Last edited by a moderator:
Digital Foundry Article Technical Discussion Archive [2013]
- Thread starter helio
- Start date
-
- Tags
- digital foundry
- Status
Why would we compare a GPU with 12 CU's and 68GB/s shared memory bandwidth to it's main memory pool plus 32MB esram at a rumoured 196GB/s bandwidth achievable only under very specific circumstances to a 32 CU GPU with 264 GB/s of dedicated and guaranteed memory bandwidth to it's main memory pool?
Major Nelson is that YOU ??
what? read the digital foundry post ms has increased the esram bandwidth to 196GB/s. 196+68 = 264.
on another forum, a programmer/engineer explained how they increased the bandwidth.
i already understood from a cu perspective, and i don't necessarily mean only to a 7970, but i mean higher parts in general. we compare the ps3 to a 7850, wouldn't it be just as fitting to compare the xbox one's gpu to a 7850 because of it's bandwidth? i mean it has more than ps4 does now so surely if ps4 can be compared why not xbox one? it just seems to me xbox one's gpu CUs aside is really beyond a 7770/7790. from what i can understand wouldn't having that much bandwidth on esram completely make a 1 to 1 comparison of CUs with desktop parts kind of pointless?
Is there a link?
How did they increase it?
Eurogamer are pretty clear on how Microsoft are coming to the larger number. It's not that performance has been increased, it's just that they have supposedly found an exploit to eek more performance out of the existing hardware. I find the claim extremely dubious to be honest, something that's more aimed at giving forum warriors some extra ammunition in console wars than something that's actually usable in the real world. I mean, are we seriously supposed to believe that Microsoft only just found this out and had no idea this was possible during the consoles design and testing phases? Here's what Eurogamer say:
Eurogamer said:
I've bolded the important bit since it makes quite clear that this exploit is only usable under very specific scenario's and even then is achieving no-where near the "theoretical 196GB/s even in this specific case, nevermind in general use.
In other words, I don't see this being in any way comparable to a large pool of GDDR5 that has a standard 264GB/s memory interface.
If you are going to pretend it has 264GB/s bandwidth, might as well pretend it has infinite bandwidth and do a thought experiment - where is the bottleneck now? Is it the 12CUs? If so what is a good amount of bandwidth to pair with 12CUs since you don't want to pay for infinite bandwidth? Well look at the discrete AMD parts, then add 20-30GB/s for the part that the CPU and IO use. What is the number you get? Probably around 80-110GB/s.
So do think MS made an unbalanced system with too much memory bandwidth?
http://forum.teamxbox.com/showpost.php?p=14035488&postcount=2623
double pumping is the assumption.
but doesnt the esram help alievate the loewer number of CUs?
I would not go with that assumption.
The use of the terminology is not consistent with what is commonly described as double pumping for the example DDR memory.
I find the idea of smudging operations so that they can fractionally spread over 15 of 16 cycles dubious. The application of quantum tunneling to make a digital interface magic up twice the performance is not compelling to me.
When eSRAM can do Flops, maybe.
This was based on earlier work in Cloud Chamber Computing
I am aware of the "bidirectional computational hole" thing from the MS insider claim and I can only say I wouldn't bet the farm on 192 gB/sec or 196 or whatever for all we know that number could be the result of simulations or driver diagnostics or who know.
Oh and the Major Nelson thing.
So much faster than the PS4 I don't know why they bothered with an update
Betanumerical
Veteran
You would be wise to take whatever astrograd says with a truckload of salt he commonly misunderstands technology, this isn't meant as a dig, its just the truth.
It should also be noted that unless Microsoft has also increased the ROPs they can't fill faster then the original bandwidth anyway, which leads me to believe that all its talking about is the compression on the GCN based GPU and comparing it to uncompressed bandwidth.
Last edited by a moderator:
ESRAM is just help feeding the CUs, the CUs are doing the work (rendering).
I think that it's possible in some games you could see evidence of fewer bandwidth restrictions on XBO, personally. Think PS2 vs 2001 Xbox.
One thing people consistently seem to give little thrift too that I continually bring up is that XBO has less flops to feed per unit BW.
