New The PC Build Thread

stefan9 said:
The YouTubers tend to talk about oc in the follow up reviews. The article reviews like guru3d and pcg did talk about the OC.
Click to expand...

I don't disagree, but Nvidia's Blackwell is being super criticized as a gaming product. I do think that expectation was a bit too high though, and strange considering that even on Nvidia's own roadmap, Blackwell would have continued on the same process node. My expectations were also a bit higher since I expected GDDR7 to be a winner. The GPU does have headroom, and a lot when compared to previous generations.

Then there is the underlying tech, that is not adopted at this time.

Day 1 reviews are day 1 reviews. How Blackwell will mature is up to Nvidia.

As much as I am on Team Red's side, the underdog, they are not acting the part. RDNA 3 is neglected at this time. Not the architecture, but the driver and stack support. I get it, the 9000 series should have launched by now. For them taking their time ironing out bugs, it must be better ironed out by the 5th.

It is sad to say, but PR in this market is akin to PR in politics. Don't believe it all.
 
I see LTT's review touched on what I said above:


I guess /cough, Nvidia gave them some extra review notes.
 
This is a very serious issue:

www.tomshardware.com

Some RTX 5090s are shipping with missing ROP units, leading to less gaming performance: Updated

5-8% slower than other 5090 models.
www.tomshardware.com www.tomshardware.com

Update:

A new report from hardware sleuth MEGAsizeGPU, who frequently gets the inside scoop from board partners, alleges Nvidia has supplied AIBs with defective GB202 chips, suggesting the issue is a hardware defect that cannot be corrected through software or BIOS updates. Consequently, all RTX 5090s may potentially ship with fewer ROPs than advertised, including Nvidia’s own Founders Edition design. The China-exclusive RTX 5090D is no exception either, being powered by the GB202 chip at its core.

RTX 5090s supplied to independent reviewers had their die stamped with the term “Press Build”. While Nvidia explicitly denied any performance enhancements, it’s speculated these chips had all 176 ROPs enabled. This has led to claims that certain retail variants might be carrying a “defective” or slightly cut-down GB202 chip with fewer functioning ROPs. Note this is not a proven fact and is still a hypothesis.

Both GPU-Z and HWiNFO acquire data through NVAPI and confirm the same readings which rules out software errors. If you’ve been lucky enough to get your hands on an RTX 5090, the only way to determine whether your model is affected is to check the ROP count. Make no mistake, Nvidia likely tests its chips rigorously to ensure validity. However, given the widespread Blackwell shortage, the use of slightly trimmed-down GB202 chips might have been a stopgap solution to meet demand. Even so, retailers are still putting customers on hold for up to 14 weeks to secure an RTX 5090.

So far, GPUs from MSI, Manli, Zotac, Gigabyte, and Nvidia’s own Founders Edition model have reported instances of reduced ROP counts. This list is likely to increase as more reports emerge.
Click to expand...

Nvidia must do a recall and replacement of these cards. This isn't even a matter concerning warranty; this is a manufacturer defect. All warranties would need to be renewed. It is just a matter of how widespread this is.
 
AMD made it known in their media invitation to the 90x0 launch event that the 9070/XT will be AIB models only. The GPU in their marketing material does not exist. This is an interesting change.

Now I am curious how MSRP will be managed since the model usually sets the base price, including that at what the GPUs may be advertised at.

Funny, maybe they don't announce a MSRP, lol.
 
Fulcrum29 said:
AMD made it known in their media invitation to the 90x0 launch event that the 9070/XT will be AIB models only. The GPU in their marketing material does not exist. This is an interesting change.

Now I am curious how MSRP will be managed since the model usually sets the base price, including that at what the GPUs may be advertised at.

Funny, maybe they don't announce a MSRP, lol.
Click to expand...
AMD is going to AMD it again I guarantee it, my guess is it's going to be $749 for the XT and it's going to be maybe 10-15% faster than the GRE but with FSR 4.0 and slightly better ray tracing performance.

And then after no one buys it it's going to drop to $649 which will be kinda interesting and intriguing but then there will be some kind of sale where you can get it for $599 and then that's when the Pascal etc holdouts might say the time has come to upgrade.
 
Last edited:
Soul Assassin said:
...than the GRE but with FSR 4.0 and slightly better ray tracing performance.
Click to expand...

I initially typed a way longer reply to the above, but to keep it short. I do think that FSR 4 will come to RDNA 3, but with some caveats. There is documentation covering rocWMMA, and then there are leaks and other things that I won't trust until I see it. FSR 4, like DLSS 4, will be a suite. Let's call this suite a stack, and I don't think that the entire stack will come to RDNA 3. RDNA 3 might get FSR 4 upscaling. It is also uncertain whether the entire RDNA 3 product stack will have the capabilities, mostly due to RAM.

At this time, we don't know how FSR 4 will compare to FSR 3 come to resource utilisation.

