Build for 1080 and HTC Vive

[bloodwulf]

Building my wife a replacement PC for her 6 year old rig. Not looking to salvage any parts as we are going to move her old computer to her office for work. I am willing to wait until post AMD announcement, and i can wait for a 3070 if that is the best deal. She wants a PC that is going to last for 4-5 years (should be easy because she doesn't typically play overly intensive games). I built her last rig, and just have been out of the CPU scene a lot lately, so I don't know what is the ideal devices atm. She does use two 27inch Benq monitors if that matters for any reason.

Here are the bullets:

Budget
$1500 (less better)
Resolution
1080p
System Purpose (gaming, workstation, HTPC, etc)
Gaming and VR
If a gaming system, what games and settings are desired?
League of Legends, World of Warcraft, VR games
Any other intensive software or special things you do (Frequent video encoding, 3D modeling, etc)
Outside of games and VR nothing special.
Do you plan to overclock?
No
Country
US
Preferred Stores / Sites
Any, No Microcenters within 3 hours of me though.
Parts that can be reused
None
Do you need an OS?
W10Home
Do you need peripherals (e.g. monitor, mouse, keyboard, speakers, etc)?
No

[Temp name]

Yeah, I'd wait until after the AMD GPU and CPU announcements. They should come in early October. We can't really say what to buy until we know what there is on offer.

[bloodwulf]

That was my thought as well. But figured id post if someone had a good build using previous gen. I'm also very doubtful that the AMD products would be reasonably obtainable in the next 3 months if they are even releases and id like to have the computer built by December. But ill at least wait til after the CPU announcement to pull the trigger.

[kukkamies]

3070 would be next best bet on release day, 3080 is already out of stock and backlogged a loong time. One store here says delivery in January if order now lmao. Big Navi is looking around 3070 levels of perf, but without the Nvidia drivers, RTX Voice, DLSS and all that.
Everything else is fine with last year tech except GPU I'd say. The new ones are so much better.

[bloodwulf]

3070 would be next best bet on release day, 3080 is already out of stock and backlogged a loong time. One store here says delivery in January if order now lmao. Big Navi is looking around 3070 levels of perf, but without the Nvidia drivers, RTX Voice, DLSS and all that.
Everything else is fine with last year tech except GPU I'd say. The new ones are so much better.

Kinda of my feels. I have the 3080 on a watchlist for my personal computer, but i plan on playing 1440p with Raytracing etc, but my wife should be more than well off with a 3070. Currently looking at a I5 9600k, but i'll at least wait til the CPU announcement next week, just to see if anything changes. I don't imagine the AMD GPUs will be ground breaking enough to justify waiting for them to be obtainable.

[kukkamies]

Kinda of my feels. I have the 3080 on a watchlist for my personal computer, but i plan on playing 1440p with Raytracing etc, but my wife should be more than well off with a 3070. Currently looking at a I5 9600k, but i'll at least wait til the CPU announcement next week, just to see if anything changes. I don't imagine the AMD GPUs will be ground breaking enough to justify waiting for them to be obtainable.

There's already i5-10600 as well. I haven't followed intel at all, so no idea if it's a big improvement or nothing at all. Seems it was released without any hype.

Yeah AMD GPU looks weaker and weaker as the leaks progress. At least the new Zen 3 CPUs are a 10-15% improvement again. Hoping 8 core for around 200e, or even new processor to start at 200e and not raise prices... 3600 is so cheap for 6core.

[Shakadam]

Big Navi is looking around 3070 levels of perf

Yeah AMD GPU looks weaker and weaker as the leaks progress.

You really don't know what you're talking about tbh.

[Thunderball]

You really don't know what you're talking about tbh.

I dunno about 3070 levels because we have no idea where that is but Big Navi indeeds looks worse and worse as more info comes out.

[Mojo03]

I dunno about 3070 levels because we have no idea where that is but Big Navi indeeds looks worse and worse as more info comes out.

You can't go by that stuff, it's rumors. You can just as easily say it's looking better and better, depending on which sites you go to or who you watch on YouTube.

You're not really going to know for a month.

[blackmartin]

Kinda of my feels. I have the 3080 on a watchlist for my personal computer, but i plan on playing 1440p with Raytracing etc, but my wife should be more than well off with a 3070. Currently looking at a I5 9600k, but i'll at least wait til the CPU announcement next week, just to see if anything changes. I don't imagine the AMD GPUs will be ground breaking enough to justify waiting for them to be obtainable.

Don’t bother buying a K-model CPU if you have no plans to overclock. Your paying a premium for the unlocked feature but then not utilizing it.

[kukkamies]

You really don't know what you're talking about tbh.

AMD was hyping it a lot, and now Navi 21 is looking more expensive than 3070 for similar or worse performance. Also of course less features since worse drivers.
Navi 22 seems to replace 5700XT for half the performance of 3070. Signs are pointing it'll be around 400usd which is a lot for that.
At least the 5700 XT was better than 2070 for cheaper. This time the AMD advantage could be availability, if they have much more stock than Nvidia.

