was that the Thinkpad only supports Sonoma in single-channel mode.
Correct me if I am wrong.

Andrew
Austin, TX

[aamsel]

I am sure that there is a T43 owner that can definitively answer this for us.

ANYONE ???

Andrew
Austin, TX

This may very well be the case. But we do not have direct proof. There were previous posts that says single and dual channel configuration show similiar system memory bandwidth scores in testing, which is not really surprising as the Dothan CPU can only take 4.2GB/s bandwidth even if dual channel setting can provide 8.4GB/s.

My understanding was that the Thinkpad only supports Sonoma in single-channel mode.
Correct me if I am wrong.

Andrew
Austin, TX

[zhaozhong]

Besides, can any T43 owner take a look at the default core and memory frequency of X300? This could also be the reason for the low score of X300 in 3dmark01.

[aamsel]

IBM always underclocks (or minimally sets the clock) for GPU's.
They are business oriented, they are not looking for framerate on some game.

Andrew
Austin, TX

Besides, can any T43 owner take a look at the default core and memory frequency of X300? This could also be the reason for the low score of X300 in 3dmark01.

[zhaozhong]

I was aware of that. I am just curious this time how much they cut off the x300 reference design, which is 350Mhz/300Mhz.

IBM always underclocks (or minimally sets the clock) for GPU's.
They are business oriented, they are not looking for framerate on some game.

Andrew
Austin, TX

Besides, can any T43 owner take a look at the default core and memory frequency of X300? This could also be the reason for the low score of X300 in 3dmark01.

[Pascal_TTH]

Very nice topic about GPU. I noticed in your fight for who is the best between Mobility Radeon X300, Mobility Radeon 9600 and Mobility FireGL T2 that you forget to care about bus size.

I buy a second hant T41 for my wife. This laptop is the same as his previous Compal CL50, but it's a Thinkpad. Joke. The main difference is the graphic memory bus. CL50 use a Mobility Radeon 9000 with 64 Mb using 128bits memory bus. T41 only have 32Mb with 64bits bus. In benchmarks or games, the performance drop is about 30% due to less memory bandwidth. CL50 reach 1100 in 3D Mark 2003 and T41 scores at 625 (lower frequencies) .

Also, you can expect better performances with Mobility FireGL not because it's a high-end GPU, but due to 128 Mb. For exemple, 3D Mark 2005 needs 128 Mb for all graphic datas. If you got less graphic memory, it will swap with central memory and performance drops. It's like Windows running out of memory and start swapping with HDD.

Also keep in mind that Dothan is far better then Banias in 3D applications. I upgrade my T41p and here is what I get :

IBM Thinkpad T41p upgraded from Banias 1,7 GHz to Dothan 1,7 GHz

Score 3D Mark 2003 : 2536 -> 2830 (+11,6%)
Score 3D Mark 2005 : 1000 -> 1247 (+24,7%)

If you compare withe this :

To finish up with the question of the 3DMark05 scores:
2668 with 64MB X300 got a score of 710 (T43 1.86GHz)
2378 with 64MB 9600 got a score of 688 (T42 1.70GHz)

[christopher_wolf]

Bus size is important, not so much the VRAM. As an example, my Dell M60 had a nVidia FX card with 128MB; now you may think it did pretty well on the graphics benchmarks...well, you would be really wrong...Problems with it;

1.) It was a nVidia FX card.
2.) It had, as far as I could tell, either crippled drivers or a limitation being imposed by the bus structure.

Consider that my T43 with a 64MB physical VRAM x300 managed to blow it out of the water when it came to games like HL2 (that exhibited the most difference between the two systems) and Far Cry (the most graphic differences showed up here).

With Far Cry on the Dell, the graphics [censored] near made it unplayable. If you were walking towards a mountain, at some random point, it would change into a flat textured ramp and then flip back and forth between a reasonably rendered hillside and a sandy/dirt-colored ramp. Bushes and other sprites would morph into blobs, etc....This has happened on two Dells I have tried it on with the latest drivers and that chipset. It only managed to stutter through HL2 even though it was at the minimal settings and not using the wide-screen on the Dell.

Compare that to my T43 with only 64MB of physical VRAM; HL2 was smooth, far above the 30FPS and 60FPS limits, throughout all the explosions and physical actions @ 1024x768 with AA on. In Far Cry, trees actually showed up as very well done trees (and not fleshy columns with green tops that would warp back into the image of a palm tree).

Watch the specs on all the cards carefully; there has to be the proper balance in everything. Simply having more VRAM isn't going to cut it.

[Pascal_TTH]

Until now, we were only speaking about X300, 9600 and FireGL T2 witch are all based on the same core or with minor changes. About memory size, 128 Mb improve performance in a bench like 3D Mark 2005 or a heavy texture game. But I only speaks about the same gpu with 64 Mb (Radeon 9600) or 128 Mb (FireGL T2 128 Mo). In other case, for exemple 3D Mark 2003, older games or games with small amout of texture, there will be only small differences.

All GeForce FX are known to have poor gaming performances. There were close to Radeon when NVIDIA only have them and always works on drivers. Now, they only care about GeForce 6 et 7 series. They even admit it when they launch the GeForce 6800 at Geneve event in 2004.

3D graphic power comme from *few things* :
- number of pixel pipelines
- number of vertex engine
- internal architecture (sur GeForce FX, GeForce 6, etc) thus DirectX supported
- core frequency
- memory bus
- memory frenquence
- type of memory (now only DDR type)

Pixel pipelines and core clock give the fillrate. It can compared to a brush for painting the screed with the 3D image. More clock or more pipelines give more power.

Architecture is important DirectX 8 or 9 graphic class can do more complex effect then DirectX 7. So they can paint the screen un one pass. An older one may need two pass to do the same effect. Also the pixel pipeline efficience may vary a lot (very poor un GeForce FX, very high in all Radeon 9600 and similar).

Vertexe engine and core clock give the geometric power. It's quiet harder to estimate how it impact performances. High power is great for CAD using wired rendering.

Memory type, memory bus and memory clock drive to memory bandwidth. You can have the better GPU, if it didn't get enough data, he wastes his time and gives poor performances.

Graphic memory size have only impact if you didn't have enough ! If you got a lot of amout but poor bandwidth it can runs fine on some games but poor on other ones. It depends how the game engine is : large texture intensive, geometric intensive, shader effect intensive.

Radeon X300 aka RV370 is the PCI-Expess version Radeon 9600 (RV360). X300 is a die shrink from 9600 using 11nm instead of 13nm. Both have about 75 millions transistors and are DirectX 9 class. They also have two vertex engines witch can handle shader 2.0 and 4 pixel pipelines with pixel shader 2.0 and 1 texturing unit per pipe. Memory interface is 64bits or 128bits but controler is the same.

Their can only differ in the way IBM use them on the Thinkpad. So only core clock, memory clock and bus size can be differents. In desktop world, X300 et 9600 with same clocks and bus width give similar performances.

Do you know the memory and the X300 clocks of your T43 ? I'am working on a mobile GPU data base (3D Mark 2003 and 2005).

[Hanson]

Default values on my T43:

Core Clock: 300 MHZ
Memory Clock: 230 MHZ

[Pascal_TTH]

For Mobility FireGL T2 and I suppose alos for Mobility Radeon 9600 :

VGA Core Clock 317 MHz
VGA Memory Clock 203 MHz

So for game/bench where memory bandwidth in more important, X300 may have better performances while in geometry or fill rate is more important, R9600/FireGL T2 is ahead.