IkarugaThey really should have gone the other way around with this whole story. The 970 is still a hell of a card. If they would have said that it only has 3.5GB, it would be all fine now.
They could have just „leaked” how to enable the last 0.5GB and enthusiast would all try it and argue and debate about it on every tech forums if it’s worth it or not because of the performance hit (which is about 2-3fps?)

There would be no drama now then.

PashaFps is ok, u got sttuter when using all memory.

HTCIf this is indeed the result of this card's problem, then no amount of price cuts or rebates can make up for it and a recall SHOULD be made for cards showing this.

xfiaok so they are saying the 980 doesnt have the partition right? and they both have the same performance loss in game testing.. hell if that is the case there is seriously nothing wrong.. I mean what happens to any other gpu if you throw it more than it can handle or cpu for that matter.
nvidia is probably laughing about it because it took them an hour to figure it out and forward a statement to tech sites that is so simple its hard for a 20 year professional to quickly figure out when they where expecting something much more complex.

buggalugsIt isn't just FPS though. The difference is the 980 still renders fluid gameplay above 3.5GB, the 970 turns into a stuttering mess above 3.5GB and some users get weird artifacts and flashes of purple and stuff across the screen.

GhostRyderThe issue is only a big deal because it was hidden and can cause issues. The solution should have been just at the beginning to advertise the card as a 3.5gb card and the last .5gb that can cause the issues to be locked out or prioritized for basics that would have just been a nice bonus for people.

SasquiSo, doing some math 3.5 vs. 4.0... they should quickly offer a 13% refund to folks who purchased the card prior to the news :p

This Nvidia-provided slide gives brief insight into how the GTX 970 is constructed. The three disabled SMs are shown at the top and 256KB L2s and pairs of 32-bit memory controllers on the bottom. Notice the greyed-out right-hand L2 for this GPU? Tied into the ROPs as they are this is a direct consequence of reducing the overall ROP count. GTX 970 has 1,792KB of L2 cache, not 2,048KB, but, as Alben points out, still has a greater cache-to-SMM ratio than GTX 980.

Historically, including up to the Kepler generation, cutting off the L2/ROP portion would require the entire right-hand quad section to be deactivated too. Now, with Maxwell, Nvidia is able to use some smarts and still tap into the 64-bit memory controllers and associated DRAM even though the final L2 is missing/disabled. In other words, compared to previous generations, it can preserve more of the performance architecture even though a key part of a quad is purposely left out. This is good engineering.

But while it's still accurate to say the GeForce GTX 970 has a 256-bit bus through to a 4GB framebuffer - the memory controllers are all active, remember - cutting out some of the L2 but keeping all the MCs intact causes other problems; there is no usual eighth L2 to access, meaning that the seventh L2 will be hit twice. The way in which the L2 work makes this a very undesirable exercise, Alben explains, because this forces all other L2s to operate at half normal speed.

Smoke and mirrors
Finally coming back to point, Nvidia gets around this L2 problem by splitting the 4GB memory into a regular 3.5GB section, constituted by seven MCs and associated DRAM, and a 0.5GB section for the last memory controller. The company could have marketed the GeForce GTX 970 as a 3.5GB card, or even deactivated the entire right-hand quad and used a 192-bit memory interface allied to 3GB of memory but chose not to do so. How does this play out with the huge memory bandwidth drop-off in the Lazygamer Nia test versus Nvidia's statement that games barely suffer from this smart engineering? The Lazygamer test at the >3.5GB metric simply probes bandwidth on a single DRAM, which is admittedly low, or 1/8th of the total speed, while in-game code, according to Nvidia, doesn't pinpoint memory in this way. There's certainly a memory-bandwidth drop-off when the 0.5GB section is called into action, Alben states, but it's not anywhere near as severe if nonrecurring code is shunted into the last MC.