I do hope AMD will still pull it through, with lowering the prices a lot. Also hope Nvidia would have more supply.... Gonna need lots of luck to get 3070 FE, otherwise might as well wait for AMD release.

[bloodwulf]

Don’t bother buying a K-model CPU if you have no plans to overclock. Your paying a premium for the unlocked feature but then not utilizing it.

I get that, just was finding one for $30 bucks cheaper than the non-K Model. K for 199 and non-K for $237.

[Shakadam]

I dunno about 3070 levels because we have no idea where that is but Big Navi indeeds looks worse and worse as more info comes out.

AMD was hyping it a lot, and now Navi 21 is looking more expensive than 3070 for similar or worse performance. Also of course less features since worse drivers.
Navi 22 seems to replace 5700XT for half the performance of 3070. Signs are pointing it'll be around 400usd which is a lot for that.
At least the 5700 XT was better than 2070 for cheaper. This time the AMD advantage could be availability, if they have much more stock than Nvidia.

I do hope AMD will still pull it through, with lowering the prices a lot. Also hope Nvidia would have more supply.... Gonna need lots of luck to get 3070 FE, otherwise might as well wait for AMD release.

I don't even know what kind of info you guys are looking at.
The most useful info we have about next gen Navi is this: https://videocardz.com/newz/amd-navi...navi-23-32-cus or straight from the source: https://www.reddit.com/r/Amd/comment...nne/?context=3

TLDR:

Navi 23 = 32 CU's, clock speed still unknown. Likely replacement for the Polaris GPU's and the RX 5500. RX 6500 most likely.

Navi 22 = 40 CU's, 2.5Ghz clock speed. Likely replacement for the RX 5700 and 5700XT, maybe 5600XT as well. RX 6700(XT) most likely, might be split into lower binned part/less memory RX 6600 also like the current gen. Same CU's as 5700XT, significantly higher clock speed. Would be about a 30% perf increase from the increased clock speed alone and not counting any architectural improvements in RDNA2. Would go up against 2080TI and likely 3070 as 2080TI is ~35% faster than a 5700XT.

Navi 21 = 80 CU's, 2050 - 2200 Mhz clock speed. Likely RX 6800(XT) and 6900(XT) to go up against the 3080 and 3090. Very difficult to know what kind of performance to expect as there was never any chip this large on either the RDNA1 or the GCN architecture.

[kukkamies]

I don't even know what kind of info you guys are looking at.
The most useful info we have about next gen Navi is this: https://videocardz.com/newz/amd-navi...navi-23-32-cus or straight from the source: https://www.reddit.com/r/Amd/comment...nne/?context=3

TLDR:

Navi 23 = 32 CU's, clock speed still unknown. Likely replacement for the Polaris GPU's and the RX 5500. RX 6500 most likely.

Navi 22 = 40 CU's, 2.5Ghz clock speed. Likely replacement for the RX 5700 and 5700XT, maybe 5600XT as well. RX 6700(XT) most likely, might be split into lower binned part/less memory RX 6600 also like the current gen. Same CU's as 5700XT, significantly higher clock speed. Would be about a 30% perf increase from the increased clock speed alone and not counting any architectural improvements in RDNA2. Would go up against 2080TI and likely 3070 as 2080TI is ~35% faster than a 5700XT.

Navi 21 = 80 CU's, 2050 - 2200 Mhz clock speed. Likely RX 6800(XT) and 6900(XT) to go up against the 3080 and 3090. Very difficult to know what kind of performance to expect as there was never any chip this large on either the RDNA1 or the GCN architecture.

I looked into the tflops. 3080 is 29tflops and Navi 21 is 22tflop. It's not even close. So far AMD has had better or equal tflop but worse game performance. Now they don't even have better tflops.
For example last gen the RTX 2070 Super and 5700 XT had similar tflops, but 2070 super was 10-30% better in actual performance.

[Shakadam]

I looked into the tflops. 3080 is 29tflops and Navi 21 is 22tflop. It's not even close. So far AMD has had better or equal tflop but worse game performance. Now they don't even have better tflops.

Tflops means absolutely nothing. It's theoretical FP32 performance and that's it. You definitely can not use Tflops as any kind of measurement between different GPU architectures.

As an example: The 2080TI has 13.45 TFLOPS of FP32, the 3080 has 29.77 TFLOPS of FP32. That's a ~120% increase yet the 3080 is only ~30% faster.

The reason the 3080 has such high FP32 Tflops is because Nvidia redesigned the Ampere architecture to be able to run double FP32 if needed instead of INT32+FP32 as in Turing. But in games it can't leverage that performance because there are other bottlenecks than simply FP32 performance.

[kukkamies]

Tflops means absolutely nothing. It's theoretical FP32 performance and that's it. You definitely can not use Tflops as any kind of measurement between different GPU architectures.

As an example: The 2080TI has 13.45 TFLOPS of FP32, the 3080 has 29.77 TFLOPS of FP32. That's a ~120% increase yet the 3080 is only ~30% faster.