This seems pretty dubious, DF has a history of digging up arcane technical documents and exposing them, the same as they did for PS4's RAM reservation. You'd have to basically think MS circulated some arcane technical document just "knowing" that DF would find it, write it up, and that people would care. Pretty far into tinfoil hat territory.
Besides all that MS has shown little interest in discussing any aspect of Durango's technical specs. They are perfectly comfortable apparently with the near ubiquitous assumption that PS4 is more powerful.
"Maybe". It seems theoretically possible but so far we've heard precious little hard evidence of this.
All one needs to do is look at furmark on PC to see GPU ALU's can be utilized more or less heavily with a large range there.
But there are many who are quick to jump to incorrect conclusions, such as the general accepted wisdom that XBO has only 68GB/s BW altogether. So I do not feel it is correct to say that ESRAM can only possibly help with memory BW alleviation as some prefer to conclude.
There have been some hints that MS may discuss how they feel their hardware is competitive, later imo. Possibly waiting for all hardware (both XBO and PS4) to be completely finalized and in production before talking about it. More may come out then on this topic.
So much faster than the PS4 I don't know why they bothered with an update
You're using as an example a console with less main BW and a small pool of EDRAM, that in BW constrained situations often outperformed a console with ~2X the main BW. Sound familiar?
The ESRAM is going to be a fun thing to learn about even without the dubious speculation of double pumping and quantum tunneling and god knows whatever else. When in doubt the simplest explanation yada yada.
Last edited by a moderator:
Let me see, cherry picking text out of context, then making an argument that has little to do with the text and all at the same time ignoring the original post and its claims/content. Well played, you got us.
So for the sake of argument, you don't think the ESRAM is feeding data to the CUs and you don't think the CUs are doing the computations? I'm unsure if you are agreeing with Solarus's claims or just trying muddy the waters with more hand waving secret sauce talk. I thought it was well understood that the ESRAM exists to alleviate the low system bandwidth (68GB/s - CPU/IO overhead is not enough for 12CUs) and I also thought even with 600GB/s bandwidth, 12CUs have a limit to what they can do. Do you disagree with these simple assumptions?
So for the sake of argument, you don't think the ESRAM is feeding data to the CUs and you don't think the CUs are doing the computations? I'm unsure if you are agreeing with Solarus's claims or just trying muddy the waters with more hand waving secret sauce talk. I thought it was well understood that the ESRAM exists to alleviate the low system bandwidth (68GB/s - CPU/IO overhead is not enough for 12CUs) and I also thought even with 600GB/s bandwidth, 12CUs have a limit to what they can do. Do you disagree with these simple assumptions?
The eSRAM stuff was apparently from developer sources. I'm not sure that keeping the dev environment up to date should count as PR from MS.The DF either got info from MS insiders or from devs who got their info from MS insiders at least when it comes to the "bidirectional computational holes" and other recent stuff, not from arcane technical documents. It's a simple PR move not unlike putting Cerny out there but at least he is answering questions. MS is quiet about the hardware because MS is quiet about everything. Remember specs don't matter so why discuss them. All kinds of speculation arises in the absence of facts and all kinds of sauce remain to be found in the gaps of knowledge. I'm reading lots of noise about the XB1 CPU being either piledriver or a "proper GC 2.0 custom APU" ...
The "developer source" could work at a Microsoft studio, and leaking "new" details through Leadbetter could be part of the same whisper campaign that brought us the reddit AMA from a "confirmed Xbox One dev".
Yeahhh. A company that has PR mouthpieces in place decides that the best way to advertise an obscure number is through the grapevine.
I think you need to up your meds, the paranoia is coming back. I worked at MS for 9 years, and in my experience, they're just not that convoluted.
Samsung did this kind of things and I think MSFT do this too. I would be surprised if they didn't - it is a multi milion dollar business - PR is handled in every possible way.
If there were concrete technical advantages they wanted to tout they'd go through an official PR mouthpiece. If the strategy is to sow FUD, then Digital Foundry is proving to be a very fertile platform for that. I don't think there's anything "convoluted" about the image rehabilitation strategy they are currently engaged in for the Xbox One.
- Status
Similar threads
