UPDATED - The Mobile Graphics Card Info Page - Most GPU Qs answered

The WEI score is completely inaccurate.

 

Chaz, I think the Shared memory section needs revision. Someone pointed to this section with regards to HyperMemory and TC, which gives a few wrong impressions IMO.

Hypermemory is actually different from typical shared memory, as is TurboCache, both are dynamic, typical shared memory is not and must be assigned in advance. Both HM and TC are assigned on the fly, the numbers you see are not reserved like shared memory is but available for addressing like AGP aperture is in older desktops using the same GART system but unlike AGP it's dynamically allocated within the range. When needed the system has full access to that memory though unlike the case with shared memory which is directly addressed by the GPU and inaccessible to the CPU use at the same time.

As can be the HyperMemory, the X300SE 128HM was 32MB VRAM and 96 HyperMemory, and the 256HM was 128+128, it's always been that way as seen in this TechReport review;
http://techreport.com/articles.x/8396/1

The one on my HD2600 isn't 831+832MB obviously;
http://img264.imageshack.us/img264/7763/hm1663iw9.jpg
The difference is what is the ratio that the OEM has set or had ATi/AMD set for them. This too can be altered.

Both are a combination of hardware and software, the dynamic part requires software, without which both would be less effective and more impactful on system resources, check their whitepapers;
http://www.nvidia.com/object/IO_17361.html
http://ati.amd.com/technology/HyperMemory_Whitepaper.pdf

I would like to see some supporting evidence of this performance difference because I haven't seen anything consistent, which if anything showed that the most important factor remains the core and VRAM characteristics more than the HM/TC involvement as seen in the TechReport review. I don't think either has an all-out edge and relies on the efficiency of the VPUs they're paired with and their ability to use extra memory.

I wouldn't put it that way makes it sounds like it's always a drawback. Consumers should really inform themselves to avoid falling for PR, but this is a beneficial technology not an attempt to trick people. The OEM/ODMs may play up the HM/TC numbers to make them sound better than they are but that's a fault of people like DELL, HP, etc, not ATi and nVidia. When put side by side the same spec RAM with the added boost of TC or HM will always offer you more options than just simply that same RAM without. IF it goes unused it goes unused, and unlike actual Shared Memory, these two solutions don't steal from the host resources, so really they should be in a separate category from shared memory. The only time either is close is when they have no discrete VRAM frame buffer and must set aside a minimum amount of system memory to even do 2D.

Anywhoo, I think that section could use and update and could be divided into dedicated, shared and then HM and the TC separately; Or even just dedicated, shared and then HM&TC together.

 

TheGreatG****Ape
I will be buying a toshiba p200 in a few weeks
and that has an ATI HD2600 video card, in your opinion
for smooth gaming at a low/med resolution of 1024x768
how well does this GPU perform?

 

I have been planning to do revisions to the shared memory section but have not gathered enough facts to do it with.

In the past, it has been shown that HyperMemory does not provide nearly the performance boost that TurboCache does, and I have been unable to find any benchmarks to prove that wrong. Also, HyperMemory is mostly software. Any ATI card can use HyperMemory. With an Nvidia card, the card must have built-in hardware support to use it. My 8800GTX cannot use TurboCache because it has no hardware support for it. However an ATI HD 2900XT could use HyperMemory since it is software.

Note that this was written before Vista so the amount of shared memory is going to be different. It was done differently in XP.

Shared memory is beneficial but advertising a card with "512MB of RAM" is misleading. For weak cards, it doesn't matter whether they have only 128MB of memory or 128MB+128MB of memory - they are barely powerful enough to use their 128MB as it is.

Thanks for the feedback though, I will update the guide sometime today.

 

It's pretty good, but it's definitely held back by drivers if you don't have another source.

For newer games like Crysis you are stuck on lower settings, but it's still good at medium, and putting less useful things like shadows on low.

1024x768 will likely do well on title in medium settings for the next year. Some may need additional tweaking but it's fine.

 

So you're saying the MMU makes this big difference over ATi's own HypeMemory Memory Controller? Unlike the PR neither really has direct access to the memory both have to go through the chipset to get to it, their effectiveness and 'efficiency' at doing this is questionable despite the claims.