The reason the 3080 has such high FP32 Tflops is because Nvidia redesigned the Ampere architecture to be able to run double FP32 if needed instead of INT32+FP32 as in Turing. But in games it can't leverage that performance because there are other bottlenecks than simply FP32 performance.

Yeah it doesn't mean much. It's still the only data we currently have. Tflops in the past being more to AMDs advantage. We could see a flip this generation sure, but not sure I believe in that.

[Shakadam]

Yeah it doesn't mean much. It's still the only data we currently have. Tflops in the past being more to AMDs advantage. We could see a flip this generation sure, but not sure I believe in that.

What do you mean it's the only data we have? We have the number of CU's and we have what is very likely the clock frequency of the chips. That is waaaaaay more useful to gauge performance than any Tflop number. FP32 Tflops is only useful in something that does nothing but FP32 compute, games do a hell of a lot more than just FP32 calculations. Hell, the GTX 1080 has quite a noticeable lead in FP32 Tflops over the 2060Super, and yet the 2060Super is noticeably faster, and that's comparing one Nvidia architecture to another not-very-different Nvidia architecture.

Likewise the Radeon VII has significantly more FP32 Tflops than the 5700XT, but it isn't any faster in gaming.

[Temp name]

Don’t bother buying a K-model CPU if you have no plans to overclock. Your paying a premium for the unlocked feature but then not utilizing it.

K skus are normally clocked higher than non-K skus though.
https://ark.intel.com/content/www/us...-5-30-ghz.html
https://ark.intel.com/content/www/us...-5-20-ghz.html

2.8-5.2 for non-k and 3.7-5.3 for k

[Thunderball]

Navi 21 = 80 CU's, 2050 - 2200 Mhz clock speed. Likely RX 6800(XT) and 6900(XT) to go up against the 3080 and 3090. Very difficult to know what kind of performance to expect as there was never any chip this large on either the RDNA1 or the GCN architecture.

220-240W TDP with those clocks and that CU count (unless these CUs are not directly comparable to RDNA1 CUs) is just unrealistic. I'd believe 250W if it used HBM, but it looks like GDDR6(X?) so far. Or if that one some new node, but it's still 7nm. It's either going to be a lot more power hungry or not able to hit those clocks.

- - - Updated - - -

Tflops means absolutely nothing. It's theoretical FP32 performance and that's it. You definitely can not use Tflops as any kind of measurement between different GPU architectures.

As an example: The 2080TI has 13.45 TFLOPS of FP32, the 3080 has 29.77 TFLOPS of FP32. That's a ~120% increase yet the 3080 is only ~30% faster.

The reason the 3080 has such high FP32 Tflops is because Nvidia redesigned the Ampere architecture to be able to run double FP32 if needed instead of INT32+FP32 as in Turing. But in games it can't leverage that performance because there are other bottlenecks than simply FP32 performance.

That's true but by saying that CU count gives you some kind of information you also assume that RDNA1 CUs are directly comparable to RDNA2 CUs, so there's just two possibilities here: 1) either CUs ARE directly comparable meaning no significant architecture changes, which means Tflops numbers are directly comparable too; 2) CU numbers are not comparable due to significant architecture changes, which means we know nothing.

EDIT: Also the fact that CU number could've stayed the same but CUs include more transistors is improbable aswell, due to no node change. Same with optimizations that would give MAJOR performance uplift on the same node and the same CU count. That never happened.

[Shakadam]

220-240W TDP with those clocks and that CU count (unless these CUs are not directly comparable to RDNA1 CUs) is just unrealistic. I'd believe 250W if it used HBM, but it looks like GDDR6(X?) so far. Or if that one some new node, but it's still 7nm. It's either going to be a lot more power hungry or not able to hit those clocks.

Note that the power limit in this info isn't the TDP or TBP for the entire card, it's just for the GPU itself as far as I understand it. So you'd add the power for the memory system and VRM losses etc on top of that. So probably a 300W card.

That's true but by saying that CU count gives you some kind of information you also assume that RDNA1 CUs are directly comparable to RDNA2 CUs, so there's just two possibilities here: 1) either CUs ARE directly comparable meaning no significant architecture changes, which means Tflops numbers are directly comparable too; 2) CU numbers are not comparable due to significant architecture changes, which means we know nothing.

EDIT: Also the fact that CU number could've stayed the same but CUs include more transistors is improbable aswell, due to no node change. Same with optimizations that would give MAJOR performance uplift on the same node and the same CU count. That never happened.

Yeah that's why I said I wasn't taking into account any architectural changes in my Navi 22 prediction, basing that on simply clock frequency increase alone. I think we can be fairly certain that AMD wouldn't design a worse CU structure in RDNA2 than in RDNA1, so it's either going to be the same as RDNA1 or better. We know RDNA2 has hardware to allow for ray tracing computation and we know the layout of the compute units from the Xbox presentation, but it doesn't necessarily have to be exactly the same for the desktop product.

RDNA2 is likely on an improved 7nm node. AMD has been tight lipped about this so we likely won't know until the announcement, it could be TSMC N7P or N7+ node.