PC Perspective Podcast #334 – 01/29/2015
Join us this week as we discuss GTX 970 Memory Issues, Samsung 840 Evo Slowdown, GTX 960 and more!
You can subscribe to us through iTunes and you can still access it directly through the RSS page HERE.
The URL for the podcast is: https://pcper.com/podcast – Share with your friends!
- iTunes – Subscribe to the podcast directly through the iTunes Store
- RSS – Subscribe through your regular RSS reader
- MP3 – Direct download link to the MP3 file
Hosts: Ryan Shrout, Jeremy Hellstrom, Josh Walrath, and Allyn Malventano
Program length: 1:27:38
-
Week in Review:
-
0:21:15 The NVIDIA GTX 970 Saga
-
News items of interest:
-
Hardware/Software Picks of the Week:
-
Jeremy: Wireless mouse for under $20
-
Josh: Best Video Card EVAR!
-
Closing/outro
Subscribe to the PC Perspective YouTube Channel for more videos, reviews and podcasts!!
video doesn’t work
video doesn’t work
Video works for me. Small
Video works for me. Small correction at around 41:00 those who bought two 970s will have that 3.5GB fast + 0.5GB slow configuration as well as far as the video games are concerned not 7GB fast + 1GB slow, despite the way dual GPU cards are sold.
(No subject)
Goddamn Ryan, they lied about
Goddamn Ryan, they lied about the bus width also. Why have you repeatedly ignoring this? Is it because NVIDIA didn’t bring it up during your warm and fuzzy conference call, and thus you only feel comfortable parroting exactly what they said about the ROPs, Cache, and Memory Config?
THEY LIED ABOUT THE BIT DEPTH! It can only be 224 or 32, IT CANNOT BE 256. Grow some balls, take off the leash NVIDIA put on you, and acknowledge this for Christ’s sake!
Also, be a real journalist for once and check into Boris’s post here:
https://www.facebook.com/enbfx
Where he’s done some research to show that the last 512MB of RAM isn’t even being utilized AT ALL, and is running directly from system RAM.
Seriously guys, this is pathetic. You’re supposed to be an independent news organization. Start acting like it!
Thanks for your constructive
Thanks for your constructive criticism ‘Anonymous’.
The 970 still has a 256 bit memory bus as it can use all of it SIMULTANEOUSLY when reading from one segment and writing from the other. However it can’t read from both segments or write to both segments at the same time.
So in summary you are technically incorrect.
No, it can NEVER use it all
No, it can NEVER use it all SIMULTANEOUSLY, so it can NEVER run at 256bit depth.
The way to a 256 bit width is via interleaving the data across all the DRAM modules. This does not happen with the 970 because the last DRAM module is either empty or filled with low priority data. This means that the high priory data is only interleaved among 7 32bit channels – thus the 224 bit depth – and never across all eight (256).
The last 32bit module cannot be read or written to while the other modules are in use. This means that in order to access the last DRAM module the 970 has to stop interacting with the other 7 channels, read or write from the last one, and then continue what it was doing.
So the 970 can never reach 256bit depth. EVER. It can only be either 224 or 32, and to hear Borris tell it (from the link above), that 32 bit channel is never accessed and is instead stored on the much slower system RAM. If that’s true (and i hope PCPer will actually look into it) it means that NVIDIA LIED AGAIN IN THEIR MEA CULPA CONFERENCE CALL TO JOURNALISTS.
Sorry, but you are wrong.
Sorry, but you are wrong.
“achieving peak memory bandwidth performance on the GTX 970 is still possible, but it requires much more effort since simple striping will not do the trick. The easiest and most effective solution in this regard is to interleave reads and writes over the segments, such that one segment is writing while another segment is reading. Interleaving in this fashion allows both segments to work at once – avoiding the blocking effect of the shared read and write buses – and makes it more likely that both segments are doing useful work rather than waiting for their turn on an operation. However because this is only applicable to situations where more than 3.5GB of VRAM is in use and both segments are necessary, this means it’s only theoretically possible to achieve 224GB/sec when more than 3.5GB of VRAM is in use. In any situations below 3.5GB we are de-facto limited to just the larger segment, in which case there are only 7 memory channels (196GB/sec) to stripe memory operations across.”
Not the same Anonymous as
Not the same Anonymous as original poster.
The stated theoretical max of 224GB/s (which cannot ever be reached anyway due to bus overhead) is purely read bandwidth. The 970 cannot manage the same bandwidth as the 980 since 2 memory controllers share one read port. You can write to one while reading from the other, I believe, but you cannot read from both partitions in the same clock. It isn’t particularly relevant to include write bandwidth in the total. GPUs read a lot more than they write.