They both have proprietary dedicated hardware, the difference is ATi made it a product-line wide features, just like Hyper-Z, etc. That nV chose to omit TC from some options doesn't make it more/less effective simply because they have decided on an optional feature for some and not for others. That would be like giving value to the RAMDACs or TMDS due to the NVIO being place outside the die on the G80 and making them optional to the core (good idea if you go multi-die to a package saves duplication).

Most tests really don't show the differences in the implementations, heck it's hard to find card with even close specs most having one with a faster memory or core. Most of the times the reviews are talking about the solutions where a GF6200TC-128 may be an overall better value/buy than an X300-32HM, not that method A or B is better.

Note that both the TechReport is using XP and is still asymmetrical.

It's funny you mention Vista because IMO the most important component of Vista towards this areas isn't even implemented yet, which would be virtualization of memory allowing greater control over the resource by the OS and the Dev. This would be a feature currently exclusive to one architecture, but also would require M$ to follow through with it's implementation with DX10.1 in Vista SP1, so if anything Vista could potentially change things even more for this type of solution, but that big change hasn't happened yet.
BUT if it does then I'd be interested in seeing how this optional DX10 mandatory DX10.1 feature impacts the performance of these solutions in rigs with a ton of extra memory space (like mine) and that may be able to use this feature (hopefully the HD2600 [if not locked out by a D3D10.1 hard cap]).

Anywhoo, could be some interesting changes shortly, especially with cheap memory leading to more people packing their PCs to the gills with memory. I use mine mainly for video editing, but I sure would love a little boost in titles like Crysis and COD4.

 

We found by testing on the forums on this site that TC was a better substitute for dedicated memory than HM. I am not against what you're saying, but I do need to see definitive proof to have any changes implemented.

I remember that we benchmarked the Go7400 64MB+64MB versus the dedicated 128MB one and found that their performance was near identical. If you benchmark a 64+64 X600 versus a 128MB dedicated one, you'll find that HM is not able to compensate for the lack of memory as well.

The shared memory portion of the guide was refreshed today as I said it would be.

Edit: Information here on how ATI HM/Nvidia TC work:
http://www.anandtech.com/video/showdoc.aspx?i=2413&p=2

 

I would just recommend then that without knowing why it occurs, then I would change the whole "because it's a hardware solution" line to "from our experiences with user feedback here TC outperforms (or appears to outperform) HM. To me it's simply a question of a conclusion I see no support for ouside, but if the random sampling here says it, I'm cool with that, hey population stats are as scientific as other techniques. Just had an issue with the reasoning being the over PR'ed MMU versus the HMMC. I still think it's probably a preference thing for users, but at least I can understand user feedback better.
I just get the impression by the way they're both painted you're simply against both technologies and see them as drawbacks rather than benefits. To me the PR BS is definitely a drawback, like I said and nV's own PR people mention in the TR article, it's really the OEMs to blame for that no nV and ATi/AMD, but overall I'd much rather have it than not. Sure dedicated > addressable all things being equal, but I'd rather have 256+1GB than just 256MB alone, especially if they didn't cheap and go DDRII 512MB instead of GDDR3 256MB limiting me in frequency and power potential.

I would need to see the comparisons in question, cause far tooo often what seemed like apples to apples at the time for may turned out to be different specs. The one I provided offers the greatest amount of variability in a nice clean package not to show superiority, but just the variability even within their own lineup.

The best would be an apples-to-apples comparison with similarly clocked and similarly speced memory on VPUs of similar power and limitations. In the example you have the X600 would be noticeably weaker to begin with than your average GF7400, in texture fill rate and vertex shader power. So it would depend alot on the tests. Also isn't the Go7400 64bit VRAM and the X600 is 128bit VRAM with clock rate difference not hgih enough to overcome the bitwidth restriction on the Go7400? That would put it relatively lower on the VRAM to host RAM differece, so if you had an equal availability of memory resources from the host then the impact would appear larger on the Go7400 than on the X600 which has less of a local VRAM choke, so of course it would prefer local to anything running over a single lane.

