Reader Mail for July Loose Ends
Added 2023-08-11 18:21:19 +0000 UTCIt's time to field questions for a livestream once again! It's expected that there will be A LOT of questions about RDNA 4, RDNA 5, and Nvidia's next gen...but note that the following news has broke since the last livestream:
- Arrow Lake failed to get HT working with Royal Core
- Details about Lion Cove+, Cougar Cove, Diamond Rapids, and Beast Lake were Leaked
- Rentable Units have been detailed as the successor to HT (or supplementing its removal)
- MLID called the GPU Pricing Bottom (and so far it was dead on!)
- Nvidia has halted Lovelace Production
- RX 7900 GRE has been Released
- AMD's Dragon Range-X V-Cache Laptop chips have made their first debuts!
You have ~4 hours to submit questions/comments here!
Comments
What exactly is RDNA 4’s purpose? Is it a fixed refresh of 3 that is cheaper with the fixed nature of it giving it the extra performance or is it better in some other way that warrants its release? I can’t imagine it selling well unless it’s cheaper, I can’t imagine people buying it over current gen stuff.
XTX 999
2023-08-11 23:07:06 +0000 UTChey tom. when RDNA2 will phase out, what prices do you expect for current gen cards around black friday/holidays?
Dylan K.
2023-08-11 22:55:19 +0000 UTCDo you know what Navi 4C is?
Jebediah
2023-08-11 22:51:53 +0000 UTCNo comment. All I can say. ;)
Moore's Law Is Dead
2023-08-11 21:49:33 +0000 UTCHey, Tom. You mentioned recently that both AMD and Nvidia are considering exiting the high-end consumer GPU space. What do you think is the likelihood that upcoming GPU generations from AMD and Nvidia will entirely exclude high-end dies from the consumer lineup? For example, would future Blackwell 102 or Navi 41 dies even make it to GeForce and Radeon products?
kjm015
2023-08-11 21:37:39 +0000 UTCTech question- With nVidia diverting Lovelace production to Hopper, it looks like they are starting to exit the high end consumer GPU market. What kind of market conditions do you think it would take for them them to consider staying in high end consumer GPU? Non tech question- I presume that Dan will bring his dog when he moves in with you. Is Jessie okay with this? :)
Woody Chang
2023-08-11 21:12:59 +0000 UTCHey Tom, how are you? Question 1: With Intel struggling, do you think they may abandon their practice of short-lived platforms (like AM4, AM5)? Cause I think that could make it more justifiable for some people to invest in. Question 2: When FSR 3 arrives and it's much closer to DLSS, more people might switch to AMD. However, there is still some people stuck with NVidia due to the wide support of CUDA. Do you think that eventually (and if more AMD cards are sold), we'll ses some talented people trying to develop something like the proton layer, but for translating CUDA to OpenCL?
Dave Scholze
2023-08-11 20:40:23 +0000 UTCMr. MLID. Will DLVR be in all future Intel CPUs or is this a specific thing to Raptor Lake? Is this coupled with the Royal Core project what will get Intel back into the efficiency game?
CompressedAIBlocks
2023-08-11 20:19:27 +0000 UTCNext year AMD should be launching Strix Point, Strix Halo, Navi 43 and Navi 44, which should nicely cover everything from low-end low-power graphics to the mid-range. Except for those wanting a new $1000+ GPU, AMD’s cancellation of Navi 42 and Navi 41 isn’t going to cost us anything except the chance to geek out over the design. In your opinion, what does Navi 43 need the most to be a success?
Chris Rijk
2023-08-11 20:01:15 +0000 UTCHey Tom, I've watched and rewatched your latest RDNA 4 video, but it's not clear what the difference between Navi 4X and 4C is/was. Was 4C chiplet based, and 4X monolithic, or am I interpreting that incorrectly? Have your sources been able to provide any detail on that, or this something we need to wait a while on, like with the explanation of Rentable Units?
Cleansweep
2023-08-11 19:15:54 +0000 UTCAMD’s Zen product line probably makes around 10 times the revenue of their Radeon product line, which probably explains why AMD can afford to have backup plans, as you laid out in your Zen 5 leak video a few months back. Going further, should AMD take a safer approach with product development for Radeon and aim lower from the start? I’d say no, but wanted to hear your thoughts. PS In yesterday’s video you suggested that maybe AMD should have delayed the release of RDNA 3, but here I'm specifically talking about the design side.