I would say that it is still incorrect for Nvidia to claim 224 GB/s for this card. They will probably not change it now though, since that would be essentially admitting that it was incorrect. The real max (read) bandwidth is 196 GB/s. I agree with the pcper crew that this is mostly irrelevant to most 970 owners; they knew what the performance was from reviews and the performance is still the same regardless of what the specs actually are. It was a major PR blunder to not make those specs known though.
I am not the original “Anonymous” in this sub-thread; the original looks like they could be a troll. Both of you seem to be partially incorrect though. Yes, you can write to one and read from the other, but as far as I can tell, you can never read from both partitions in the same clock due to the shared read port. This means you can never achieve the same bandwidth as the 980, at least where bandwidth is assumed to be read bandwidth only, which is generally the case.
Agreed. Which is why I stated
Agreed. Which is why I stated that this can only be achieved as a combination read/write but not when purely reading or writing. However, in order to reach the theoretical maximum the GPU must use above 3.5GB.
Under normal use it will be limited to 192 GB/s.
You are being a bit loose
You are being a bit loose with your definition of bandwidth from what has generally been used. What is the write bandwidth on 980? Do we care? The “max theoretical bandwith” is 196 GB/s whether under normal use or not. Quoting it as 224 GB/s would imply to most people that all of it is available as read bandwidth which is not the case.
Anyway, I suspect the performance would be only marginally different if the extra partition was just not used at all. It is 196 GB/s read bandwidth or 28 GB/s read bandwidth but not both at once. Being able to write the second partition at the same time as reading the main partition is of limited value. The second partition can only be read at 28 GB/s and it can not be read at the same time as the other partition. Also, if this is managed by the OS, it may not be used effectively without added optimizations. I don’t think we have had a card with such unbalanced partitions before.
Not any of the above
Not any of the above anonymous posters…
Why do you suspect that the original poster might be a troll? It sounds to me as if he/she is just pissed off at both Nvidia’s lies and the press’s whitewashing of them.
Nvidia put out 4 major errors in their publishing of the 970 specifications, and did nothing to correct them until enthusiast caught on, at which point they had a number of cozy one on one conference calls with a bunch of tech site representatives.
After the calls piratically the whole tech journalism community responded to the controversy with a unified front: “Nvidia made an honest mistake about the specs, here’s why, it’s kind of crappy, but your 970 is no less awesome today than yesterday.”
When a backlash ensued most waded into their respective comment threads to defend Nvidia further and to marginalize anyone who rightfully felt cheated by the lies.
The job of a journalist is to question everything, not rush to the defense of the subjects they cover. Their job isn’t to sweep things under the rug, but to pull the rug back and show everyone what’s under it. They are not supposed to be the public relations arm of the companies they investigate, but a check against their improprieties. I’d argue that with regard to this story, they failed their role as consumer guardians completely, and reading the original poster’s comments, my guess is that that is where the anger comes from.
Ryan if you’re reading this, please reflect on what your role is as a journalist.
I really don’t see how this
I really don’t see how this ‘issue’ effects the original reviews or why anyone would change their opinion of the 970 based on these results.
The FCAT results in the original reviews remain the same. The cards performance has not changed, it did not get magically slower in review site benchmarks once this information was disclosed, only some peoples perception of the card have changed.
SLI users “may” have a legitimate issue but I find it extremely hard to believe single GPU users will actually be effected by this.
Again, there is something
Again, there is something much bigger going on here than the card’s performance.
1)It is never ok for a company to lie about the specifications of something they are selling. When they do they need to be held accountable in order to ensure they don’t do it again. A big component of this process is a free press.
2)It is not the role of the press to take Nvidia’s explanation at their word, defend their word, and marginalize anyone who agrees with point #1.
3_It is their role to question Nvidia, independently investigate their claims, independently investigate how the impropriety happened, shine a big spotlight on it, and find out exactly what Nvidia will do to make it right.
The press has largely been doing the things in point 2 when they should have been doing the things in point 3. This is a failure on behalf of the press, and i’d argue it’s a bigger deal than the lies that spawned it.
I agree with the reviewers.
I agree with the reviewers. It would be very stupid for Nvidia to lie about this. There is little chance that it would go unoticed forever so this seems like a rather massive oversight. From now on, reviewers will probably be testing bandwidth to make sure that it matches with the specifications.
Absolutely agree.
I am
Absolutely agree.
I am getting so tired of “journalists” outlining the reasons why Nvidia’s marketing department did what they did. We are all aware of how marketing works, what we need are accurate specs from which to make purchasing decisions.
Additionally, a few notable points from the podcast:
– Josh’s dismissal of the individuals who attempted to equalize the FLOPs of the 980 and 970 in order to find performance differences. Yes Josh, there are additional differences between the two cards (ROPs as you mentioned), something we were not aware of until last week.