Just FYI, If you read the Anand link you'll see it says pretty much everything I did (including the hardware vs more hardware and virtualization in Vista/DX10.1), although it does draw conclusions about the implication of the differences (somewhat ommiting the differences with the VPU's handling of memory, which always favoured ATi up until the G80 came along) without actually testing them using something more synthetic which would focus on both memories' limitations and bus influence.

Anywhoo, I've said my peace and I'll leave it to you to absorb or discard, just wanted to point out the issues I noticed as I'd been pointed to this thread by someone else and trying to reduce confusion.

 

Sounds good to me, I'll implement the changes.

The Anandtech article says the same thing yeah, I only wanted to show the page I linked to about how ATI's solution is more software-based.

Thanks for the feedback GreatG****Ape, I do appreciate it. I'll give you credit for the revisions I make to the shared memory section. It may have seemed like I was giving you a hard time but I wasn't - I simply need to verify everything that goes into this guide. The GPU Guide been viewed well over a half-million times so I have to make sure everything's backed up with solid information. It's nice to see members like you using their time to help improve the site. Much appreciated, again.

 

Sounds good.

No worries about the 'hard-time' I understand, for something like this you wanna make sure it's not some fanboi's fantasies by forum 'irregulars' (in both senses of the word), better you push for clarification than just accept people's claims.

I think I was just getting little dissapointed/frustrated that I may have not been clear about things and perhaps causing more confusion than clarifying like I had originally intended. Being dyslexic doesn't help when you start posting long explanations and occasoinally cross over yourself.

Anywhoo, glad I could be of some help.

 

so what are the some portable laptops (5.4lb or lower) that uses Performance Video Cards??

 

The Apple MacBook Pro (15.4", 5.4 lb, 8600M-GT); Asus F8 (14", HD 2600); Asus Lamborghini VX2S (15.4", 5.6 lb, 8600M-GT), Dell XPS M1530 (15.4", 5.8lb, 8600M-GT).

Those are all the ones I can find. Your options are quite limited as you can see.

 

ouch v. limited indeed xD..

if i O/C a 8400 will it reach similar speed compared to the crippled MBP GPU?

 

Hi im buying a laptop and have an option for the graphics cards but dont know which one to go for. i could go for GE force 8600M GT or Geforce 860M GT. Which one should i go for and which one is better.
Thanks

 

Those two cards are the exact same... you just misspelled them both times

 

What about Nvidia 8600m GS and 8600m GT ?

What is difference ?

 

The 8600M GT has double the stream processors (32 vs 16), and is a better card by a fair amount.

 

Ok, so ive heard mixed opinions. Is it any better? Is it worth me sacrificing a 17" for a 15" with this? The 9500M has 512MB and the 8600 has 1767MB of VRAM, even though that matters little. the 8600M is DDR2 and im not sure about the 9500M, here is the deal though...
http://1toppc.com/Merchant2/merchant.mv?Screen=PROD&Product_Code=M50SV-A1&Category_Code=

So, is it worth it? Thanks.

Also, can anyone find a 17" verson of that PC with those specs? for under $1500? Thanks again in advance.

 

9500m GS = 8600m GT + dx10.1 compliance.

 

Hmm, ok. Can anyone find thosespecs on a 17"? I cant, but that dosnt mean its not out there... Thanks Lithus +1rep

 

you should see if you can get the 9600m gt (dont know the release date)

 

theres a toshiba 17in with an 8800m gts for like $1300 @ best buy.

 

Also, I have aquestion about this from the website:


^Now im assuming, these media buttons on the touchpad dissapear when you tap the "mode" button. Thats what it looks like, but i want to make sure. Can anyone confirm this? thanks

 

The geforce 9 don't support DX10.1. According to microsoft it will almost only be used for X360 ports anyway

 

Wow, can you give me the link? Well, ill go look myself too, but if i dont find it, please post it. Thanks.

I would love to wait for the GT, but ive already waited around a month 1/2 and i want to order in the beginning of March.
Thanks again

 

It's a gateway, not a toshiba. It's already been discountinued in Canada, should be the last shipping in the US so.

