QuoteAMD-ATI Merger Looks Likely
Read the rest HERE (http://www.forbes.com/2006/05/31/amd-ati-technologies-0531markets10.html?partner=yahootix)
could be nice if it did go through, likely see some SHINY stuff coming out if the companies tied themselves together properly and developed stuff alongside each other.
interesting, could result in some sweet AMTI CPU/Chipset/GPU combos if they do it right.
Could we also see the emergence of motherboards that take a drop in CPU/GPU all in one core? :twisted:
Quote from: WalrusbonzoCould we also see the emergence of motherboards that take a drop in CPU/GPU all in one core? :twisted:
i thought that, would be nice, though it would mean not being able to upgrad GFX very often. Good for SFF systems though :D
What i was thinking was if AMD had full control of the GFX Chipset as well they could likely build it into the Hypertransport bus, now THAT would be impressive. Likely need a less powerful GPU because the connection to the CPU would be much much faster. Could be interesting to say the least
Quote from: BeakerQuote from: WalrusbonzoCould we also see the emergence of motherboards that take a drop in CPU/GPU all in one core? :twisted:
i thought that, would be nice, though it would mean not being able to upgrad GFX very often. Good for SFF systems though :D
What i was thinking was if AMD had full control of the GFX Chipset as well they could likely build it into the Hypertransport bus, now THAT would be impressive. Likely need a less powerful GPU because the connection to the CPU would be much much faster. Could be interesting to say the least
Err. NO.
AGP and PCI-Express transfer rates have shown it makes virtually no differerence. GPUs rely on their own RAM speed and not a lot else.
Quote from: WalrusbonzoQuote from: BeakerQuote from: WalrusbonzoCould we also see the emergence of motherboards that take a drop in CPU/GPU all in one core? :twisted:
i thought that, would be nice, though it would mean not being able to upgrad GFX very often. Good for SFF systems though :D
What i was thinking was if AMD had full control of the GFX Chipset as well they could likely build it into the Hypertransport bus, now THAT would be impressive. Likely need a less powerful GPU because the connection to the CPU would be much much faster. Could be interesting to say the least
Err. NO.
AGP and PCI-Express transfer rates have shown it makes virtually no differerence. GPUs rely on their own RAM speed and not a lot else.
still not go direct connection to the CPU though, still using a shared bus, and still having to go through the the nothbridge to get the CPU. Thats a bottleneck and 8xAGP wasnt exactly stretched even with the latest GPUs. If you can skip the bridge, and directly connect to the CPU then you would be on a winner. However with present tech is isnt feasable, if all the hardware was in-house then its more likely someone would try it. Think of all those wasted cycles you get at the moment when the PCI/AGP/PCIe isnt feeding data to the GPU fast enough....
No, what I was saying was, look at the difference in performance going from AGP1x(133MB/s) to 2x(266MB/s), to 4x(533MB/s) to 8x(1.06GB/s) to PCIe(About 5GB/s for a 16x link, someone correct me as I may be wrong). Massive bandwidth difference, and does it cause any actually increase in GFX performance, no, not a lot at all.
So, even if we did have a direct GPU to CPU link, would it help, no, not a lot.
Quote from: WalrusbonzoNo, what I was saying was, look at the difference in performance going from AGP1x(133MB/s) to 2x(266MB/s), to 4x(533MB/s) to 8x(1.06GB/s) to PCIe(About 5GB/s for a 16x link, someone correct me as I may be wrong). Massive bandwidth difference, and does it cause any actually increase in GFX performance, no, not a lot at all.
So, even if we did have a direct GPU to CPU link, would it help, no, not a lot.
but at this point you are talking present technology. If you can directly connect all of your components to the CPU then the overall speed differentials improve. if the tech can be integrated together properly then you can do away with the proprietory interfaces. That would speed things up a hell of a lot. At present the PCI/PCIe bus system works ok, but if you can bypass the north and southbridge you are looking at a massive performance boost provided the software is there to support it.
Quote from: WalrusbonzoCould we also see the emergence of motherboards that take a drop in CPU/GPU all in one core? :twisted:
The way things are going you might end up with a computer on a chip. TBH the most recent CPUs have plenty of room for northbridge and a load of memory so you ed up with a two chip computer, in a notebook with low power a priority it could mean a much longer battery life and reasonable performance.
Quote from: SeriousQuote from: WalrusbonzoCould we also see the emergence of motherboards that take a drop in CPU/GPU all in one core? :twisted:
The way things are going you might end up with a computer on a chip. TBH the most recent CPUs have plenty of room for northbridge and a load of memory so you ed up with a two chip computer, in a notebook with low power a priority it could mean a much longer battery life and reasonable performance.
already been done back in the days of the 486. Cyrix did it IIRC, but the chip would have been far too expensive to produce. I have/had a copy of byte with the article in it. IIRC they where talking over $5k for the ship alone. Nobody is going to pay that much for a CPU.
Someone will no doubt do it again though, its well within present technical possibility to put every single controller inside the chip itself, and then just use external connectors and the odd diode and cap to smooththe signals out before they terminate on the connectors. Similar technology is already used in very specialist applications.
I read the article too, the main problem with it though was that it was slow, very slow IIRC.
Hopefully they will be able to do something better.
Cyrix produced that chip?
The same people that didnt see fit to install a math co-processor on many product lines? :roll:
No wonder they went under.
Quote from: SeriousI read the article too, the main problem with it though was that it was slow, very slow IIRC.
it was a Cyrix, what do you expect? they survived as long as they did by lying to punters about how fast their chips were. The number of people I spoke to when I worked in a comp store who said "My new P200 is only doing 150Mhz, whats up with that?" and I had to explain to them the whole PR thing (as long as you run integer apps its faster than a Pentium 200, just whatever you do dont try and run anything that needs floating point maths). Not to mention of course you had to overclock your FSB just to run a Cyrix 200+ at all cause they ran on 75Mhz * 2 and the chipsets of the time officially topped out at 66.
stupid bunch, good ridance I say.
If AMD/ATI give the same thing a try Im sure theyll do a much better job
Quote from: MongooseQuote from: SeriousI read the article too, the main problem with it though was that it was slow, very slow IIRC.
it was a Cyrix, what do you expect? they survived as long as they did by lying to punters about how fast their chips were. The number of people I spoke to when I worked in a comp store who said "My new P200 is only doing 150Mhz, whats up with that?" and I had to explain to them the whole PR thing (as long as you run integer apps its faster than a Pentium 200, just whatever you do dont try and run anything that needs floating point maths). Not to mention of course you had to overclock your FSB just to run a Cyrix 200+ at all cause they ran on 75Mhz * 2 and the chipsets of the time officially topped out at 66.
stupid bunch, good ridance I say.
If AMD/ATI give the same thing a try Im sure theyll do a much better job
Lol was a funny time indeed :D