– “you get what you pay for”, well I paid for 64 ROPs, and a true 256 bit memory bus and got neither
Wasn’t necessarily a
Wasn't necessarily a dismissal, but I just don't agree that their measures would prove to be accurate enough for a good comparison.
I know people are mad about not being told the truth about the architecture. If you are upset about it, that is certainly your perogative. I am not telling you that you are wrong or right about that stance. My opinion of it is different. That's the wonderful thing about this type of interaction, I'm not forcing you to think a different way. Hell, I'm not even trying to cajole you. I am merely expressing my opinion about the situation. NVIDIA should have been upfront about how they set up the 970, just as they did with the 660 Ti.
So my "you get what you paid for" remains true (though slightly less so, as AMD and its partners have dropped their prices on the R9 290X in a pretty significant way). The 970 still performs very well in a variety of situations. Certainly more testing on the SLI aspect is needed though. That is actually looking like a really troubling situation.
I think it’s pretty clear you
I think it’s pretty clear you don’t understand. The double standards this site has shown over the last 2-3 years is shocking. What did NV show this site and a few others from those 100,000 stolen AMD documents anyway? It was something Ryan said a while ago that gave it away.. 😉
You have to be joking. Slower
You have to be joking. Slower 512mb is virtually inconsequential to what AMD put me through. Even if SLI is a problem at over 3.5Gb usage it’s still inconsequential to what AMD did.
One of my 6870s was practically useless in Crossfire for 3 years! 3 YEARS!!!!
That’s nothing, I haven’t
That’s nothing, I haven’t been able to use one of my NV GPU’s in SLI for over a bloody decade! That’s over *5* years!!!!!! Just sitting there collecting dust because it’s worthless. Not that I’d want to put anything in SLI considering the terrible stutter and frame latency compared to the much smoother Crossfire. But you won’t see sites like PCPER and Techreport reporting on that with anything more than a whisper. It’s things like this that lost these sites any credibility they had left after NV paid them off with FCAT.
Really? So if it had two
Really? So if it had two 128bit buses, that couldn’t use simultaneously for reading, that would be enough to call it an 256bit card? Are you a comedian or something?
here is a direct quote if you
here is a direct quote if you even watched the episode
“NVIDIA screwed up when they gave us the wrong specifications for the part and that’s really crappy.”
yep sound like they are on NVIDIA’s leash.
SHOCKING interview with
SHOCKING interview with Nvidia engineer about the 970 fiasco: http://youtu.be/spZJrsssPA0
“God bless the reviewers!”
“God bless the reviewers!”
Damn nVidia got all the right
Damn nVidia got all the right answers… maybe they can tell us “hey… where all the white women at?” 01:09:57 Just a joke, anyways…
970 Saga/Sammy Evo 840 thoughts… they should have called these the “GTX970 Trap Edition”, and the “Sammy 840 Evo Trap Edition” because:
Pretend you are walking down the street, you meet a girl… take her home. You are in bed and playing, come to find out… she lied/failed to tell you/misrepresented/overlooked/didn’t know/etc… You will just say, “Ok MAN… go ahead and give it to me up the butt, at least I’m getting some lovin’.” Some of you would even ask for another date, and the real freaks would not even care about the initial violation… then ask the freak be his wife. WTF??? Even if the freak got a sexchange (or firmware update to meet your expectations) it is just not what you were expecting in the first place right?
Honestly is the best policy… that should be a companies first priority, and they should check many times over to make sure they are representing honestly. Yeah… you sick freaks, you need to re-evaluate your reality. Nobody likes to be misled, well some of y’all weirdos might like that fat lying businessman to violate you repeatedly… yeah you know who you are.
Even if a company has much fancier products with higher specifications, I would never do business with them if they are hella untrustworthy… especially if they come at me with a much higher price for basically the same performance. Am I wrong?
ye allyn, bitmap brothers
ye allyn, bitmap brothers gods was awesome!
Something an ap said above
Something an ap said above struck me and i thought it would be best to break this out into a new thread:
The mention of the “unified front” is spot on, and to be honest looking back on how everything went down it appears that there almost had to have been more than met the eye with regard to everyone coming out in unison with virtually the same responses.
So i must ask if that response was coordinated in any way between journalists from different tech sites: Were any of you guys discussing how you were going to respond to this story with journalists or representatives from other news/blog outlets in back channel chat rooms, emails, or other types of communication?
This business is all about
This business is all about traffic and scoops. We do not willingly share info with other sites for fear of either showing our hand, or at the very least giving them an idea of what we had stumbled upon. So to answer your question, afaik, there was no communication between us and other sites about this story. We want to keep our information private until we are ready to publish.
Litigate, litigate,
Litigate, litigate, litigate:
http://bursor.com/investigations/nvidia/