 

EDIT: oh, Gateway. I dont care how cheap. Thanks anyway +1 rep everyone so far.

 

I would go with the 8600GT with the DDR3 memory. It will beat the other card time and time again with that kind of speed.

 

Well, actually, thats not an option. I can either get the 8600M GS (GDDR2) in a 17" -OR- a 9500M GS in a 15"
--Im stilll not sure if the 9500M is GDDR2 or GDDR3, no one has confirmed that. But the thing is, im dying for a 17" after having a 14" and before that a 15". Im ready for something bigger. the thing is, im going to have this for probably two to three years, so ill be stuck with something for a while, and I dont want to have a big screen and a lacking GPU, or a better GPU and a small screen. -which is why i wish i could find these specs ina a 17" to solve my problems.

 

Are you going to game on the computer at all? If you plan on gaming I would choose the 8600M. The 9500M GS does infact use DDR2 memory. Basically the only difference between the 2 cards is that the 9500M has more on board memory. The DDR3 that the 8600M GT has though, will lap the 9500M time and time again. So go with the 8600M GT.

 

It doesn't matter what RAM the 9500m GS uses. As long as it's clocked at 475/950, it'll have the performance of a 475/950 8600m GT (GDDR3 version is stock clocked at that).

 

But what im saying is: I dont have the GT or GDDR3 option. Its either 8600M GS -OR- 9500M GS, both in GDDR2. So: since the 9500M=8600M GT, since the GT>GS, then that means the 9500M GS is a little bit better than the 8600M GS. So now its just, do i sacrifice the bigger screen for a lightly better GPU? that, i have to think about.

 

Oh sorry, I totally forgot that you dont have the DDR3 option. If that's the case then yes you want the 9500M GS as it simply has more on board memory.

 

The onboard memory doesn't really affect anything from 256 to 512. So at this point it's WHEN you want to buy the computer.

 

I never said it did. But if he has the option, he wants to get more memory on the card if they are both otherwise identical.

 

The 9500M GS is definitely better than the 8600M GS.

 

would it be possible to get some benchmarks for sli setups in games as im curious to as what they are like.

 

We have a review of an SLI notebook:
http://www.notebookreview.com/default.asp?newsID=4121
Bottom line - not worth it.

 

Could you add 9500m GS? It should be just above 8600m GT, GDDR3 and DDR2 respecively. Though I don't know if there actually is a GDDR3 version.

 

May be the 9650 GS too...as chaz mentioned, it seems to be the die-shrink version of the 8700GT.

 

But the performance is not yet determined.

 

meh true... any Bench marks of the 9500GS faz?

 

Yup. Gentechpc made one. http://forum.notebookreview.com/showthread.php?t=222056

Ah-burn.

 

I'll add the 9650M-GS above the 8700M-GT and the 9500M-GS above the 8600M-GT DDR2. Thanks for the suggestions!

 

Coolio. You forgot to add it to the Big Chart though.

 

hey there! just wanna ask some q's for some of my friends... they are getting an 15.4" notebook and currently are getting it at xoticpc.com and some other place... you see, my friends are trying to get the a notebook with a good 8600-GT. All they found was the 8600-GT with ddr2 and since i've read the 1st page that has GDDR3, i really want to ask... is there a BIG difference in gaming with ddr2 or GDDR3??? graphic-wise..

 

GDDR 3 is a die-shrink version of DDR2, so it runs cooler enabling higher clock speeds, and better OC'ing ability I'd say potentially about 20%+ faster.

 

Hi All

Quick question, I am looking to buy a laptop in the next day or so. It will primarily be used for gaming.

I am looking at two laptops one with the 9500GS and one with the 8600GT.
They have similar processors(actually the laptop with the 9500GS has a faster processor), same RAM but the one with the 8600GT is $500 AUD more expensive.

Is it worth it for the extra money or should I get the 9500GS.

Cheers
Scott

 

They are pretty much the same, b/c the 9500 gs is comparable in game perfromance to the 8600gt. I say go with the 9500 laptop since its cheaper and theres basically no reason to get the more expensive 8600.

 

The 9500GS has the same exact specifications as a 8600GT. Proceed with confidence.