We know that Blackwell have dedicated hardware that will support their new RTX Neural Texture Compression tech. From the samples we have seen the compression looks phenomenal. According to Nvidia, the minimum requirement is Shader Model 6.0, but there is a cost without dedicated hardware. This is big.

AMD have a similar tech that they published in 2024, Neural Texture Block Compression (NTBC). Their white paper on the technologies made capable with this was using an RX 7900 XT as a target, and though there are dramatic improvements it comes at a cost. The question then is, will RDNA 4 have dedicated hardware to negate this cost, or will it be introduced with UDNA?

I can see NTBC being worked into the FSR 4 stack, and being exclusive to RDNA 4. There are many other things I can touch on, but then I am rambling. We don't know what will be new with RDNA 4. All I know is, AMD as a business, they would not wish to maintain too many branches accross a product stack that varies in architecture. Nvidia can do it because they have the resources. For people who bought RDNA 3, AMD never promised FSR 4 to you. Say the FSR 4 upscaling model is trained using FP8, which it does, to make that 'workable' with FP16/BF16/??? would require some serious optimisations, not only due to the RAM needed, but also the computation needed. Naturally I would assume it to be a lighter model, that is not as good, but is a improvement. Perhaps they will call it "FSR 3.5" as some have suggested. It is possible, but I am not a engineer so I know nothing.

Anyhow, on to the 28th.
 
It seems that Videocardz has received some intel on the reviewer notes:


Like we told you earlier this week, AMD was planning a press briefing this week for the media. We were not invited, but we sent our best agents to gather intel, and the mission was a success.

Before you look at the figures, AMD actually confirmed they do not have the RTX 5070 Ti at their labs and are in the process of obtaining one for their internal tests, hence it was not included in comparisons. Perhaps they will do some further tests for February 28, perhaps naming the card after the competition wasn’t enough of a reason to include the series.

AMD did, however, confirm that their RX 9070 series should be competitive against the RTX 4070/4080 series, and they should also be competitive against the RTX 50 series without going into details.
Click to expand...

There is a table in the link. According to Videocardz, and I won't quote it as AMD, the numbers look like this over a 30+ game average:

RX 9070 vs 7900 GRE (highest graphics):

1440p +20%
2160p +21%

RX 9070 XT vs 7900 GRE (highest graphics):

1440p +38%
2160p +42%

Say the above is true, then it is a big improvement; however, we don't know the precise details concerning this in-house testing.

Looking at TechPowerUp, the relative comparison at 1080p.

7900 XT vs 7900 GRE

18%

7900 XTX vs 7900 GRE

37%

5070 Ti vs 7900 GRE

35%

Odd that they didn't compare it to the XT and XTX then. Though, as said, not much is known about this testing.

NOTE: According to VideoCardz, AMD unlike Nvidia, didn't use FSR in their benchmarks. I doubt that.
 
Last edited:
Anyhow, this guy posted the VideoCardz screenshots:

https://twitter.com/x/status/1893631198116606285

The "FIRST PARTY" things make little sense. Some are now saying that AMD only used base clocks and that partner cards will have higher clocks. /Ugh. Same old same old. These same people ignore that AMD won't have reference cards at launch. The review models sent to their partners will be AIB cards. Nvidia did the same; they sent ASUS Prime cards out for the 5070 Ti review.
 
Fulcrum29 said:
AMD made it known in their media invitation to the 90x0 launch event that the 9070/XT will be AIB models only. The GPU in their marketing material does not exist. This is an interesting change.

Now I am curious how MSRP will be managed since the model usually sets the base price, including that at what the GPUs may be advertised at.

Funny, maybe they don't announce a MSRP, lol.
Click to expand...
Why wouldn't there be an MSRP? All 5070ti are AIB, still has a MSRP (even if its totally misleading).
 
stefan9 said:
Why wouldn't there be an MSRP? All 5070ti are AIB, still has a MSRP (even if its totally misleading).
Click to expand...
My point is that there isn't a reference model to attach a sticker price to. If AMD have stocked up, then they can theoretically enter the market below MSRP, and then gradually raise the price for meeting their ends. The RRP for the consumer could be very different from that in agreement with their AIB partners.

It is a pro for early buyers, and for magnificent day 1 reviews.

We won't know until the 28th, and the actual day of the product being available for purchase.
 
Longboi said:
Strange.

ASRock was always looked upon as a solid brand in Mainboards.
Click to expand...
From the thread the issue isn't limited to Asrock or the 9800X3D. Just so happens to be the largest sample.

Now. I had anomalies with the latest AGESA. Is it related I don't know. The temps is what had me worried, not as much the driver issues which I think have more to do with Gigabyte's own packaging.

It is hard to compare issues when people have dissimilar systems. I don't know anyone other than me with a 7950X3D in a 870E. Then there is memory too.

This whole thing is the Zen 4 launch all over again.
 
You must log in or register to reply here.
Back
Top
Sign up to the MyBroadband newsletter
X