Chris Rijk
2023-08-11 19:12:52 +0000 UTCHey Tommm, is AMD planning on making any fundamental changes to their approach to raytracing with rdna 4, or 5? For example AMD's architeture has been said to be poor for multiple light bounces. Will we only see a small performance dip with raytracing on in the future at least?
Swiggles
2023-08-11 19:12:15 +0000 UTCShould preface with looking at current gen hardware. And the real question to answer is, does an 7900XTX accomplish the goal. Can future improvements come solely from FSR DLSS?
BaconHouse Collective
2023-08-11 19:11:18 +0000 UTCWouldn't it be better for the industry as a whole to have a Big launch - small launch cadence for GPU releases? With the rising cost and complexity of designing new GPU architectures it seems to me that having longer cycles between new architectures with a refresh in the middle that is merely a refinement of what's already out would help GPU companies deliver better products and save money and time in the long run.
Rootknight
2023-08-11 18:54:07 +0000 UTCTom, this question is in regards to Rentable Units. I found myself talking to someone in Discord (insert shameless plug for MLID discord here) that these kind of sound like CMT (clustered multithreading) approach that AMD attempted (and failed) with Bulldozer HOWEVER these seem to have none of the downsides of CMT. Is this where the idea stemmed from? Is this more so to make use of resources that aren't being touched in a more effective way than HT can or is this more of an evolved philosophy of CMT? Bonus question: Will HT be making a comeback with Rentable Units at any point in the future after Arrow Lake?
CompressedAIBlocks
2023-08-11 18:43:25 +0000 UTCWelcome back from vacation Lord Tometh. So this is a question based off your recent video that was made in response of AMD quitting high end and FSR 3. This one is a decently long one here. Hypothetical here: Say AMD releases RDNA 3.5/4 (whatever it is) in Q3/Q4 2024 as planned, and it's a Radeon VII > RX 5700 XT situation where depending on situation it's a wash in performance (5-10% performance in either direction) BUT the newer one (RX 5700 XT in that case) has increased power efficiency, smaller die size, can be cheaper. Is this a win for consumers? 2-ish years later, same performance but everything outside of performance IS improved. I ask if this is a win for consumers because I understand as a business standpoint, with how the market is and Nvidia's plans, this way makes the most business sense IMO. Hell, I'd personally be happy with RX 7900 XTX performance if it was a smaller die, lower power consumption AND cheaper to buy BUT does this feed into the perception that Radeon has such a hard time to shake off? The one that they can't compete with GeForce cards? I think the more educated/informed enthusiast/consumer in this space would and could understand why this is happening, but for the average consumer (which are the vast majority of sales) does this feed into that perception? What are your thoughts?
CompressedAIBlocks
2023-08-11 18:38:38 +0000 UTCWe know that it's going to be hard to get a gaming card to work well with multiple compute chiplets, but DLSS and other AI tech heavily leverages the AI (tensor) cores in GPUs, these cores do only one thing: fused multiply add (matrix multiply) on large datasets. Do you think it may be possible for NVidia to break these cores out into a separate chiplet? Could the first MCM design from NVidia be RDNA3 style external memory controllers, feeding a single compute chip and one or more AI chips? Seems like this design would open up more room for compute cores on the main die, while also allowing NVidia to play with the compute vs AI core balance on their cards.
Timothy Baldridge
2023-08-11 18:36:05 +0000 UTCHey Tom. Wondering what performance levels (terrflops, Vram, etc) are needed on avg for 4k 120+ fps high settings gaming on the AAA 2023 releases? Pulling this thread with regards to the no RDNA 4 high end. I think we need to characterize where we are as far as hardware needs for top tier gaming with DLSS/FSR quality turned on. This should frame the argument for whether we actually need high end going forward.
BaconHouse Collective
2023-08-11 18:33:44 +0000 UTCIt makes sense for AMD to wait on a high end successor card to the 7900xtx until there is a full generational leap in raytracing and efficiency. Any guesses on the timeframe a successor might arrive and what performance lift AMD could be targeting? What aspects of RDNA4 are giving them trouble?
Frobozz
2023-08-11 18:33:36 +0000 UTCDo you see a dragon range version of the 7800x3d coming down the line?
ishould
2023-08-11 18:32:03 +0000 UTCHello Tom, xtx owner here. With your FSR 3 leak last night if Navi 32 is delayed for any reason or they chose to not launch it next month, do you think it will be pushed back as well?
Sad XTX 999
2023-08-11 18:31:40 +0000 UTCWhy did gamers refuse a $900 80-class GPU this generation, instead opting for a $1200 80-class entry option?
Jen-Hsun Huang
2023-08-11 18:22:30 +0000 UTC