In a high-level nutshell Nvidia is using smart engineering to get the most out of the GTX 970's architecture. The lack of total ROPs is relatively unimportant because this GPU cannot make use of them - the 13 SM units, running at four pixels per clock (so 52 in total), are limiting the GPU more so than the 56 processed by the ROPs. The GeForce GTX 970's performance hasn't changed, obviously, but Nvidia wasn't clear on how the back-end works... and it has taken investigation by enthusiasts to uncover the real reason why this 256-bit architecture isn't as good as the GTX 980's

The Lazygamer test at the >3.5GB metric simply probes bandwidth on a single DRAM, which is admittedly low, or 1/8th of the total speed, while in-game code, according to Nvidia, doesn't pinpoint memory in this way

SasquiSo, doing some math 3.5 vs. 4.0... they should quickly offer a 13% refund to folks who purchased the card prior to the news :p

trenterIt's not a "bug", it's a total slap in the face of 970 owners. Nvidia just realized that this whole time they were selling 970's the specifications they were quoting were actually wrong. They admitted the 970 only really has 3.5gb of useable memory because the other .5gb is accessed at 28 gbps....WTF. Now let's hear the nvidia fangirls find a way to defend this joke for a company.....starting with W11zzardd.

trenterIt's not a "bug", it's a total slap in the face of 970 owners. Nvidia just realized that this whole time they were selling 970's the specifications they were quoting were actually wrong. They admitted the 970 only really has 3.5gb of useable memory because the other .5gb is accessed at 28 gbps....WTF. Now let's hear the nvidia fangirls find a way to defend this joke for a company.....starting with W11zzardd.

64KI will take the 13% ($45.50) or Jen-Hsung Huang can come to my house and detail my Camry. :D

NVIDIA’s Senior VP of GPU EngineeringTo those wondering how peak bandwidth would remain at 224 GB/s despite the division of memory controllers on the GTX 970, Alben stated that it can reach that speed only when memory is being accessed in both pools.

HumanSmokeHow auspicious. 2 posts in and you've already called out two staff members.
:shadedshu::shadedshu::shadedshu:

XzibitGTX 970 now has 52 ROPs instead of 64 and 1792KB of L2 Cache instead of 2048KB

Rahmat SofyanFor Fun, W1zz should change the Value and Conclusion for GTX 970 right now :D
Oh, and nVidia seems fucked :p

XzibitDon't forget the difference in ROP and L2 cache.

GTX 970 now has 52 ROPs instead of 64 and 1792KB of L2 Cache instead of 2048KB



This is entertaining to see develop :toast:

HumanSmokeLike AMD did with Bulldozer when people found out that 4 modules isn't the necessarily the same as 8 cores ?:laugh:

XzibitHere is PcPer take on it.

With that said, our 4K test did pick up a potential discrepancy in Shadows of Mordor. While the frame rates were equivalently positioned at both 4K and 1080p, the frame times weren’t. The graph below shows the 1% frame times for Shadows of Mordor, meaning the worst1% times (in milliseconds). www.extremetech.com/extreme/198223-investigating-the-gtx-970-does-nvidias-penultimate-gpu-have-a-memory-problem/2

HumanSmokeThe revised total of ROPs is 56 not 52.

Why? What changed? Is the performance of the card any less? Does it use more power than it did when it launched?

XzibitDon't forget the difference in ROP and L2 cache.

GTX 970 now has 52 ROPs instead of 64 and 1792KB of L2 Cache instead of 2048KB



This is entertaining to see develop :toast:

XzibitHere is PcPer take on it.

the54thvoidKinda what Anandtech are saying too.

Nvidia has egg on it's face for 'ahem' lying about it's card, no doubt but the performance of it isn't an issue. Each reviewer looking at it in turn, (PCper, Anand, Hexus) has the same conclusion which is threefold:
1) Nvidia have slipped up and undoubtedly their PR and engineering sections have 'misled' the public somewhat. (IMO, I don't believe it was innocent but hey)
2) The real performance impact isn't there. The card, according to all sites so far, is still great.
3) People are trying and failing so far to find a real world gaming example that kills the cards performance, outside of a load that would do that anyway based on it's SMM units etc.

FWIW, IMO, Nvidia knew fine well what they were releasing and probably expected no fall out from it, due to the fact it has no impact on real scenario's. But techy people like to dig and found an anomaly. Now NV have to explain it and it's hard to make this one sound like a genuine 'miss'. Even if it was a genuine lapse, it's very hard to sell to us, the public.

But hey, this ugly truth (bad move NV, but still a great card) won't stop people throwing those ignorance stones.